03.07.2013 Views

PDF (1.7 Mb) - Laboratoire de Télécommunications et Télédétection ...

PDF (1.7 Mb) - Laboratoire de Télécommunications et Télédétection ...

PDF (1.7 Mb) - Laboratoire de Télécommunications et Télédétection ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

LABORATOIRE DE TÉLÉCOMMUNICATIONS<br />

ET TÉLÉDÉTECTION<br />

B - 1348 Louvain-la-Neuve Belgique<br />

ABOUT MAXIMUM-LIKELIHOOD PHASE ESTIMATION<br />

IN DS-CDMA COMMUNICATION SYSTEMS<br />

Laurent SCHUMACHER<br />

Thèse présentée en vue <strong>de</strong> l’obtention du gra<strong>de</strong> <strong>de</strong><br />

Docteur en Sciences Appliquées<br />

Jury composé<strong>de</strong><br />

Luc VANDENDORPE (UCL - FSA/ELEC/TELE) - Promoteur<br />

Michel GEVERS (UCL - FSA/INMA/AUTO) - Examinateur<br />

Marc MOENECLAEY (Universiteit Gent - FTW/TELIN) - Examinateur<br />

Paul DELOGNE (UCL - FSA/ELEC/TELE) - Examinateur<br />

Marco LUISE (Università di Pisa - DII) - Examinateur<br />

Piotr SOBIESKI (UCL - FSA/ELEC/TELE) - Prési<strong>de</strong>nt<br />

Décembre 1999


Remerciements<br />

Le texte que voici synthétise les résultats <strong>de</strong> plusieurs années <strong>de</strong> recherche.<br />

Un tel aboutissement n’est jamais l’œuvre d’une personne seule, mais la<br />

conjugaison, à travers elle, <strong>de</strong> multiples contributions, parfois mo<strong>de</strong>stes,<br />

souvent déterminantes, parfois involontaires, souvent décidées. Nombreuses<br />

sont dès lors les personnes auxquelles je souhaiterais exprimer ici<br />

toute ma gratitu<strong>de</strong> pour leurs précieux conseils <strong>et</strong> leur soutien <strong>de</strong> chaque<br />

instant.<br />

Primus inter pares, mes remerciements vont à mon promoteur, le Professeur<br />

Luc Van<strong>de</strong>ndorpe. Son magistère scientifique m’a guidé dans les arcanes<br />

du traitement du signal <strong>et</strong> son inébranlable confiance a su ranimer<br />

la flamme chaque fois que celle-ci vacillait.<br />

Je tiens aussi à saluer les membres <strong>de</strong> mon comité d’encadrement, les Professeurs<br />

Michel Gevers <strong>et</strong> Marc Moeneclaey, ainsi que les autres membres<br />

du jury, les Professeurs Paul Delogne, Marco Luise <strong>et</strong> Piotr Sobieski, pour<br />

s’être investis dans une tâche qui en rebutait plus d’un. Leurs remarques<br />

pertinentes ont sensiblement contribué à l’amélioration du document final.<br />

Au sortir <strong>de</strong> mes étu<strong>de</strong>s universitaires, rien n’indiquait que le diplômé<br />

montois que je suis viendrait arpenter le plateau <strong>de</strong> Lauzelle. Je sais gré<br />

au Professeur Auguste Laloux <strong>de</strong> m’avoir ouvert les portes <strong>de</strong> Louvain-la-<br />

Neuve, me perm<strong>et</strong>tant ainsi d’accomplir un rêve d’enfant.<br />

S’il est vrai que le découragement menace souvent le doctorand, les nuages<br />

noirs ne s’attar<strong>de</strong>nt pas longtemps dans le ciel <strong>de</strong> TELE. La convivialité<br />

qui prévaut au laboratoire est l’écrin rêvé pour l’épanouissement <strong>de</strong>s<br />

compétences scientifiques qui s’y développent. Que tous ses membres,<br />

passés <strong>et</strong> présents, en soient remerciés. Je désire remercier tout spécia-


ii<br />

lement mes condisciples <strong>de</strong> bureau, Stéphane Pigeon, Laurent Cuvelier,<br />

Benoît Maison <strong>et</strong> François Deryck, pour l’atmosphère chaleureuse qu’ils y<br />

ont fait régner. Mes pensées vont aussi à ceux qui, dans l’ombre, œuvrent<br />

pour nous assurer le meilleur cadre <strong>de</strong> travail. J’adresse en outre mes<br />

voeux <strong>de</strong> succès à Mamoun Guenach, qui m<strong>et</strong> ses pas dans les miens un<br />

peu plus chaque jour.<br />

Ma reconnaissance va également au Fonds National <strong>de</strong> la Recherche Scientifique<br />

dont le soutien financier m’a permis <strong>de</strong> mener à bien c<strong>et</strong>te thèse <strong>de</strong><br />

doctorat, libéré <strong>de</strong> toute contingence matérielle.<br />

Un Special Award revient sans conteste à Sarah Zandona, qui s’est astreinte<br />

à plusieurs relectures méticuleuses, en dépit <strong>de</strong> son absence d’affinités<br />

avec le domaine. Sans elle, ce texte eût été d’une piètre qualité linguistique.<br />

Enfin, je ne peux terminer sans remercier du fond du coeur mes proches,<br />

qui m’accompagnent <strong>et</strong> me supportent <strong>de</strong>puis le premier jour.<br />

Laurent Schumacher<br />

Décembre 1999


Contents<br />

1 Introduction 1<br />

1.1 A paradigm shift ......................... 1<br />

1.2 Motivation ............................. 3<br />

1.3 Structure of the thesis ....................... 4<br />

1.4 Notations .............................. 7<br />

2 State of the art 9<br />

2.1 DS-CDMA, a technique whose time has come ......... 9<br />

2.1.1 Spread-spectrum in a nutshell ............. 9<br />

2.1.2 Applications ........................ 16<br />

2.2 Multiuser reception for DS-CDMA systems .......... 24<br />

2.2.1 D<strong>et</strong>ection .......................... 25<br />

2.2.2 Param<strong>et</strong>er estimation ................... 27<br />

2.2.3 Joint <strong>de</strong>tection and param<strong>et</strong>er estimation ....... 33<br />

2.3 Phase estimation ......................... 34<br />

2.3.1 Estimation structures ................... 34<br />

2.3.2 Performance characterisation of phase estimators .. 37<br />

2.3.3 Multiuser Phase estimation ............... 40<br />

2.4 Conclusions ............................ 43<br />

3 Tools 45<br />

3.1 System <strong>de</strong>scription ........................ 45<br />

3.1.1 System un<strong>de</strong>r investigation ............... 45<br />

3.1.2 Definition of Energy-to-Noise ratios .......... 49<br />

3.2 Maximum-Likelihood estimation ................ 51<br />

3.2.1 Maximum A Posteriori and Maximum-Likelihood .. 51<br />

3.2.2 Likelihood function ................... 52<br />

3.2.3 ML condition ....................... 57<br />

3.3 Optimal estimator performance ................. 58


iv CONTENTS<br />

3.3.1 Cramér-Rao Lower Bound ................ 58<br />

3.3.2 ML performance ..................... 60<br />

3.3.3 CRLB for multiuser phase estimation ......... 60<br />

3.4 FF estimation ........................... 62<br />

3.4.1 Closed form of the estimator .............. 62<br />

3.4.2 Variance approximation ................. 66<br />

3.5 Performance evaluation of DD estimators ........... 67<br />

3.5.1 Direct space - Gaussian probability integral ...... 69<br />

3.5.2 Reciprocal space - Characteristic function ....... 71<br />

3.6 Conclusions ............................ 72<br />

4 Data-Ai<strong>de</strong>d 75<br />

4.1 Feedback .............................. 76<br />

4.1.1 Open-loop study ..................... 78<br />

4.1.2 Closed-loop study .................... 84<br />

4.2 Feedforward ............................ 94<br />

4.2.1 Pdf of an SU estimator in a multiuser context ..... 94<br />

4.2.2 Linearised multiuser estimator in 2-user system ... 98<br />

4.3 Feedback-Feedforward correspon<strong>de</strong>nce ............103<br />

4.4 Conclusions ............................105<br />

5 Decision Directed 109<br />

5.1 Feedback ..............................110<br />

5.1.1 Decisions assumed correct ................112<br />

5.1.2 Actual <strong>de</strong>cisions - Open-loop study ..........116<br />

5.1.3 Actual <strong>de</strong>cisions - Closed-loop study .........141<br />

5.2 Feedforward ............................142<br />

5.2.1 SU DD ML FF estimator .................143<br />

5.2.2 MU DD ML FF estimator ................144<br />

5.2.3 Decisions assumed correct ................147<br />

5.2.4 Actual <strong>de</strong>cisions .....................148<br />

5.3 Conclusions ............................149<br />

6 Conclusions 151<br />

6.1 Achievements ...........................151<br />

6.2 Perspectives ............................152<br />

A Correlation function of the loop noise in a DA recovery loop 157<br />

A.1 BPSK modulation .........................158<br />

A.2 QPSK modulation .........................159


CONTENTS v<br />

B Pdf of Single-User DA ML FF phase estimator 161<br />

B.1 First step: characteristic function ψˆxu,ˆyu (ωr,ωi) ........161<br />

B.2 Second step: pdf Tˆxu,ˆyu (ˆxu, ˆyu) .................162<br />

B.3 Third step: change of variables .................164<br />

B.4 Fourth step: pdf T∆u(∆u) ....................165<br />

B.5 Analytical validation .......................165<br />

C Variance of DA ML FF phase estimators 167<br />

C.1 Multiuser estimator ........................167<br />

C.1.1 BPSK modulation .....................167<br />

C.1.2 QPSK modulation ....................169<br />

C.2 Single-user estimator .......................171<br />

C.2.1 BPSK modulation .....................171<br />

C.2.2 QPSK modulation ....................171<br />

D First or<strong>de</strong>r statistics in a linear channel 173<br />

D.1 Expectations of data ¢ <strong>de</strong>cision products ...........174<br />

D.1.1 User ¢ User ........................174<br />

D.1.2 User ¢ Interferer .....................176<br />

D.2 Expectations of <strong>de</strong>cision ¢ <strong>de</strong>cision products .........177<br />

D.2.1 Same I/Q branch .....................177<br />

D.2.2 Cross-talk .........................178<br />

D.3 Conclusion .............................178<br />

E Expectations for DD FB open-loop performance evaluation 179<br />

E.1 BPSK Modulation .........................179<br />

<br />

E.1.1 Derivation of E âm u an v ˆ <br />

Φ=0, Φ=∆ .........180<br />

<br />

E.1.2 Derivation of E âm u ânv ˆ <br />

Φ=0, Φ=∆ .........182<br />

E.2 QPSK Modulation .........................184<br />

<br />

E.2.1 Derivation of E âm u an v ˆ <br />

Φ=0, Φ=∆ .........185<br />

<br />

E.2.2 Derivation of E âm u bnv ˆ <br />

Φ=0, Φ=∆ .........187<br />

<br />

E.2.3 Derivation of E âm u ân v ˆ <br />

Φ=0, Φ=∆ .........188<br />

<br />

E.2.4 Derivation of E âm u ˆb n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆ .........189<br />

F Expressions of Uu,DD <strong>de</strong>rived in the reciprocal space 191<br />

F.1 BPSK modulation .........................191<br />

F.2 QPSK modulation .........................194


vi CONTENTS<br />

G COST 207 Channel Mo<strong>de</strong>ls 197<br />

H Curriculum vitae 199<br />

Bibliography 203


List of Figures<br />

1.1 Voice traffic on public n<strong>et</strong>works continues to grow at steady,<br />

predictable rates, while data traffic is growing exponentially<br />

and may surpass voice traffic in many countries by the year<br />

2000 (Source: [3]) ......................... 3<br />

2.1 Block diagram of a digital DS/SS transmitter for radio communications<br />

............................ 10<br />

2.2 Spectrum of a DS/SS signal (Bandwidth expansion factor = 4) 11<br />

2.3 Block diagram of a digital DS/SS receiver for radio communications<br />

.............................. 11<br />

2.4 Spectrum of a FH/SS signal (Bandwidth expansion factor =<br />

4) .................................. 12<br />

2.5 Representation of FDMA, TDMA and CDMA in the timefrequency<br />

plane (Source: [12]) .................. 14<br />

2.6 Illustration of user separation in DS-CDMA systems ..... 15<br />

2.7 UMTS components ........................ 18<br />

2.8 Path diversity (Source: [25]) ................... 21<br />

2.9 Some applications of CDMA nowadays ............ 25<br />

2.10 MUD systems <strong>de</strong>scribed in [41, 42] ............... 26<br />

2.11 DA estimator ........................... 28<br />

2.12 DD estimator ........................... 28<br />

2.13 NDA estimator .......................... 29<br />

2.14 FB and FF implementations ................... 36<br />

2.15 Hang-up and cycle slip ...................... 39<br />

3.1 Uplink of a coherent CDMA communication system ..... 46<br />

3.2 Sub-domains in the plane (νm u , νn v ) ............. 70<br />

4.1 2-user DA phase recovery loop ................. 77


viii LIST OF FIGURES<br />

4.2 Power spectral <strong>de</strong>nsity of Additive Noise, Self- and Cross-<br />

Noise ................................ 83<br />

4.3 DA BPSK PLL ........................... 86<br />

4.4 Inci<strong>de</strong>nce of the quadratic term of the Taylor-series expansion<br />

at equilibrium of the variance expression ......... 88<br />

4.5 Variance of DA FB estimators in AWGN channel (BPSK) .. 89<br />

4.6 Near-Far effect on DA FB estimators (BPSK) .......... 90<br />

4.7 Variance of DA FB estimators in dispersive channels (BSPK) 91<br />

4.8 Variance of DA FB estimators in AWGN channel (QPSK) .. 92<br />

4.9 Near-Far effect on DA FB estimators (QPSK) ......... 92<br />

4.10 Variance of DA FB estimators in dispersive channels (QSPK) 93<br />

4.11 Pdf of the SU DA ML FF phase estimate in a 2-user, RA<br />

channel context .......................... 96<br />

4.12 Variances of the SU DA ML FF phase estimation error as a<br />

function of the number of user Nu and of the channel type . 97<br />

4.13 Variance of DA FF estimators in an AWGN channel .....102<br />

4.14 Near-Far effect on DA FF estimators ..............103<br />

4.15 Inci<strong>de</strong>nce of ISI on DA FF estimators ..............104<br />

4.16 Correspon<strong>de</strong>nce b<strong>et</strong>ween DA FB and FF estimators .....107<br />

5.1 2-user DD phase recovery loop .................111<br />

5.2 Variance of DD ML FB estimators in ISI-free scenario (BPSK) 114<br />

5.3 Variance of DD ML FB estimators in presence of ISI (BPSK) . 115<br />

5.4 Variance of DD ML FB estimators in ISI-free scenario (QPSK) 116<br />

5.5 Variance of DD ML FB estimators in presence of ISI (QPSK) . 117<br />

5.6 S-curves in a 2-user non-dispersive synchronous system, xv,u =<br />

0 ...................................121<br />

5.7 S-surfaces of a 2-user non-dispersive synchronous system,<br />

uncoupled scenario (a: BPSK, b: QPSK) ............122<br />

5.8 S-curves function of ∆u, param<strong>et</strong>rised on ∆v - 2-user nondispersive<br />

synchronous system, uncoupled scenario (a: BPSK,<br />

b: QPSK) ..............................123<br />

5.9 S-curves function of ∆v, param<strong>et</strong>rised on ∆u - 2-user nondispersive<br />

synchronous system, uncoupled scenario (a: BPSK,<br />

b: QPSK) ..............................124<br />

5.10 S-surfaces of a 2-user non-dispersive synchronous system,<br />

coupled scenario (a: BPSK, b: QPSK) ..............125<br />

5.11 S-curves function of ∆u, param<strong>et</strong>rised on ∆v - 2-user nondispersive<br />

synchronous system, coupled scenario (a: BPSK,<br />

b: QPSK) ..............................126


LIST OF FIGURES ix<br />

5.12 S-curves function of ∆v, param<strong>et</strong>rised on ∆u - 2-user nondispersive<br />

synchronous system, coupled scenario (a: BPSK,<br />

b: QPSK) ..............................127<br />

5.13 S-surfaces of a 2-user non-dispersive synchronous system,<br />

Near-Far scenario (a: BPSK, b: QPSK) .............128<br />

5.14 S-curves function of ∆u, param<strong>et</strong>rised on ∆v - 2-user nondispersive<br />

synchronous system, Near-Far scenario (a: BPSK,<br />

b: QPSK) ..............................129<br />

5.15 S-curves function of ∆v, param<strong>et</strong>rised on ∆u - 2-user nondispersive<br />

synchronous system, Near-Far scenario (a: BPSK,<br />

b: QPSK) ..............................130<br />

5.16 U BPSK<br />

u,DD (0) as a function of δv,u (- simulation, ¢ computation<br />

direct space, Æ computation reciprocal space) ......133<br />

5.17 Phasor contributions of user u and interferer v to matched<br />

filter output ym u for BSPK-modulated data symbols ......134<br />

QP SK<br />

5.18 Uu,DD (0) as a function of δv,u (- simulation, ¢ computation<br />

direct space, Æ computation reciprocal space) ......136<br />

5.19 Phasor contributions of user u and interferer v to matched<br />

filter output y m u<br />

5.20 U BPSK<br />

u,DD<br />

5.21 U BPSK<br />

u,DD<br />

for QSPK-modulated data symbols .....136<br />

=30dB ....139<br />

where user u is the strongest and Eb<br />

N0<br />

where user u is the weakest and Eb<br />

N0<br />

=10dB .....140<br />

5.22 2-user DD phase recovery loop .................142<br />

5.23 SU DD ML FF estimator - Fastest update implementation . . 143<br />

5.24 SU DD ML FF estimator - Slow update implementation ...144<br />

5.25 2-user parallel MU DD ML FF estimator ............145<br />

5.26 2-user successive MU DD ML FF estimator ..........146<br />

5.27 Variance of ML FF estimators in presence of ISI (BSPK) ...149<br />

G.1 The Rural Area (RA) channel mo<strong>de</strong>l is ma<strong>de</strong> of 6 taps and<br />

its power <strong>de</strong>lay profile spreads over 0.5 µs. ..........197<br />

G.2 The Typical Urban (TU) channel mo<strong>de</strong>l is ma<strong>de</strong> of 12 taps<br />

and its power <strong>de</strong>lay profile spreads over 5 µs .........198<br />

G.3 The Hilly Terrain (HT) channel mo<strong>de</strong>l is ma<strong>de</strong> of 12 taps<br />

and its power <strong>de</strong>lay profile spreads over 20 µs. ........198


List of Tables<br />

2.1 Air interface param<strong>et</strong>ers of IS-95, cdmaOne and WCDMA<br />

(Sources: [4, 5, 19]) ........................ 19<br />

2.2 Comparison of 2nd-generation Globalstar and Ellipso and<br />

3rd-generation Skybridge (Source: [21, 25, 26, 28]) ...... 22<br />

2.3 Analog and digital phase recovery implementations ..... 35<br />

4.1 Asymptotical variance expressions of DA estimators in a 2user<br />

case ..............................106<br />

6.1 Synth<strong>et</strong>ic view of the achievements of the thesis .......151


LIST OF ABBREVIATIONS xiii<br />

List of abbreviations<br />

ACRB Asymptotic CRLB<br />

ATM Asynchronous Transfer Mo<strong>de</strong><br />

AWGN Additive White Gaussian Noise<br />

BER Bit Error Rate<br />

BPSK Binary Phase Shift Keying<br />

BRAN Broadband Radio Access N<strong>et</strong>work<br />

BS Base Station<br />

CATV Community Area TeleVision<br />

CDG CDMA Development Group<br />

CDMA Co<strong>de</strong> Division Multiple Access<br />

COST European Cooperation in the field of Scientific and<br />

Technical Research<br />

CRLB Cramér-Rao Lower Bound<br />

CSMA/CA Carrier Sense Multiple Access with Collision Avoidance<br />

DA Data Ai<strong>de</strong>d<br />

DD Decision Directed<br />

DFE Decision-Feedback Equaliser<br />

DOA Direction Of Arrival<br />

DS Direct Sequence<br />

EKF Exten<strong>de</strong>d Kalman Filtering<br />

EM Expectation-Maximisation<br />

ETSI European Telecommunications Standards Institute<br />

EVD Eigenvalue Decomposition<br />

FB Feedback<br />

FC Full-Carrier<br />

FDD Frequency Division Duplex<br />

FDMA Frequency Division Multiple Access<br />

FF Feedforward<br />

FH Frequency Hopping<br />

FPLMTS Future Public Land Mobile Telecommunication System<br />

GEO Geostationary Earth Orbit<br />

GMPCS Global Mobile Personal Communications by Satellite<br />

GSM Global System for Mobile communications<br />

HIPERLAN High Performance Radio Local Area N<strong>et</strong>work<br />

HT Hilly-Terrain<br />

IMT-2000 International Mobile Telecommunications 2000


xiv LIST OF ABBREVIATIONS<br />

IS Intermediate Standard<br />

ISI Inter-Symbol Interference<br />

ISM Industrial, Scientific and Medical<br />

ISO International Organisation for Standardization<br />

ITU International Telecommunication Union<br />

JD Joint D<strong>et</strong>ection<br />

LAN Local Area N<strong>et</strong>work<br />

LEO Low Earth Orbit<br />

LOS Line-Of-Sight<br />

MAI Multiple Access Interference<br />

MAP Maximum A Posteriori<br />

MC-CDMA Multi-Carrier CDMA<br />

MCRB Modified CRLB<br />

MEO Medium Earth Orbit<br />

MF Matched Filter<br />

MIPS Million of Instructions Per Second<br />

ML Maximum-Likelihood<br />

MSE Mean Square Error<br />

MMSE Minimum MSE<br />

MU Multiuser<br />

MUD Multiuser D<strong>et</strong>ection<br />

MUSIC Multiple Signal Classification<br />

NDA Non Data Ai<strong>de</strong>d<br />

OFDM Orthogonal Frequency Division Multiplex<br />

OHG Operators Harmonisation Group<br />

OSI Open Systems Interconnection<br />

PDA Personal Digital Assistant<br />

pdf Probability Density Function<br />

PIC Parallel Interference Canceller<br />

PLL Phase Locked Loop<br />

psd Power Spectral Density<br />

QPSK Quaternary Phase Shift Keying<br />

RA Rural Area<br />

RLS Recursive Least Squares<br />

S-CDMA Synchronous-CDMA<br />

SC Suppressed-Carrier<br />

SAGE Space-Alternating Generalised EM<br />

SIC Successive Interference Canceller<br />

SIR Signal-to-Interference Ratio<br />

SMR Signal-to-Multipath Ratio


LIST OF ABBREVIATIONS xv<br />

SNIR Signal-to-Noise-and-Interference Ratio<br />

SNR Signal-to-Noise Ratio<br />

SS Spread Spectrum<br />

SU Single-User<br />

SVD Singular Value Decomposition<br />

TD-CDMA Time/Co<strong>de</strong> Division Multiple Access<br />

TDD Time Division Duplex<br />

TDMA Time Division Multiple Access<br />

UMTS Universal Mobile Telecommunication Service<br />

UTRA UMTS Terrestrial Radio Access<br />

WATM Wireless ATM<br />

WCDMA Wi<strong>de</strong>band CDMA<br />

WDM Wavelength Division Multiplexing<br />

WLAN Wireless LAN<br />

ZF Zero-Forcing


Chapter 1<br />

Introduction<br />

1.1 A paradigm shift<br />

With his Technology Reports [1], George Gil<strong>de</strong>r is a respected but feared observer<br />

of the infocom world. Owners of the techniques he supports never<br />

fail to praise him, but he is as well strongly criticised for his views on technologies<br />

he believes will not be successful. There was thus no surprise<br />

to see the mix of enthusiastic and negative reactions that his article ”Telecosm<br />

and Beyond: Over the Paradigm Cliff” [2] generated when it was<br />

published in February 1997, as he discussed nothing less than a paradigm<br />

shift rooted in the tra<strong>de</strong>-off b<strong>et</strong>ween power and bandwidth inherent in<br />

Shannon’s law.<br />

According to this law, every engineer willing to perform reliable transmissions<br />

of information over a noisy channel has to make the most efficient<br />

balance b<strong>et</strong>ween power and bandwidth. In or<strong>de</strong>r to reach a certain bit rate<br />

while mitigating the effect of the noise, s/he can either increase the transmitted<br />

power within a limited bandwidth or use a wi<strong>de</strong>r bandwidth with<br />

a limited power.<br />

For <strong>de</strong>ca<strong>de</strong>s, the latter option has not been consi<strong>de</strong>red. In the ”Industrial<br />

Age”, as George Gil<strong>de</strong>r calls it, bandwidth was regar<strong>de</strong>d as scarce,<br />

and power abundant. It may sound strange to consi<strong>de</strong>r that the electromagn<strong>et</strong>ic<br />

spectrum might suffer from scarcity since it is physically infinite.<br />

However, the fact that engineers thought they had to <strong>de</strong>al with a bandwidth<br />

shortage had more to do with the way the use of the spectrum was<br />

constrained by government regulations. Only the lower part of the spec-


2 Introduction<br />

trum was consi<strong>de</strong>red at that time and it was then shared on an exclusive<br />

base b<strong>et</strong>ween all kinds of application. As a result, transmissions were to<br />

occur over narrow noisy channels plagued by interference. Quality of service<br />

was then ensured by exploiting the remaining <strong>de</strong>grees of freedom,<br />

namely emitting and switching powers.<br />

In<strong>de</strong>ed, service provi<strong>de</strong>rs were first inclined to boost up power within the<br />

limited frequency band allocated to their applications in or<strong>de</strong>r to overcome<br />

the poor quality of their noisy channel. However, this strategy alone<br />

could not provi<strong>de</strong> enough channels with enough quality to sustain the<br />

increasing <strong>de</strong>mand for communications. In or<strong>de</strong>r to accommodate more<br />

services and more users per service within a band-limited environment,<br />

n<strong>et</strong>work switches were then s<strong>et</strong> on work. Their aim was to improve the<br />

efficiency of the time-frequency resource sharing mechanism. It en<strong>de</strong>d<br />

up in exclusive sharing techniques such as Frequency Division Multiple<br />

Access (FDMA) and Time Division Multiple Access (TDMA). For George<br />

Gil<strong>de</strong>r, the paradigm of this time period was ”long and strong”: long<br />

wavelengths, i.e. small and low frequency bands, and strong power. Watts<br />

and MIPS helped to overcome the bandwidth shortage.<br />

These multiple access mechanisms have been <strong>de</strong>signed in times when<br />

analog-based voice services were dominant. However, the move from<br />

analog to digital processing led to the <strong>de</strong>velopment of data services. Y<strong>et</strong>,<br />

data transmissions patterns do not necessarily match with patterns <strong>de</strong>signed<br />

to fit voice applications. As the share of data applications rises,<br />

and that of voice applications correspondingly <strong>de</strong>clines (Figure 1.1), the<br />

shortcomings of the use of powerful transmitters and switches in or<strong>de</strong>r to<br />

overcome the bandwidth scarcity become obvious.<br />

On the other hand, communications have entered an era of bandwidth<br />

abundance on the wired scene as well as on the wireless one. Fiber optics<br />

<strong>de</strong>velopments <strong>de</strong>monstrate ever increasing throughput, while the constraints<br />

that restricted the use of the electro-magn<strong>et</strong>ic spectrum are being<br />

lifted as government regulations loosen their grip. Techniques enabling<br />

to exploit this bandwidth abundance have emerged: Wavelength Division<br />

Multiplexing (WDM) on optical fibers and Co<strong>de</strong> Division Multiple Access<br />

(CDMA) in the wireless world.<br />

Focusing on wireless applications, George Gil<strong>de</strong>r claims that CDMA is the


1.2 Motivation 3<br />

Figure 1.1: Voice traffic on public n<strong>et</strong>works continues to grow at steady,<br />

predictable rates, while data traffic is growing exponentially and may surpass<br />

voice traffic in many countries by the year 2000 (Source: [3])<br />

wireless access technique of the ”Information Age” in that it appropriately<br />

responds to the new balance b<strong>et</strong>ween abundance and scarcity: abundance<br />

of bandwidth, scarcity of power. In<strong>de</strong>ed, present and future <strong>de</strong>vices exhibit<br />

more and more stringent constraints on their power requirements,<br />

wh<strong>et</strong>her one speaks of portable <strong>de</strong>vices or of on-board switches. Power<br />

is no longer the key factor it used to be. This role has been overtaken by<br />

bandwidth. Communications are eager to be convoyed by weak signals<br />

using wi<strong>de</strong> bandwidths, leading to a new paradigm for the ”Information<br />

Age”, ”wi<strong>de</strong> and weak”.<br />

1.2 Motivation<br />

Although praising George Gil<strong>de</strong>r’s brilliant <strong>de</strong>monstration, many critics<br />

pointed out its <strong>de</strong>ficiencies. The least of them is not that his <strong>de</strong>monstration<br />

is too technology-oriented, implicitly un<strong>de</strong>restimating the weight of<br />

business constraints in the success of technical solutions. In the real world,<br />

they said, the success of a technique does not <strong>de</strong>pend only on its own<br />

merits, but also involves commercial issues, which George Gil<strong>de</strong>r has not


4 Introduction<br />

taken into account. Among these issues is the question wh<strong>et</strong>her end-users<br />

can sustain the abundant bandwidth which will provi<strong>de</strong> them with ever<br />

increasing data flows. In<strong>de</strong>ed, another limiting factor now comes in the<br />

picture, besi<strong>de</strong>s power and bandwidth: the information processing power<br />

of a human being.<br />

The subject of this thesis, however, is not to sort out the pros and contras<br />

of CDMA in view of all the constraints that <strong>de</strong>fine a successful communication<br />

system. Its aim is to gather knowledge about the estimation of an explicit<br />

synchronisation param<strong>et</strong>er, viz. the phase in the uplink of a wireless<br />

coherent Direct-Sequence CDMA (DS-CDMA) communication system.<br />

Dealing with the uplink leads to a situation where the Base Station (BS) is<br />

at the receiving end. From this point of view it is valuable to <strong>de</strong>sign an<br />

estimation structure that will simultaneously encompass all users active<br />

in the system instead of focusing on one user at a time and neglecting the<br />

others as interferers. The main objective of this thesis is to <strong>de</strong>monstrate<br />

analytically that the information content of the Multiple Access Interference<br />

(MAI) can be used to improve the quality of the estimation.<br />

However, one might argue that this work is only of aca<strong>de</strong>mic interest,<br />

since reception is not performed coherently in IS-95, the wireless spreadspectrum<br />

communication system currently in commercial use [4, p. 544].<br />

In<strong>de</strong>ed, while a pilot signal <strong>de</strong>dicated to each mobile receiver is inserted<br />

in the downlink, such facility is not used in the uplink by fear of interference.<br />

Y<strong>et</strong>, <strong>de</strong>velopers of third-generation systems are consi<strong>de</strong>ring to<br />

perform coherent reception also in the uplink [5]. This kind of reception<br />

will be ma<strong>de</strong> easier by the insertion of time-multiplexed pilot signals. As<br />

a result, efficient estimators of the phase param<strong>et</strong>er are required.<br />

1.3 Structure of the thesis<br />

This thesis is divi<strong>de</strong>d in four main chapters and seven appendices. The<br />

appendices <strong>de</strong>tail the mathematical <strong>de</strong>velopments leading to the relations<br />

studied and illustrated in the chapters of this thesis.<br />

Chapter 2 will introduce the issue of phase estimation in multiuser spreadspectrum<br />

context. It consists of three sections, each of them <strong>de</strong>aling with<br />

an aspect of this problematic.


1.3 Structure of the thesis 5<br />

The first section will present spread-spectrum communication techniques<br />

and focus on the Direct-Sequence Spread-Spectrum (DS/SS) technique.<br />

It will be shown that DS-CDMA is a technique well suited for providing<br />

multiple access to communication resource. Although this work only<br />

<strong>de</strong>als with mobile applications of DS-CDMA, its possible applications in<br />

several other communication environments, wired and wireless, will be<br />

briefly reviewed.<br />

DS-CDMA has long been regar<strong>de</strong>d as a m<strong>et</strong>hod for providing multiple<br />

access without having to <strong>de</strong>sign multiuser receivers. A simple correlator<br />

can perform users separation thanks to the inherent orthogonality of the<br />

users in DS-CDMA. However, as the technique gained in popularity, it appeared<br />

that this simple correlator had some shortcomings. Asynchronous<br />

transmissions, dispersive channels, highly loa<strong>de</strong>d systems, and power imbalance<br />

b<strong>et</strong>ween users plague the system with self- and cross-interference,<br />

respectively Inter-Symbol Interference (ISI) and MAI. This has led to a shift<br />

in the <strong>de</strong>sign of the receivers. Noticing that interference has an informative<br />

structure, contrary to additive white noise, <strong>de</strong>velopers started to regard interference<br />

as a useful contribution whose exploitation might improve the<br />

performance of the receiver. This new approach has been applied for both<br />

symbol <strong>de</strong>tection and param<strong>et</strong>er estimation, as it will be <strong>de</strong>scribed in the<br />

second section of Chapter 2.<br />

The scope of this work, however, will be limited to the param<strong>et</strong>er estimation<br />

issue, and more precisely, to the estimation of the phase param<strong>et</strong>er.<br />

The third section of Chapter 2 will review the proposed estimation structures<br />

and the means to perform the characterisation of their performance.<br />

None of the three sections of Chapter 2 briefly <strong>de</strong>scribed here above claims<br />

to be a thorough presentation of the related issue. They are rather broad<br />

overviews aimed at introducing the subject to newcomers and at pointing<br />

to relevant references.<br />

Following Chapter 2, Chapter 3 will first compl<strong>et</strong>e the s<strong>et</strong>ting of this<br />

work’s background. The communication system in which the phase estimation<br />

issue has been studied will be presented. Notations will be s<strong>et</strong> for the<br />

following chapters. Elements of estimation theory will be given in or<strong>de</strong>r<br />

to introduce the chosen m<strong>et</strong>hod of Maximum-Likelihood (ML) estimation.


6 Introduction<br />

Since the phase jitter variance is the performance benchmark of the following<br />

<strong>de</strong>velopments, a lower-bound, the Cramér-Rao Lower Bound (CRLB)<br />

will be introduced. Some problems faced during the study as well as the<br />

tricks consi<strong>de</strong>red to alleviate them will be <strong>de</strong>scribed afterwards.<br />

The background of this thesis having thus been explained in the previous<br />

two chapters, its subject, namely the estimation of the phase param<strong>et</strong>er<br />

in a multiuser spread-spectrum environment, shall then be tackled. The<br />

central theme of the next chapters will be the relationship b<strong>et</strong>ween <strong>de</strong>tection<br />

and estimation stages.<br />

In Chapter 4 the param<strong>et</strong>er estimation will rely on a perfect knowledge<br />

of transmitted symbols. This situation, called Data-Ai<strong>de</strong>d (DA) estimation,<br />

occurs in training periods when transmitter and receiver exchange<br />

pre<strong>de</strong>fined sequences aimed at helping to characterise the communication<br />

environment. The specificity of this chapter lies in that it starts from the<br />

premise that the <strong>de</strong>tection stage has no inci<strong>de</strong>nce on the estimation one. In<br />

this context two different implementations of the ML phase estimator will<br />

be consi<strong>de</strong>red: feedback (FB) and feedforward (FF).<br />

Contrary to Chapter 4, Chapter 5 will take into account possible interactions<br />

b<strong>et</strong>ween <strong>de</strong>tection and estimation stages. Firstly, <strong>de</strong>cisions will be<br />

assumed to be correct and the difference b<strong>et</strong>ween DA and DD estimators<br />

due to causality will be un<strong>de</strong>rlined. Secondly, the assumption of correct<br />

<strong>de</strong>cisions will be lifted and the inci<strong>de</strong>nce of <strong>de</strong>cision errors on the openloop<br />

performance of a Decision-Directed (DD) ML FB estimator will be<br />

<strong>de</strong>rived and illustrated. The closed-loop study will be mentioned in or<strong>de</strong>r<br />

to illustrate the benefit of multiuser estimation.<br />

Some <strong>de</strong>velopments have been ma<strong>de</strong> in the field of Non Data Ai<strong>de</strong>d (NDA)<br />

estimation while the present thesis was being written. They are nevertheless<br />

much too incompl<strong>et</strong>e to be presented here.<br />

By way of conclusion, the achievements of this thesis will be summarised<br />

and potential future <strong>de</strong>velopments will be outlined.


1.4 Notations 7<br />

1.4 Notations<br />

The following typographic conventions are used throughout this work unless<br />

explicitly specified otherwise.<br />

Scalar variables are <strong>de</strong>noted by normal-faced symbols, while vectors and<br />

arrays are <strong>de</strong>noted by bold-faced symbols:<br />

x or x (n) <strong>de</strong>notes a scalar variable<br />

x <strong>de</strong>notes a vector or an array<br />

Accents are wi<strong>de</strong>ly used:<br />

ˆx <strong>de</strong>notes the estimate of variable x<br />

x ⋆ <strong>de</strong>notes the complex conjugate of variable x<br />

As far as operators are concerned, the following notations are used:<br />

Pr (X >0) <strong>de</strong>notes the probability that the random variable X<br />

is positive<br />

E (X) <strong>de</strong>notes the expectation of the random variable X<br />

Finally, s<strong>et</strong>s of variables are <strong>de</strong>noted by curly braces xk.


Chapter 2<br />

State of the art<br />

2.1 DS-CDMA, a technique whose time has come<br />

2.1.1 Spread-spectrum in a nutshell<br />

Single-user perspective<br />

The main and common characteristic of spread-spectrum techniques [4,<br />

6, 7] is the expansion of the modulated symbol bandwidth from the minimum<br />

required to transmit the information to a wi<strong>de</strong>r one. However these<br />

techniques differ in the way they use this wi<strong>de</strong>r bandwidth.<br />

DS/SS system When the whole bandwidth is permanently occupied by<br />

the spread signal, one speaks of Direct Sequence Spread-Spectrum (DS-<br />

/SS). The bandwidth expansion is piloted by a periodic co<strong>de</strong> sequence<br />

whose Nc elementary components are called chips. Its rate, the chip rate<br />

1<br />

1<br />

T<br />

, is higher than the symbol rate Tc T , so that the ratio represents the<br />

Tc<br />

bandwidth spreading factor in the frequency domain. Usually one prefers<br />

to mention the processing gain which is the ratio Tb b<strong>et</strong>ween the chip rate<br />

Tc<br />

and the bit rate [8]. The bandwidth spreading factor is thus the product of<br />

the processing gain by the dimension of the data modulation T 1 . Diffe-<br />

Tb<br />

1 Authors of [8] note that the bandwidth spreading factor and the processing gain are<br />

som<strong>et</strong>imes confused, both of them being used to <strong>de</strong>signate the ratio b<strong>et</strong>ween the spread<br />

bandwidth and the original one (see for instance [9, p. 1]). Moreover, calculating the processing<br />

gain as a bandwidth ratio is regar<strong>de</strong>d in [4, chapter 2] as an approximation of the<br />

true processing gain, <strong>de</strong>fined as a measure of the performance improvement achieved by<br />

systems using spread-spectrum techniques with respect to systems not using them.


10 State of the art<br />

rent possibilities exist as regards the ratio b<strong>et</strong>ween the co<strong>de</strong> sequence N cTc<br />

and the symbol length T . The present work is only concerned with co<strong>de</strong><br />

sequences whose period is equal to the symbol length (NcTc = T ).<br />

A digital DS/SS transmitter is presented in Figure 2.1. The spreading operation<br />

is performed in two steps. First, the rate of the information sequence<br />

( 1<br />

1<br />

T ) is increased up to the rate of the spreading sequence ( ). This<br />

Tc<br />

increased rate sequence is then passed through a filter whose taps are the<br />

chips of the co<strong>de</strong> sequence. The sequence output by this filter is the spread<br />

sequence, ready to be shaped and transmitted.<br />

1, -1, 1<br />

Nc<br />

Tc<br />

T T<br />

Tc<br />

Chip<br />

shaping<br />

Figure 2.1: Block diagram of a digital DS/SS transmitter for radio communications<br />

Since the time resolution of the co<strong>de</strong> sequence is Nc times higher than the<br />

one of the symbol sequence, the spreading operation provokes a corresponding<br />

bandwidth expansion in the frequency domain (Figure 2.2). As<br />

a result, the spread signal tends to melt down into the background noise.<br />

It becomes thus hardly noticeable for third parties, achieving a low probability<br />

of interception. This ability to hi<strong>de</strong> pertinent information in the<br />

background noise is a first interesting characteristic of spread-spectrum<br />

transmissions.<br />

At the receiving end (Figure 2.3), the spread signal is correlated with a<br />

synchronised version of the periodic co<strong>de</strong> sequence so as to cancel the effect<br />

of spreading and to restore the original narrow symbol bandwidth.<br />

This <strong>de</strong>spreading operation <strong>de</strong>fines the strict requirements with which<br />

a periodic co<strong>de</strong> sequence has to comply. To be suitable for single-user<br />

spread-spectrum transmissions, the periodic co<strong>de</strong> sequence ought to exhibit<br />

Dirac-like auto-correlation properties. Pseudo-random co<strong>de</strong> sequences<br />

me<strong>et</strong> such requirements [10].


2.1 DS-CDMA, a technique whose time has come 11<br />

2<br />

Tc<br />

Figure 2.2: Spectrum of a DS/SS signal (Bandwidth expansion factor = 4)<br />

Chip-matched<br />

filter<br />

2<br />

T<br />

Nc<br />

1<br />

Nc<br />

1<br />

Figure 2.3: Block diagram of a digital DS/SS receiver for radio communications<br />

1,-1,1


12 State of the art<br />

While symbols are recovered at the receiving end by compacting their energy<br />

back into the symbol bandwidth, the correlation operation is in<strong>de</strong>ed<br />

a spreading operation by the bandwidth expansion factor for unspread<br />

incoming signals like narrowband interferers. This introduces another interesting<br />

feature of spread-spectrum techniques: the resistance to narrowband<br />

interferers.<br />

FH/SS system Instead of permanently using the whole bandwidth as<br />

DS/SS signals do, the frequency resource can be divi<strong>de</strong>d into slots as large<br />

as the symbol bandwidth (Figure 2.4).<br />

2<br />

Tc<br />

Figure 2.4: Spectrum of a FH/SS signal (Bandwidth expansion factor = 4)<br />

The co<strong>de</strong> sequence is then used to <strong>de</strong>fine the slot to be used for transmission,<br />

commanding jumps from one slot to the other over time. Hence the<br />

name of the technique: Frequency-Hopping Spread-Spectrum (FH/SS).<br />

On the one hand, the hop scheme is governed by the co<strong>de</strong> sequence. In<br />

or<strong>de</strong>r to make it hardly predictable, pseudo-random sequences are also<br />

used in FH/SS systems. On the other hand, the jump rate leads to distinguishing<br />

b<strong>et</strong>ween Slow-Frequency-Hopping (SFH) and Fast-Frequency-<br />

Hopping (FFH). In SFH systems, several symbols are transmitted b<strong>et</strong>ween<br />

two frequency jumps whereas several jumps occur within a single symbol<br />

duration in FFH transmissions [4, section 2.4].<br />

2<br />

T


2.1 DS-CDMA, a technique whose time has come 13<br />

Wh<strong>et</strong>her DS/SS or FH/SS, the bandwidth expansion of the transmitted<br />

signal is a common characteristic of spread-spectrum techniques. As mentioned<br />

earlier, they both exhibit a low probability of interception since<br />

the spread signal is hardly distinguishable from the background noise<br />

(DS/SS), or hard to catch due to the hops (FH/SS). These techniques are<br />

also robust against jammers thanks to the spreading of interferers at the<br />

<strong>de</strong>correlating end (DS/SS) or the avoidance of continuous emission within<br />

a jammed band (FS/SS). Those features are pr<strong>et</strong>ty interesting for military<br />

applications. That is why spread-spectrum techniques were first applied<br />

in military projects before moving to the civilian world in the last <strong>de</strong>ca<strong>de</strong>s.<br />

In the following sections only DS/SS techniques will be consi<strong>de</strong>red. Besi<strong>de</strong><br />

their resistance to narrowband interferers, another interesting property<br />

of DS/SS signals is their inherent frequency diversity. Since they occupy<br />

a large bandwidth, they are subject to frequency-selective fading [7,<br />

chapter 7]. The resulting multipath transmission can be exploited to improve<br />

the reception, using a RAKE receiver which properly combines the<br />

<strong>de</strong>layed versions of the transmitted signal [4, subsection 8-4.5].<br />

Multiuser perspective<br />

The preceding features of DS/SS communication systems, viz. resistance<br />

to narrowband interferers and frequency diversity, have been consi<strong>de</strong>red<br />

from a single-user perspective. Spread-spectrum techniques are also well<br />

suited for organising multiple access to digital transmissions.<br />

While conventional multiple access schemes like FDMA (Figure 2.5a) and<br />

TDMA (Figure 2.5b) share the time/frequency resources exclusively b<strong>et</strong>ween<br />

active users, DS-CDMA gives them the opportunity to share the<br />

same bandwidth at the same moment. Transmissions are spread over the<br />

time-frequency plane by co<strong>de</strong> multiplication (Figure 2.5c). This last approach<br />

seems more efficient when the resource is spare [11].<br />

To separate users, DS-CDMA communication systems rely on a proper<br />

choice of co<strong>de</strong> sequences whose mutual correlations ought to be as small<br />

as possible, so that <strong>de</strong>spreading with one co<strong>de</strong> sequence the signal spread<br />

with another one produces zeros. In Figure 2.6, the receiver <strong>de</strong>correlates<br />

the signal produced by the transmitter shown in Figure 2.1 using a co<strong>de</strong><br />

sequence other than the one used at the transmitting end. Since these two<br />

co<strong>de</strong> sequences are orthogonal, the <strong>de</strong>tector output is null.


14 State of the art<br />

(a) FDMA: a narrow frequency<br />

band per user<br />

(b) TDMA (with FDMA component):<br />

several time slots per<br />

frequency channel<br />

(c) CDMA: signal spread over<br />

time-frequency plane<br />

Figure 2.5: Representation of FDMA, TDMA and CDMA in the time-frequency plane (Source: [12])


2.1 DS-CDMA, a technique whose time has come 15<br />

Chip-matched<br />

filter<br />

Nc<br />

1<br />

Nc<br />

1<br />

Figure 2.6: Illustration of user separation in DS-CDMA systems<br />

As far as co<strong>de</strong> sequences are concerned, orthogonal ones are the best choice.<br />

However, they require perfectly synchronous and non-dispersive transmissions<br />

to compl<strong>et</strong>ely cancel MAI. Quasi-orthogonal sequences are thus<br />

often preferred to orthogonal ones due to their ability to still separate users<br />

when neither synchronism nor non-dispersiveness can be ensured.<br />

FDMA, TDMA and CDMA are thus different in the way they distinguish<br />

users. As a result, the limit on their capacity is also of a different nature<br />

[13].<br />

The number of users that FDMA/TDMA systems can accommodate is<br />

limited by the fractioning into user slots of the global time/frequency resource<br />

they can allocate. The exclusive allocation of these slots avoids interference.<br />

To guarantee that interference is avoi<strong>de</strong>d, FDMA/TDMA systems<br />

respect guard times/bands, which results in resource wasting. These<br />

techniques are thus said to be resource-limited [11].<br />

On the other hand, in DS-CDMA schemes, interference is unavoidable.<br />

Each new user is spread over the whole bandwidth and, as a consequence,<br />

raises the interference level up to a point where the Signal-to-Noise-and-<br />

Interference Ratio (SNIR) at the receiver is too poor to permit reception.<br />

As a result, DS-CDMA is interference-limited.<br />

Interfering users can be seen as wi<strong>de</strong>band jammers. A conventional singleuser<br />

receiver, relying only on the correlation properties of the co<strong>de</strong> sequence<br />

to recover the signal transmitted by one user, requires stringent<br />

power control so as to keep the interference level plaguing the reception<br />

below the minimum SNIR level required for reception. Should this power<br />

control fail, the single-user receiver would be unable to recover the trans-<br />

0,0,0


16 State of the art<br />

mitted signal. This happens with the Near-Far effect. It refers to a situation<br />

of power imbalance b<strong>et</strong>ween users which is encountered, for instance, in<br />

cellular communication systems [6] when a user situated far away from<br />

the BS is affected by the wi<strong>de</strong>band interference generated by another user<br />

located near the BS. Without power control to level users’ energies, the<br />

near user masks the far user at the BS. The Near-Far effect has long been<br />

thought to be an inherent limitation of DS-CDMA. In fact, it is due to the<br />

single-user <strong>de</strong>sign of the receiver, which postulates that MAI has been cancelled<br />

at the <strong>de</strong>correlator [14]. This question will be <strong>de</strong>veloped in the following<br />

sections.<br />

Other interesting features of spread-spectrum techniques that can be mentioned<br />

from a multiuser system perspective are the soft handover and the<br />

overlay over any existing system. Being able to combine several unsynchronised<br />

versions of the same signal, CDMA receivers can communicate<br />

with several BSs, and thus softly switch from one to the other when they<br />

change serving area. On the other hand, the wi<strong>de</strong>band spreading of the<br />

signals enables such systems to overlay over existing systems, as long as<br />

the new system appears as a low-power wi<strong>de</strong>band interferer for them.<br />

2.1.2 Applications<br />

Nowadays, spread-spectrum techniques have found their way in the communication<br />

world, mainly for wireless applications. This section will briefly<br />

review the communication systems already using spread-spectrum techniques<br />

or consi<strong>de</strong>ring to do so in the near future. The <strong>de</strong>scription of standards<br />

mentioned in the following section is limited to the physical layer<br />

(Layer 1 of International Organization for Standardization (ISO) [15] Open<br />

Systems Interconnection (OSI) reference mo<strong>de</strong>l), which is the global scene<br />

of the present work.<br />

Land mobile services<br />

Mobile applications are concerned with wireless connections within a cellular<br />

n<strong>et</strong>work, i.e. b<strong>et</strong>ween a mobile receiver able to move within the<br />

serving area at typical car speeds and base stations interfacing the mobile<br />

receiver with other mobiles and/or a fixed n<strong>et</strong>work. Two scenes are<br />

consi<strong>de</strong>red here: the terrestrial and the satellite.


2.1 DS-CDMA, a technique whose time has come 17<br />

Terrestrial-based cellular systems The first generation of terrestrial cellular<br />

system was analog. With second-generation systems like IS-95 and<br />

Global System for Mobile Communications (GSM), a switch has been ma<strong>de</strong><br />

from analog to digital. While GSM organises the multiple access according<br />

to a Frequency Division Duplex (FDD)/TDMA scheme, IS-95 standard [4,<br />

section 9-4.1] relies on CDMA. IS-95 <strong>de</strong>scribes a single-carrier DS-CDMA<br />

digital cellular system <strong>de</strong>dicated to voice and data services for rates up to<br />

9.6 to 14.4 kbps (up to 115.2 kbps with IS-95-B revision). The choice for<br />

CDMA has been ma<strong>de</strong> in or<strong>de</strong>r to enjoy interference rejection and spectral<br />

efficiency promised by this system, as illustrated in [13] which <strong>de</strong>rives capacity<br />

equations of a spread-spectrum terrestrial cellular n<strong>et</strong>work by taking<br />

into account the number of users, bit rates, the reuse factor, possible<br />

sectorisation, and inter-cell interference.<br />

Although second-generation n<strong>et</strong>works are still in <strong>de</strong>ployment in many<br />

places, third-generation systems are already un<strong>de</strong>r close investigation. The<br />

incentive of these research activities has been the <strong>de</strong>mand for speed connections<br />

over cellular n<strong>et</strong>works higher than what is affordable today with<br />

second-generation systems. The walk towards third-generation systems<br />

takes place in the framework of International Telecommunication Union<br />

(ITU)’s International Mobile Telecommunications-2000 (IMT-2000) 2 which<br />

<strong>de</strong>fines the requirements for next generation services: rates up to 144 kbps<br />

for vehicular applications, 384 kbps for pe<strong>de</strong>strian services and 2 <strong>Mb</strong>ps<br />

for indoor systems, with improved spectrum efficiency and service flexibility<br />

within 1,885-2,025 and 2,110-2,200 MHz frequency bands [16, 17]. All<br />

over the world, in IS-95 served areas as well as in GSM countries, spreadspectrum<br />

techniques are r<strong>et</strong>ained as candidates for implementing multiple<br />

access in those third-generation systems, with different issues according to<br />

the cellular n<strong>et</strong>works that are already installed.<br />

As far as IS-95 is concerned, the evolution is quite obvious. The next step,<br />

initiated by CDMA Development Group (CDG) [18], will be cdma2000 or<br />

Wi<strong>de</strong>band cdmaOne, heading to bit rates up to 2<strong>Mb</strong>ps. A crucial aspect<br />

of its implementation is the backward compatibility with IS-95. To accommodate<br />

higher speeds, nominal bandwidths have been enlarged from 1.25<br />

MHz in IS-95 to 5 MHz in Wi<strong>de</strong>band cdmaOne. However, a multicarrier<br />

scheme is <strong>de</strong>signed so as to enable overlay of Wi<strong>de</strong>band cdmaOne over<br />

2<br />

formerly known until 1996 as Future Public Land Mobile Telecommunication System<br />

(FPLMTS)


18 State of the art<br />

IS-95. Other enhancements to the physical layer are consi<strong>de</strong>red, among<br />

which coherent <strong>de</strong>modulation in the uplink, faster power control, use of<br />

turbo-co<strong>de</strong>s, and optional Multiuser D<strong>et</strong>ection (MUD) [19].<br />

On the other hand, the evolution from GSM to spread-spectrum techniques<br />

is not as natural as the move from IS-95 to Wi<strong>de</strong>band cdmaOne,<br />

since the nature of the multiple access scheme has to change. In Europe,<br />

where GSM was <strong>de</strong>veloped and where IMT-2000 translates into Universal<br />

Mobile Telecommunication Service (UMTS), the European Telecommunications<br />

Standards Institute (ETSI) [20] chose in early 1998 two terrestrial air<br />

interfaces (UTRA): Wi<strong>de</strong>band CDMA (WCDMA) for paired FDD bands<br />

(1,920-1,980 and 2,110-2,170 MHz), and Time/Co<strong>de</strong> Division Multiple Access<br />

(TD-CDMA) for unpaired Time Division Duplex (TDD) bands (1,900-<br />

1,920 and 2,010-2,025 MHz). The main issue is to ensure backward compatibility<br />

with GSM <strong>de</strong>spite the differences in multiple access schemes.<br />

This needs to be done by <strong>de</strong>riving the clock rates of the third-generation<br />

systems from the GSM clock rate (13 MHz or 26 MHz) and by placing carriers<br />

over a common frequency grid with 200 kHz-spacing.<br />

UMTS<br />

Terrestrial component (UTRA)<br />

WCDMA TD-CDMA<br />

Satellite component (S-UMTS)<br />

Figure 2.7: UMTS components<br />

Table 2.1 synthesises the main param<strong>et</strong>ers of the air interfaces for secondgeneration<br />

IS-95 and third-generation Wi<strong>de</strong>band cdmaOne and WCDMA 3 .<br />

3 An agreement on a globally harmonised third-generation CDMA radio standard was<br />

reached by the Operators Harmonisation Group (OHG) in May 1999 and later endorsed by<br />

all other concerned standardisation bodies. There should be three mo<strong>de</strong>s in the harmonised<br />

3G CDMA standard: a FDD single-carrier DS mo<strong>de</strong> for WCDMA, a FDD multi-carrier


Generation 2nd 3rd<br />

System IS-95 Wi<strong>de</strong>band cdmaOne WCDMA<br />

RF channel bandwidth [MHz] 1.25 1.25/5/10/15/20 5/10/20<br />

Downlink RF channel structure<br />

Chip rate [Mcps] 1.2288<br />

Direct spread<br />

1.2288/3.6864<br />

Multicarrier<br />

n¢ 1.2288 (n =1,3,<br />

Direct spread<br />

4.096/8.192/16.384<br />

/7.3728/11.0593<br />

/14.7456<br />

6, 9, 12)<br />

Frame length [ms] 20 20 (data and control)/5 (control information) 10/20 (optional)<br />

Data Downlink BPSK QPSK<br />

modulation Uplink 64-ary<br />

gonalortho-<br />

BPSK<br />

Coherent <strong>de</strong>- Downlink Common Common pilot channel + auxiliary pilots Common pitection<br />

pilot channel<br />

lot channel +<br />

Uplink - Time-multiplexed pilot<br />

user-<strong>de</strong>dicated<br />

time-multiplexed<br />

pilots<br />

User-<strong>de</strong>dicated<br />

time-multiplexed<br />

pilots<br />

Data rates 9.6-14.4 kbps<br />

(IS-95-A)<br />

9.6-115.2 kbps<br />

(IS-95-B)<br />

9.6 kbps - 2 <strong>Mb</strong>ps 128 kbps - 2 <strong>Mb</strong>ps<br />

Table 2.1: Air interface param<strong>et</strong>ers of IS-95, cdmaOne and WCDMA (Sources: [4, 5, 19])<br />

2.1 DS-CDMA, a technique whose time has come 19


20 State of the art<br />

Satellite-based cellular systems The success of cellular systems leads<br />

to a point where users are no longer asking for mobility but for ubiquity.<br />

Land-based cellular n<strong>et</strong>works are unfortunately not available everywhere.<br />

From this point of view, satellite-based cellular systems appear as complementary<br />

to land-based n<strong>et</strong>works, backing them up or even substituting<br />

them in non served areas.<br />

Satellite-based communication services have been available for many years<br />

now, mainly for broadcasting, using Geostationary Earth Orbit (GEO) systems.<br />

However, the GEO orbit (35,860 km above the Earth) is not the most<br />

appropriate for the mobile applications targ<strong>et</strong>ed nowadays due to latency<br />

(250 ms round-trip propagation <strong>de</strong>lay) and small link margins [21]. Lower<br />

orbits, like Low Earth Orbit (LEO, 160-480 km) and Medium Earth Orbit<br />

(MEO, 9,660-19,110 km), solve these issues but also introduce new problems.<br />

In<strong>de</strong>ed, such low orbit satellites move quickly in the sky above the<br />

ground user, which means that handover and correction of large Dopplershifts<br />

(up to 60 kHz for a satellite altitu<strong>de</strong> of 1,500 km at 2.4 GHz [22]) shall<br />

be <strong>de</strong>alt with.<br />

Moreover, such satellite-based systems work at frequencies higher than<br />

the land-based ones, from one to the tens GHz. At such frequencies,<br />

electro-magn<strong>et</strong>ic fields do not pen<strong>et</strong>rate buildings and are blocked by obstacles.<br />

As a result, transmission is only viable as long as Line-of-Sight<br />

(LOS) visibility is ensured.<br />

For several years, efforts have been ma<strong>de</strong> un<strong>de</strong>r ITU’s Global Mobile Personal<br />

Communications by Satellite (GMPCS) label in or<strong>de</strong>r to provi<strong>de</strong> voice<br />

and data services to hands<strong>et</strong>s worldwi<strong>de</strong> using LEO and MEO satellites<br />

[23].<br />

Commercial service was opened in 1998 by the Iridium consortium [24].<br />

Iridium offers voice and data services up to 9.6 kbps through a 66 LEO<br />

satellite-based cellular n<strong>et</strong>work. Multiple access is based on a FDD/TDMA<br />

scheme. Communications b<strong>et</strong>ween the user and the satellite occur in the<br />

L-band (1.616-1.6265 GHz), while satellites and gateways use K-bands<br />

(19.4-19.6 GHz and 29.1-29.3 GHz). An innovative feature in the Iridium<br />

system is the direct handling of calls from one satellite to the other without<br />

mo<strong>de</strong> for cdmaOne, and a TDD CDMA mo<strong>de</strong>. First and third mo<strong>de</strong>s will operate at 3.84<br />

Mcps chip rate, while the FDD multi-carrier will use 3.6864 Mcps chip rate.


2.1 DS-CDMA, a technique whose time has come 21<br />

ground-based relay.<br />

While Iridium is on the air, other second-generation satellite-based cellular<br />

systems are g<strong>et</strong>ting ready for opening services. Among them, Globalstar<br />

[25] and Ellipso [26] projects plan to offer multiple access based<br />

on CDMA. Thanks to spread-spectrum, a soft handover b<strong>et</strong>ween satellite<br />

footprints is possible, as already mentioned in the case of land-based cellular<br />

systems. Moreover, the receiver enjoys path diversity. With a RAKE<br />

receiver several versions of the same signals, transmitted by the different<br />

satellites in view, can be combined so as to improve reception quality and<br />

to avoid blocking (Figure 2.8).<br />

Figure 2.8: Path diversity (Source: [25])<br />

Similarly to the evolution in terrestrial-based applications, third-generation<br />

systems are already knocking at the door. Tele<strong>de</strong>sic [27] and Skybridge<br />

[28] are among the few projects known so far. The latter has to be<br />

mentioned in the current presentation, since it is based on CDMA. Skybridge<br />

announces the <strong>de</strong>ployment of a 64-LEO satellite-based cellular system<br />

working in Ku-band (12-18 GHz) and offering bit rates up to 60 <strong>Mb</strong>ps<br />

in the downlink, and 2 <strong>Mb</strong>ps in the uplink.


22 State of the art<br />

Generation 2nd 3rd<br />

Project Globalstar Ellipso Skybridge<br />

Operator Motorola MCH Inc., Alcatel<br />

Opening of service 1999<br />

Lockheed,<br />

Harris<br />

2001 2001<br />

Number of satellites 48 + 8 spares 6 equatorial +<br />

8 elliptical + 3<br />

spares<br />

64<br />

Orbit LEO MEO LEO<br />

User link Downlink S-band L-band Ku-band<br />

Uplink L-band<br />

Rates Downlink 9.6 kbps 60 <strong>Mb</strong>ps<br />

Uplink 9.6 kbps 2 <strong>Mb</strong>ps<br />

Table 2.2: Comparison of 2nd-generation Globalstar and Ellipso and 3rdgeneration<br />

Skybridge (Source: [21, 25, 26, 28])<br />

Cordless/portable/WLAN service provision<br />

The domain of cordless/portable communication <strong>de</strong>vices is the intermediate<br />

step from the mobile communication world to the fixed communication<br />

one, trying to combine both advantages, viz. mobility and high-speed<br />

transmissions. On the one hand, such portable <strong>de</strong>vices are giving the user<br />

some freedom of position and/or movement within a restricted serving<br />

area: static connections can be engaged from any point and low-speed<br />

mobility is ensured through n<strong>et</strong>work and handover management. On the<br />

other hand, these portable terminals are often regar<strong>de</strong>d as wireless gateways<br />

to backbone n<strong>et</strong>works, first Local Area N<strong>et</strong>works (LAN) then Asynchronous<br />

Transfer Mo<strong>de</strong> (ATM) n<strong>et</strong>works. They are thus expected to offer<br />

higher bit rates than the ones available through mobile <strong>de</strong>vices.<br />

In the late 1980’s and early 1990’s, the first wave of cordless/portable communication<br />

<strong>de</strong>vices, introduced un<strong>de</strong>r the tra<strong>de</strong>mark of Wireless LAN<br />

(WLAN), was expected to break through thanks to the ease of <strong>de</strong>ployment.<br />

In<strong>de</strong>ed, wireless connections enable to s<strong>et</strong> up a communication n<strong>et</strong>work<br />

without rewiring the communication scene. In fact, these <strong>de</strong>vices<br />

mainly gained popularity thanks to the connection possibilities they ad<strong>de</strong>d<br />

to portable <strong>de</strong>vices like Personal Digital Assistants (PDA). Among the<br />

several air interface solutions consi<strong>de</strong>red, spread-spectrum was r<strong>et</strong>ained<br />

for radio-based WLAN thanks to its ability to coexist with already implemented<br />

services in the band used for transmission. Moreover, promoters


2.1 DS-CDMA, a technique whose time has come 23<br />

of spread-spectrum claimed that it would enable cooperation of products<br />

from different vendors without prior dialogue. However, power control<br />

issues appeared to request such concertation [29]. Nevertheless, spreadspectrum<br />

techniques are implemented in the American IEEE 802.11 WLAN<br />

standard to provi<strong>de</strong> immunity with respect to narrowband interferers in<br />

the Industrial, Scientific and Medical (ISM) band (2.4 GHz). Multiple<br />

access is organised using Carrier Sense Multiple Access with Collision<br />

Avoidance (CSMA/CA) mechanism. The IEEE 802.11 standard provi<strong>de</strong>s<br />

bit rates up to 2 <strong>Mb</strong>ps [30]. Its equivalent in Europe is ETSI High Performance<br />

Radio Local Area N<strong>et</strong>work/1 (HIPERLAN/1) which was formalised<br />

in 1997. Unlike IEEE 802.11, multiple access techniques normalised<br />

in HIPERLAN/1 do not rely on CSMA/CA but on a FDMA/TDMA combination.<br />

HIPERLAN/1 addresses bit rates up to 23.529 <strong>Mb</strong>ps in the 5.15-<br />

5.30 GHz-band [31, 32].<br />

Nowadays, the next wave of cordless/portable applications, called Wireless<br />

ATM (WATM), is targ<strong>et</strong>ing bit rates much higher than the 2 <strong>Mb</strong>ps<br />

rate of IEEE 802.11. For applications offering such high bit rates, spreadspectrum<br />

techniques have been disregar<strong>de</strong>d. In<strong>de</strong>ed, with a bandwidth<br />

expansion factor as small as 15, spreading data symbols produced at 155<br />

<strong>Mb</strong>ps requires a prohibitive bandwidth of 2 GHz. Moreover, synchronisation<br />

issues become troublesome [33]. Other modulation schemes offering<br />

orthogonality b<strong>et</strong>ween users and immunity to multipath, like Orthogonal<br />

Frequency Division Multiplex (OFDM), have then been consi<strong>de</strong>red in the<br />

European Broadband Radio Access N<strong>et</strong>work (BRAN) project [34]. This<br />

project paves the way beyond HIPERLAN/1 in or<strong>de</strong>r to me<strong>et</strong> <strong>de</strong>mands for<br />

high bit rates transmission. Un<strong>de</strong>r the BRAN project, three different standards<br />

(HIPERLAN/2, HIPERACCESS and HIPERLINK) are <strong>de</strong>veloped for<br />

broadband cordless/portable communications. These standards targ<strong>et</strong> different<br />

bit rates (25-155 <strong>Mb</strong>ps) and environments (static/mobile, indoor-<br />

/outdoor).<br />

Nevertheless spread-spectrum techniques have not compl<strong>et</strong>ely disappeared<br />

from the cordless/portable communication scene. The parallel transmission<br />

implemented in the OFDM scheme provi<strong>de</strong>s a means of spreading<br />

at chip rates lower than the ones requested by DS/SS techniques used<br />

alone. A mix of OFDM and DS/SS, named Multi-Carrier CDMA (MC-<br />

CDMA), is a solution un<strong>de</strong>r investigation [33].


24 State of the art<br />

Fixed service provision<br />

The last application to be mentioned in this section is <strong>de</strong>livery of highspeed<br />

data services over cable TV coaxial n<strong>et</strong>works. While the applications<br />

of spread-spectrum techniques for providing communication services<br />

<strong>de</strong>scribed so far are all wireless, CDMA has also found its way in<br />

the wired world in or<strong>de</strong>r to help Community Area Television (CATV) operators<br />

to turn into multimedia service provi<strong>de</strong>rs. Since the <strong>de</strong>mand for<br />

such value-ad<strong>de</strong>d services have been i<strong>de</strong>ntified, these operators have invested<br />

much time and money to adapt their n<strong>et</strong>works so that they could<br />

provi<strong>de</strong> these services. From this point of view, CDMA has received much<br />

attention as a modulation technique that helps to sustain impairments<br />

encountered on the cable, namely narrowband interference, ingress, and<br />

impulse noise, while enabling to <strong>de</strong>liver data services without requiring<br />

much change to the infrastructure of a two-way pure coaxial n<strong>et</strong>work [35].<br />

Synchronous-CDMA (S-CDMA) systems [36] have <strong>de</strong>monstrated their ability<br />

to work robustly within the unused low frequency bands of the CATV<br />

medium (5-42 MHz in the United States). Implementation of CDMA communication<br />

systems on CATV n<strong>et</strong>works is the object of ongoing standardisation<br />

work within IEEE 802.14 group [37].<br />

Figure 2.9 illustrates the applications reviewed here above.<br />

2.2 Multiuser reception for DS-CDMA systems<br />

The implementation of DS-CDMA as a multiple access technique has revealed<br />

the inherent limitation of the single-user correlating receiver. Out<br />

of i<strong>de</strong>al conditions (orthogonal co<strong>de</strong> sequences, synchronous transmissions,<br />

perfect power-control), efficient reception can no longer rely only<br />

on the co<strong>de</strong> correlation properties to separate users. A MAI component<br />

plagues the receiving end, <strong>de</strong>grading performance, particularly when the<br />

power of the users is not balanced (Near-Far effect).<br />

The <strong>de</strong>sign of efficient receivers, in or<strong>de</strong>r to work in DS-CDMA systems,<br />

has to take this MAI component into account so as to exploit its information<br />

to improve reception. In a word, the receiver ought to <strong>de</strong>al with all<br />

active users. Clearly, this increases the complexity of the receiver, that is<br />

why multiuser reception is usually only consi<strong>de</strong>red in the uplink. In<strong>de</strong>ed,<br />

the receiving end, where signals from several users converge, is a n<strong>et</strong>work


2.2 Multiuser reception for DS-CDMA systems 25<br />

CATV<br />

cdmaOne<br />

UTRA<br />

Globalstar<br />

Skybridge<br />

IEEE 802.14 IEEE 802.11<br />

Figure 2.9: Some applications of CDMA nowadays<br />

point which should <strong>de</strong>al with all of them. Hence the obvious benefit of<br />

multiuser reception. However, this advantage has a cost: namely complexity.<br />

A tra<strong>de</strong>-off b<strong>et</strong>ween performance and complexity needs thus to<br />

be ma<strong>de</strong> in or<strong>de</strong>r to keep receivers affordable.<br />

The following sections will <strong>de</strong>scribe the advances of multiuser reception.<br />

MUD has been the subject of much interest, while the question of multiuser<br />

param<strong>et</strong>er estimation has less often been studied. After a short introduction<br />

to MUD, multiuser param<strong>et</strong>er estimation will be presented. Combined<br />

MUD and multiuser param<strong>et</strong>er estimation will close this section.<br />

2.2.1 D<strong>et</strong>ection<br />

The optimum <strong>de</strong>tector for multiuser DS-CDMA transmissions, a maximum-likelihood<br />

sequence <strong>de</strong>tector, has been <strong>de</strong>scribed in [38]. Consi<strong>de</strong>ring<br />

that the activity of the Nu users results in a signal which is similar to<br />

a single-user transmission over a dispersive channel, this <strong>de</strong>tector applies<br />

the Viterbi algorithm to the outputs of a bank of conventional single-user<br />

matched filters. However, the exponential complexity of the Viterbi algorithm<br />

in the number of users Nu makes it hardly practicable. Neverthe-


26 State of the art<br />

less, the way to MUD has been opened. In the sequel of [38], the linear <strong>de</strong>correlating<br />

<strong>de</strong>tector has been introduced in [14]. Many other sub-optimal<br />

structures have been proposed afterwards. Their performance, in terms of<br />

Near-Far resistance [39] and asymptotic efficiency [40], is similar to that<br />

achieved by the Viterbi algorithm but with a complexity only linear in Nu.<br />

A review of them can be found in [41, 42]. It is sk<strong>et</strong>ched on Figure 2.10.<br />

The issue of MUD in frequency-selective environments, shortly tackled<br />

in [41], is discussed in a more <strong>de</strong>tailed and more up-to-date way in [43].<br />

ZF<br />

MMSE<br />

PIC<br />

Conventional<br />

Matched filter<br />

Optimum<br />

Bank of matched filters<br />

+ Viterbi algorithm<br />

Suboptimum<br />

Linear<br />

DFE<br />

SIC<br />

Multipath fading ?<br />

No<br />

Yes<br />

Conventional<br />

RAKE receiver<br />

Optimum<br />

Bank of RAKE receivers<br />

+ Viterbi algorithm<br />

Figure 2.10: MUD systems <strong>de</strong>scribed in [41, 42]<br />

To alleviate the Near-Far effect, most of proposed MUD schemes require<br />

knowledge beyond that assumed by the conventional receiver, namely the<br />

channel impulse responses and the users’ signature waveforms. This information<br />

is often collected by using training sequences. To avoid this<br />

loss of throughput while maintaining performance, blind techniques relying<br />

on the same information as the conventional receiver but exhibiting<br />

optimum Near-Far resistance have recently been proposed [43, 44].


2.2 Multiuser reception for DS-CDMA systems 27<br />

2.2.2 Param<strong>et</strong>er estimation<br />

The ultimate objective of any receiver is to recover transmitted information.<br />

However, this requires most often to recover synchronisation prior to<br />

performing <strong>de</strong>tection. Synchronisation, to be un<strong>de</strong>rstood here in a broad<br />

sense, concerns recovery of all param<strong>et</strong>ers of the link: timing, frequency,<br />

phase, amplitu<strong>de</strong>, channel response, user’s waveform, <strong>et</strong>c. A thorough review<br />

of the synchronisation issues in spread-spectrum systems is presented<br />

in [8]. However, it does not really <strong>de</strong>al with multiuser aspects, assimilating<br />

MAI to a supplementary Gaussian noise contribution (Gaussian approximation,<br />

Section 2.3.3). The present work <strong>de</strong>velops the opposite view.<br />

It does regard MAI as an informative contribution which can be exploited<br />

in or<strong>de</strong>r to improve the performance of the param<strong>et</strong>er estimator. Among<br />

others, this is the main and most innovative contribution of this work.<br />

At this point, a remark should be ma<strong>de</strong> with respect to the interaction<br />

b<strong>et</strong>ween <strong>de</strong>tection and param<strong>et</strong>er estimation stages. To <strong>de</strong>rive the estimates<br />

of the param<strong>et</strong>ers, some structures rely on transmitted symbols<br />

while others do not. Moreover, the symbols used in the estimation process<br />

can be either the true symbols or the <strong>de</strong>cisions produced by the <strong>de</strong>tector.<br />

This leads to the distinction b<strong>et</strong>ween Data-Ai<strong>de</strong>d (DA), Decision-Directed<br />

(DD), and Non Data-Ai<strong>de</strong>d (NDA) estimation.<br />

At least at start-up, and maybe also periodically during transmission if<br />

the param<strong>et</strong>ers to estimate are time-varying, systems transmit training sequences<br />

aimed at helping receivers to collect information about the context<br />

of the transmission. These sequences, known by both transmitter and receiver,<br />

convey no information. Param<strong>et</strong>er estimators use them in or<strong>de</strong>r to<br />

perform estimation, minimising a cost function (I,θ) which <strong>de</strong>pends on<br />

the training sequences and on the param<strong>et</strong>ers (Figure 2.11). Since in this<br />

case symbols are known, the estimation is said to be DA.<br />

Obviously, DA estimation cannot be performed permanently since this<br />

would suppress any information throughput. When param<strong>et</strong>ers have been<br />

acquired thanks to the training sequences, estimators can switch from<br />

training sequences to <strong>de</strong>tector’s outputs. The cost function to minimise<br />

no longer <strong>de</strong>pends on the true symbols I but on the <strong>de</strong>cisions Î (Figure<br />

2.12). In such case one speaks of DD estimation. Provi<strong>de</strong>d that the <strong>de</strong>cisions<br />

are mostly correct, DD estimation performs as well as DA, with<br />

the advantage over DA that these are now informative bits and no longer


28 State of the art<br />

r (t)<br />

Phase estimator<br />

maxθ (I,θ)<br />

Training sequence<br />

Figure 2.11: DA estimator<br />

training ones which are transmitted over the channel.<br />

r (t)<br />

Phase estimator<br />

maxθ Î,θ<br />

<br />

Figure 2.12: DD estimator<br />

Of course, any <strong>de</strong>cision error <strong>de</strong>gra<strong>de</strong>s the DD estimation process, which<br />

in turn <strong>de</strong>gra<strong>de</strong>s the <strong>de</strong>cision process since the <strong>de</strong>tector often needs good<br />

param<strong>et</strong>er estimates in or<strong>de</strong>r to perform properly. This coupled influence<br />

can lead to a compl<strong>et</strong>e collapse of the receiver’s performance. A solution<br />

to such failure is to make the estimation process in<strong>de</strong>pen<strong>de</strong>nt of the <strong>de</strong>tector.<br />

NDA 4 structures are <strong>de</strong>signed in this perspective. They are also<br />

required in short burst transmissions when one cannot afford to waste<br />

throughput with training sequences [45]. The cost function they minimise<br />

has been ma<strong>de</strong> in<strong>de</strong>pen<strong>de</strong>nt of the symbols, for instance by averaging it<br />

over their pdf (Figure 2.13). Since NDA estimators do not exploit all the<br />

information available at the receiver, their performance is not as good as<br />

DA or DD structures. On the other hand, they still can work in situations<br />

where <strong>de</strong>cision errors multiply.<br />

Wh<strong>et</strong>her DA, DD or NDA, single-user estimation m<strong>et</strong>hods are <strong>de</strong>gra<strong>de</strong>d<br />

by MAI [46, 47]. Optimum estimators can be <strong>de</strong>rived, but similarly to<br />

the situation of MUD, their complexity plays against them. To perform<br />

multiuser param<strong>et</strong>er estimations two options are possible: splitting the<br />

4 NDA estimators are som<strong>et</strong>imes called ”blind” by analogy with blind <strong>de</strong>tectors [45].<br />

Î


2.2 Multiuser reception for DS-CDMA systems 29<br />

r (t)<br />

Phase estimator<br />

maxθ EI [ (I,θ)]<br />

Figure 2.13: NDA estimator<br />

multiuser estimation problem into single-user ones ([48], Approximate<br />

Maximum-Likelihood in [49]), or performing joint estimation over all active<br />

users. The latter option will be <strong>de</strong>scribed in the following paragraphs.<br />

Most of the contributions in the field of multiuser param<strong>et</strong>er estimation<br />

relate either to Expectation-Maximisation (EM), Singular Value Decomposition<br />

(SVD), or Exten<strong>de</strong>d Kalman Filtering (EKF). Each m<strong>et</strong>hod will be<br />

<strong>de</strong>alt with in a specific paragraph. A fourth paragraph will briefly encompass<br />

other contributions not based on one of the first three m<strong>et</strong>hods.<br />

Expectation-Maximisation EM is a two-step estimation algorithm which<br />

is able to produce the ML estimate of a vector of param<strong>et</strong>ers θ when observations<br />

y are the result of a many-to-one mapping of un<strong>de</strong>rlying variables<br />

x, also called ”compl<strong>et</strong>e data” [50, 51]. The compl<strong>et</strong>e data contains extra<br />

information that would ease the param<strong>et</strong>er estimation but, unfortunately,<br />

these compl<strong>et</strong>e data are usually unobservable. The true ML estimation<br />

would request to maximise the log-likelihood function ΛL (x θ) of the un<strong>de</strong>rlying<br />

variables with respect to the vector param<strong>et</strong>er. However, since<br />

these variables are not available, estimation is performed recursively using<br />

the EM algorithm. It is ma<strong>de</strong> of two steps: E-step (Expectation) and<br />

M-step (Maximisation), iterated successively until convergence is reached.<br />

In the E-step, the likelihood function to maximise ¯ ΛL is <strong>de</strong>fined as the expectation<br />

over the un<strong>de</strong>rlying variables x of their log-likelihood function<br />

ΛL assuming the observations y and the estimate of the vector param<strong>et</strong>er<br />

θk produced at previous iteration k:<br />

¯ΛL (x θ) =Ex [ΛL (x θ) y,θk] . (2.1)<br />

Following the E-step, the M-step updates the estimate by maximising the<br />

averaged log-likelihood function (2.1) over the vector param<strong>et</strong>er θ:<br />

ˆθk+1 =argmax<br />

θ<br />

¯ΛL (x θ) . (2.2)


30 State of the art<br />

The algorithm iterates b<strong>et</strong>ween E- and M-steps until convergence to the<br />

ML estimate, or at least a local extremum is reached, since the likelihood<br />

function does not <strong>de</strong>crease during iterations [51].<br />

The EM algorithm is well-suited to estimate param<strong>et</strong>ers from the received<br />

signal of a multiuser DS-CDMA system. In<strong>de</strong>ed, it is a composite signal<br />

built from contributions of Nu users transmitting over possibly multipath<br />

(Np-path) channels. Direct access to the compl<strong>et</strong>e data is not possible since<br />

the Nu¢Np signal contributions and noise are mixed tog<strong>et</strong>her into each reception<br />

filter output. At first sight, finding ML estimates of corresponding<br />

Nu ¢ Np s<strong>et</strong>s of synchronisation param<strong>et</strong>ers would require a joint maximisation<br />

over all param<strong>et</strong>ers, involving all Nu ¢ Np contributions. However,<br />

applying the EM algorithm enables to split this joint problem into<br />

Nu single ones and to <strong>de</strong>fine a likelihood function relying on one user at a<br />

time, assuming the other ones.<br />

Furthermore, an evolution of the EM algorithm, the Space-Alternating<br />

Generalised EM (SAGE) algorithm, exhibits faster convergence and lower<br />

complexity. The evolution is twofold. First, the SAGE algorithm involves<br />

additional iteration loops trying to produce refined estimates of a subs<strong>et</strong><br />

of the whole vector of param<strong>et</strong>ers to be estimated while keeping the other<br />

param<strong>et</strong>ers fixed. Second, the mapping from compl<strong>et</strong>e data to incompl<strong>et</strong>e<br />

data is no longer necessarily <strong>de</strong>terministic, but might be random. In [52],<br />

the joint estimation of <strong>de</strong>lay, azimuth, Doppler frequency, and complex<br />

amplitu<strong>de</strong> is performed in mobile radio environments using the SAGE algorithm.<br />

Singular Value Decomposition-based estimation m<strong>et</strong>hods SVD helps<br />

to build the pseudo-inverse matrix bringing out the minimum-norm solution<br />

of a linear least-squares problem [53, p. 414]. This might be used<br />

to <strong>de</strong>gra<strong>de</strong> a multiuser problem into single-user ones that are easier to<br />

solve [48], or to perform multiuser param<strong>et</strong>er estimation using the Multiple<br />

Signal Classification (MUSIC) algorithm [53, p. 452]. The MUSIC<br />

algorithm is known to be a m<strong>et</strong>hod of estimating frequencies of uncorrelated<br />

complex sinusoids in additive noise or to solve Direction Of Arrival<br />

(DOA) problems. Consi<strong>de</strong>ring the received signal samples of a sum of<br />

uncorrelated signals, the MUSIC algorithm performs an Eigenvalue De-


2.2 Multiuser reception for DS-CDMA systems 31<br />

composition (EVD) of the sample correlation matrix 5 , so as to distinguish<br />

two subspaces: the signal subspace and the noise subspace. Optimally,<br />

the param<strong>et</strong>ers of the signals are orthogonal to the noise subspace. A reliable<br />

estimate is thus obtained by searching for the param<strong>et</strong>er values which<br />

minimise the norm of their projection onto the noise subspace.<br />

In [49], a modified version of the MUSIC algorithm is introduced to estimate<br />

the propagation <strong>de</strong>lays, the phases, and the amplitu<strong>de</strong>s for all users<br />

of a DS-CDMA system in an Additive White Gaussian Noise (AWGN)<br />

channel. Besi<strong>de</strong>s performance results presented in [49], the performance<br />

of this estimator has also been <strong>de</strong>rived in [54] by applying an alternative<br />

perturbation analysis of the second-or<strong>de</strong>r statistics. A similarly modified<br />

algorithm is used in [55] for param<strong>et</strong>er estimation in static fading channels,<br />

while the estimation of propagation <strong>de</strong>lays in time-varying fading<br />

channels is performed in [56] using the same algorithm than in [49]. Back<br />

to static fading channels, another MUSIC-based algorithm is presented in<br />

[57] for channel estimation. In all these works, estimators are shown to be<br />

Near-Far resistant and not to rely on information from the <strong>de</strong>tector. They<br />

are thus suited for acquisition as well as for tracking.<br />

This separation property of SVD is also used in [58] in or<strong>de</strong>r to mo<strong>de</strong>l<br />

MAI as a coloured Gaussian noise and to <strong>de</strong>rive channel param<strong>et</strong>ers as<br />

ML estimates in coloured Gaussian noise. This requires a preamble (DA<br />

estimation).<br />

Exten<strong>de</strong>d Kalman Filtering Kalman filters have received much attention<br />

for their ability to perform adaptive least-squares estimation with a<br />

time-varying gain in the update equation. It ensures faster convergence<br />

[53]. However, Kalman filters are not directly applicable to the problem<br />

of param<strong>et</strong>er estimation since they only apply to linear systems. Unfortunately,<br />

the received signal is a non-linear function of the synchronisation<br />

param<strong>et</strong>ers. EKF, introduced as an extension of the standard Kalman filter<br />

to non-linear systems [59, section 13.7] [60, p. 386], is thus well suited for<br />

estimation in non-linear systems. Moreover, it performs b<strong>et</strong>ter than the<br />

Recursive Least Squares (RLS) algorithm because it incorporates a priori<br />

knowledge [61].<br />

5<br />

or, equivalently, a SVD of the received signal samples, since EVD is a particular case<br />

of SVD [53, p. 408, 456]


32 State of the art<br />

EKF is implemented in [62] for the DA estimation of timing and complex<br />

coefficients of a tapped-<strong>de</strong>lay line channel mo<strong>de</strong>l in a single-user wi<strong>de</strong>band<br />

communication system. The time-varying nature of the param<strong>et</strong>ers<br />

is mo<strong>de</strong>lled with first-or<strong>de</strong>r auto-regressive processes. Sequels of [62] add<br />

rejection of narrowband interference [63] and estimation of Doppler shift<br />

[64]. In [63] the narrowband interference is mo<strong>de</strong>lled as an N-th or<strong>de</strong>r<br />

auto-regressive process, estimated by a DD EKF estimator relying on the<br />

<strong>de</strong>cisions provi<strong>de</strong>d by a RAKE receiver. The lack of knowledge about the<br />

Doppler velocity is solved in [64] by implementing a bank of EKF estimators<br />

assuming this velocity. Their outputs are then combined according<br />

to their a posteriori probabilities so as to form a weighted estimate of the<br />

timing and channels param<strong>et</strong>ers.<br />

Miscellanea Besi<strong>de</strong> those implementing one of the three algorithms mentioned<br />

here above, some other contributions <strong>de</strong>al with multiuser param<strong>et</strong>er<br />

estimation in DS-CDMA systems.<br />

Four estimators of the complex amplitu<strong>de</strong> of the signal are introduced<br />

in [65], <strong>de</strong>pending on wh<strong>et</strong>her data are known (DA) or not (NDA) and<br />

wh<strong>et</strong>her timing has been recovered previously or not. These four estimators<br />

make use of the outputs of a bank of matched filters. Regarding [65]as<br />

based on second-or<strong>de</strong>r stationary statistics (the outputs of the matched filter<br />

bank), an estimator based on the second-or<strong>de</strong>r cyclostationary statistics<br />

of the received signal collected at the output of the sampler is presented<br />

in [66]. Using these statistics, complex amplitu<strong>de</strong> and timing are obtained<br />

as NDA least-squares estimates based on the Fourier transform of the cyclic<br />

auto-correlation function.<br />

While most contributions are concerned with steady-state performance,<br />

the joint acquisition of both time <strong>de</strong>lay and Doppler velocity is studied<br />

in [67] using a two-dwell correlator system. Acquisition is also an issue<br />

in [68]. The choice of midamble training co<strong>de</strong>s in burst transmission is<br />

discussed with respect to the performance of DA ML- and Matched Filter<br />

(MF)-channel estimators. Instead of wasting throughput in training<br />

sequences, one could rely on blind param<strong>et</strong>er estimation. Blind channel<br />

estimators are classified into two main categories [69]: statistics-based and<br />

subspace-based [70, 71].


2.2 Multiuser reception for DS-CDMA systems 33<br />

2.2.3 Joint <strong>de</strong>tection and param<strong>et</strong>er estimation<br />

In or<strong>de</strong>r to <strong>de</strong>sign a multiuser receiver, the combination of one of the MUD<br />

algorithms and one of the multiuser param<strong>et</strong>er estimation m<strong>et</strong>hods mentioned<br />

previously can be consi<strong>de</strong>red.<br />

The EM algorithm has often been applied to perform the param<strong>et</strong>er estimation<br />

step in the framework of a Gauss-Sei<strong>de</strong>l scheme. This scheme performs<br />

successively a <strong>de</strong>tection step, using previously produced param<strong>et</strong>er<br />

estimates, and an estimation step fed with the outputs of the <strong>de</strong>tector and<br />

applying the EM algorithm. Consi<strong>de</strong>ring timing known, a multistage algorithm<br />

is implemented for <strong>de</strong>tection in [72]. Complex amplitu<strong>de</strong>s are<br />

estimated through the EM algorithm. In [73, 74], timing is embed<strong>de</strong>d into<br />

the synchronisation param<strong>et</strong>ers to be estimated using the EM algorithm,<br />

while multistage <strong>de</strong>tection is performed.<br />

Subspace-based m<strong>et</strong>hods are used in [75] for the estimation of the propagation<br />

<strong>de</strong>lays, while Minimum Mean Square Error (MMSE) techniques are<br />

applied for the estimation of the path gains and for the <strong>de</strong>tection of the<br />

data bits. A refinement of [75] is presented in [76] where all the param<strong>et</strong>ers<br />

are estimated using SVD while <strong>de</strong>tection is performed according to the<br />

MMSE criterion.<br />

Finally, a tree-search for <strong>de</strong>tection and an adaptive recursive least-squares<br />

multiuser param<strong>et</strong>er estimator are associated in [77].<br />

However, the <strong>de</strong>sign of multiuser receivers can be modified so as to perform<br />

only one global estimation process, involving both data and param<strong>et</strong>ers.<br />

In<strong>de</strong>ed, the emergence of estimation structures <strong>de</strong>aling with different<br />

kind of param<strong>et</strong>ers have led to proposals which regard data symbols<br />

as another param<strong>et</strong>er so that the distinction b<strong>et</strong>ween <strong>de</strong>tection and param<strong>et</strong>er<br />

estimation disappears. The algorithms mentioned here above are<br />

then applied to solve this global estimation problem. In that perspective,<br />

the EKF estimation of data symbols (in<strong>de</strong>ed, hard <strong>de</strong>cisions taken over a<br />

param<strong>et</strong>er embedding both information and all synchronisation param<strong>et</strong>ers<br />

but timing) and timing is <strong>de</strong>scribed in [61]. Convergence and Near-Far<br />

resistance are evaluated and compared respectively with RLS and singleuser<br />

EKF. Both comparisons <strong>de</strong>monstrate how much more efficient the<br />

multiuser EKF is.


34 State of the art<br />

A last word in this review, about blind techniques. Most often, receivers<br />

are ma<strong>de</strong> of cooperating <strong>de</strong>tectors and param<strong>et</strong>ers estimators. Since <strong>de</strong>tectors<br />

require estimates while estimation can be ma<strong>de</strong> NDA, param<strong>et</strong>er<br />

estimation might be performed before <strong>de</strong>tection. Blind techniques r<strong>et</strong>urn<br />

this paradigm. Enabling the <strong>de</strong>tection stage to work autonomously, they<br />

lead to structures where data <strong>de</strong>tection can be initiated without preliminary<br />

param<strong>et</strong>er estimation. However, a param<strong>et</strong>er estimator might still be<br />

required at the output of the <strong>de</strong>tector. For instance a Phase Locked Loop<br />

(PLL) is used in [78] to mitigate the phase rotation of the Constant Modulus<br />

Algorithm (CMA).<br />

2.3 Phase estimation<br />

After the review of the literature about multiuser reception presented in<br />

the previous section, the present section shall <strong>de</strong>al with the estimation of<br />

the phase param<strong>et</strong>er. As explained in Section 1.2, this is the subject of the<br />

present study.<br />

2.3.1 Estimation structures<br />

Analog implementations<br />

Phase recovery structures have been <strong>de</strong>signed first for analog transmissions<br />

[7, section 4.5]. The major actor in this context is the PLL [79], used to<br />

track the carrier in Full-Carrier (FC) modulations or any other pilot signal.<br />

Analytical study shows that the PLL performs ML estimation of the phase<br />

param<strong>et</strong>er. The situation looks different when consi<strong>de</strong>ring Suppressed-<br />

Carrier (SC) modulations. At first sight, there is no frequency component<br />

to track. Still, the PLL can be used to recover the phase information of SC<br />

modulations. In or<strong>de</strong>r to do so, it tracks the output of a Mth-power nonlinearity<br />

which contains a pertinent frequency component thanks to the<br />

cyclostationarity of the received signal [80]. The squaring loop is an example<br />

of such a Mth-power loop (M = 2). Its study shows that it performs<br />

NDA ML phase estimation. Being well suited for binary modulations,<br />

this squaring loop is however helpless for higher-or<strong>de</strong>r balanced modulations,<br />

since applying them the squaring operation results in the vanishing<br />

of the information content [81, p. 279]. Correspondingly, these modulations<br />

require higher-or<strong>de</strong>r non-linearities or a slightly different treatment<br />

that avoids calling upon such high-or<strong>de</strong>r operations [82]. Other tracking


2.3 Phase estimation 35<br />

Analog Digital<br />

PLL tracking reference wave Waveform regenerator<br />

Costas loop Trackers<br />

Param<strong>et</strong>er extractor<br />

Param<strong>et</strong>er search<br />

Table 2.3: Analog and digital phase recovery implementations<br />

loops exist, like the Costas loop. Such loops do not explicitly track a frequency<br />

component. However, their study shows some equivalence with<br />

the squaring loop.<br />

Digital implementations<br />

Digital phase recovery structures have succee<strong>de</strong>d to analog implementations.<br />

As far as their analytical study is concerned, it is interesting to note<br />

that the ML estimation theory serves as an unifying theor<strong>et</strong>ical framework<br />

for their analysis [83, chapter 2]. Digital implementations are not just digitalised<br />

versions of previously <strong>de</strong>veloped analog structures. Among digital<br />

structures, one can distinguish waveform generators, param<strong>et</strong>er search,<br />

trackers, and param<strong>et</strong>er extractors. The major analog and digital phase<br />

estimator types are summarised in Table 2.3.<br />

Waveform regenerators are reminiscent of PLL tracking the phase of a FCmodulated<br />

signal. However, they appear ill-suited for digital implementation<br />

since they require several samples per symbol [83, p. 57]. Param<strong>et</strong>er<br />

search is a brute force m<strong>et</strong>hod. The value of the param<strong>et</strong>er which best fulfills<br />

a performance criterion is searched over the interval of possible values<br />

by testing all of them or a s<strong>et</strong> of uniformly distributed ones over this interval.<br />

The efficiency of the m<strong>et</strong>hod is obtained at the cost of exhaustive<br />

search. Neither waveform generator nor param<strong>et</strong>er search will be consi<strong>de</strong>red<br />

in the following sections.<br />

The current study will restrict its attention to trackers (FB phase estimators)<br />

and to param<strong>et</strong>er extractors (FF phase estimators). The performance<br />

criterion can be manipulated so as to produce an error signal which<br />

will drive a recovery loop. Here comes the tracker (Figure 2.14 a). On<br />

the other hand, the param<strong>et</strong>er extractor is a new form of estimator which<br />

is impossible to <strong>de</strong>sign in the analog world. It explicitly calculates the


36 State of the art<br />

closed-form estimate of the param<strong>et</strong>er according to ML theory. The value<br />

of the estimate is used further in the receiver to correct the phase of the<br />

received samples (Figure 2.14 b). FF structures are preferred to FB ones<br />

in burst transmissions. The relevant information is quickly collected at<br />

the receiver and FF estimators introduce no <strong>de</strong>lay due to acquisition. On<br />

the other hand, FB estimation is more suited for continuous transmissions<br />

since it is able to track variations of the param<strong>et</strong>ers instead of always starting<br />

the estimation process from scratches as FF structures do.<br />

Several digital phase estimators are listed in [83, chapter 2]. This classification<br />

has been systematically organised in [84, section I.2] on the basis<br />

of synergy of the estimation stage with the <strong>de</strong>tection process (DA/DD-<br />

/NDA), the un<strong>de</strong>rlying estimation theory (ML, non-linearities [82] among<br />

which squaring loop, <strong>et</strong>c.), the structure type (FB/FF), and the signal modulation.<br />

r (t)<br />

Phase<br />

estimator<br />

(a) FB<br />

r (t)<br />

Phase<br />

estimator<br />

(b) FF<br />

Figure 2.14: FB and FF implementations<br />

The introduction of param<strong>et</strong>er extractors is not the only difference b<strong>et</strong>ween<br />

analog and digital phase recovery structures. Another one is the<br />

fact that digital phase estimators can be implemented after timing recovery,<br />

which is not the case in analog structures [85, p. 233]. In digital receivers,<br />

phase estimation requires only one sample per symbol, while timing<br />

recovery needs oversampling. Then, the information initially provi<strong>de</strong>d<br />

to the timing estimator is <strong>de</strong>cimated before entering the phase estimation<br />

process [85, p. 275]. This introduces a <strong>de</strong>pen<strong>de</strong>ncy of the phase estimation<br />

with respect to the timing recovery which translates into a sensitivity to<br />

timing offs<strong>et</strong>, function of the pulse shape [83, pp. 251-255].<br />

However, phase estimators working in<strong>de</strong>pen<strong>de</strong>ntly of timing estimators<br />

can be <strong>de</strong>signed. This means that they rely on non-synchronised samples<br />

taken at the output of a prefilter. As a result, oversampling is requested to


2.3 Phase estimation 37<br />

avoid aliasing [84, section I.2.1.4].<br />

2.3.2 Performance characterisation of phase estimators<br />

FB estimation<br />

A lot of work has been done for the performance characterisation of recovery<br />

loops implementing FB estimators. The properties of these loops are<br />

most often <strong>de</strong>scribed using their open- and closed-loop transfer functions,<br />

in terms of closed-form expressions, Bo<strong>de</strong> plots, and root loci [79, chapter<br />

2]. These specifications are then used to perform two different kinds of<br />

analysis, linear and non-linear, <strong>de</strong>pending on the kind of performance to<br />

<strong>de</strong>rive, either steady-state or dynamic. The following paragraphs are a<br />

short introduction to linear and non-linear analysis.<br />

Linear analysis The linear analysis is performed un<strong>de</strong>r the assumption<br />

of small phase error to <strong>de</strong>rive steady-state performance. It is well known<br />

in the estimation literature [85] that the stable operating points of a recovery<br />

loop are located at the positive zero-crossings of its S-curve. The<br />

S-curve is the plot of the mean of the error signal driving the loop, with<br />

respect to the estimation error, consi<strong>de</strong>ring open-loop conditions. At the<br />

operating points this mean is equal to zero.<br />

The open-loop configuration is obtained by breaking the feedback path of<br />

the loop. Despite this lack of feedback, the influence of the recovery process<br />

is taken into account by substituting the estimation error of the loop<br />

for the param<strong>et</strong>er to be recovered.<br />

The sensitivity of the recovery loop with respect to the estimation error is<br />

measured on the S-curve as the slope at equilibrium. This value is used to<br />

build a simplified version of the loop. Using this linear mo<strong>de</strong>l, the firstor<strong>de</strong>r<br />

moments of the phase estimate (bias and variance) can be <strong>de</strong>rived,<br />

as well as the noise bandwidth [79, section 3.1] or the steady-state error in<br />

tracking, <strong>de</strong>pending on the phase stimulus and the shape of the loop filter<br />

[79, section 4.1].<br />

Non-linear analysis The assumption of small phase error is not always<br />

applicable especially when the loop starts working. A non-linear analysis<br />

is then performed. It is used to characterise dynamic performance of the


38 State of the art<br />

loop either during acquisition or during tracking.<br />

The main tool in that account is the Fokker-Planck m<strong>et</strong>hod for solving<br />

stochastic differential equations. A compl<strong>et</strong>e analysis of an analog firstor<strong>de</strong>r<br />

loop is performed in [86, chapter 4] with the help of the Fokker-<br />

Planck m<strong>et</strong>hod. It <strong>de</strong>rives the steady-state pdf of the phase error and characterises<br />

both acquisition and tracking performance with the time to lock<br />

and the cycle slip frequency. Although not applicable strictly speaking to<br />

discr<strong>et</strong>e time problems, the Fokker-Planck technique can been exten<strong>de</strong>d to<br />

the treatment of the stochastic difference equations mo<strong>de</strong>lling the working<br />

of digital loops [84, p. I-47].<br />

The analysis of the acquisition is aimed at examining the convergence of<br />

the estimate leading to the locking of the loop. This property is measured<br />

by two ranges, the lock-in range and the pull-in range. The lock-in range<br />

<strong>de</strong>fines the span of possible param<strong>et</strong>er values for which the FB structure<br />

will converge without missing any param<strong>et</strong>er cycle [79, p. 68]. On the<br />

other hand, the pull-in range, wi<strong>de</strong>r than the lock-in range, guarantees<br />

convergence but with possibly missing cycles. The error signal slowly<br />

drives the loop into its lock-in range where convergence is achieved [79,<br />

p. 72].<br />

A phenomenon that needs to be kept in view when studying acquisition<br />

is the hang-up problem. Due to the periodicity of the mean error signal<br />

with respect to the estimation error, illustrated on the S-curve [83, p. 221],<br />

the structure exhibits unstable tracking points. These points are the zerocrossings<br />

of the S-curve with negative slope (Figure 2.15). At these points,<br />

the loop wrongly appears to have reached equilibrium since the error signal<br />

is null. However, this is an unstable situation. Even a small change of<br />

the param<strong>et</strong>er value leads to a change of operating point. Measures will<br />

be taken during acquisition so as to avoid being trapped in such a position<br />

[79, p. 68]. Unless the acquisition time is prolonged due to the small value<br />

of the error signal until a change of operating point occurs [80].<br />

Moving to tracking performance, two kinds of inci<strong>de</strong>nt are to be consi<strong>de</strong>red,<br />

namely cycle slips and loss of lock.<br />

Cycle slips occur when the tracked phase param<strong>et</strong>er exhibits a variation<br />

so large that the loop moves its operating point to a neighbouring stable


2.3 Phase estimation 39<br />

Uu (∆)<br />

Linear<br />

working<br />

area<br />

Hang-up<br />

Cycle slip<br />

Figure 2.15: Hang-up and cycle slip<br />

∆<br />

Stable working point<br />

tracking point (Figure 2.15). This provokes <strong>de</strong>cision errors if the modulation<br />

exhibits rotational symm<strong>et</strong>ry. This failure roots in the narrowness<br />

of the loop bandwidth with respect to the spectral <strong>de</strong>nsity of the phase<br />

noise [81, p. 224], which prevents the loop from efficiently following variations<br />

of the param<strong>et</strong>er to track. However, this problem is not as easy<br />

to solve as it might appear at first sight. In<strong>de</strong>ed, enlarging the loop bandwidth<br />

helps to reduce the cycle slip problem but at the expense of a greater<br />

noise contribution to the system. A <strong>de</strong>gradation of steady-state performance<br />

thus occurs. A tra<strong>de</strong>-off has thus to be ma<strong>de</strong> [83, p. 63]. This slip<br />

phenomenon is characterised by the time average b<strong>et</strong>ween slips. These<br />

are rare inci<strong>de</strong>nts which are pr<strong>et</strong>ty difficult to study using computer simulations<br />

[85, section 6.4].<br />

Finally, situations might occur where the loop is driven out of lock. This<br />

loss of lock is characterised by the Mean-Time to Lock Loss (MTLL) [87,<br />

88].<br />

FF loops<br />

The study of FF implementations have not attracted as much attention as<br />

FB structures have. There are many reasons amounting for this lack of<br />

interest, the main one being the fact that FB and FF implementations produce<br />

the same estimate provi<strong>de</strong>d that their loop/averaging bandwidth<br />

coinci<strong>de</strong> [84, section I.3.2]. Steady-state performance <strong>de</strong>rived for FB estimators<br />

are thus directly exten<strong>de</strong>d to FF estimators.


40 State of the art<br />

As far as dynamic performance is concerned, FF implementations are not<br />

plagued by some of the effects encountered with FB loops like hangup<br />

[84, p. I-7]. This comes from the fact that there is no periodic behaviour<br />

like the one illustrated by the S-curve of a FB loop. However, this does not<br />

imply that FF loops are more efficient than FB structures in or<strong>de</strong>r to follow<br />

param<strong>et</strong>er dynamics [85, section 6.4.4].<br />

2.3.3 Multiuser Phase estimation<br />

So far, the review of phase estimation has been mainly concerned with<br />

single-user systems. What happens when the system un<strong>de</strong>r investigation<br />

exhibits MAI, like in spread-spectrum communication systems ? Consi<strong>de</strong>ring<br />

that Nu users are active, a rigorous analysis would request to lead<br />

an analytical study in the Nu-dimensional param<strong>et</strong>er space. There are very<br />

few works addressing this question directly and without approximation.<br />

It is usually preferred to simplify the problem in one way or another.<br />

Gaussian approximation<br />

The first simplification that comes to mind can be implemented at the<br />

mo<strong>de</strong>lling step. Bearing in mind the central limit theorem which states<br />

that the sum of a large number of mutually in<strong>de</strong>pen<strong>de</strong>nt random variables<br />

approaches a Gaussian distribution, it is tempting to mo<strong>de</strong>l the MAI<br />

as an additive Gaussian noise contribution which modifies the reception<br />

performance of the user un<strong>de</strong>r consi<strong>de</strong>ration, conditioned on the operating<br />

conditions. The averaging over these operating conditions, also called<br />

interference averaging [4, section 9-3.2], can then be performed either on<br />

the pdf of the MAI (Standard Gaussian Approximation) or on the performance<br />

<strong>de</strong>rived using this conditioned pdf (Improved Gaussian Approximation)<br />

[89]. For instance, a Single-User Maximum-Likelihood (SUML) synchroniser<br />

is presented in [90] in which the MAI has been mo<strong>de</strong>lled as a<br />

zero-mean Gaussian random variable. It is <strong>de</strong>monstrated that the loss of<br />

performance due to the <strong>de</strong>viation with respect to strict ML estimation is<br />

compensated by performance improvement (Near-Far resistance) with respect<br />

to standard synchronisation approaches. However, such Gaussian<br />

approximations are only valid as long as the central limit theorem applies,<br />

that is to say, in the case of large populations and without dominant term<br />

among the contributing variables. The limits of these Gaussian approximations<br />

are illustrated in [89] for a scarcely populated system and in the<br />

case of a dominant interferer.


2.3 Phase estimation 41<br />

Monte-Carlo simulations<br />

On the other hand, the study is not necessarily led in an analytical way.<br />

Rather than <strong>de</strong>riving their closed-form expressions, performance is measured<br />

through computer simulations of the communication system (Monte-<br />

Carlo simulations [91, section 5.6.1]). Computer tools enable to build a<br />

block diagram representation of the system by simulating its operation.<br />

Operating conditions are specified through system param<strong>et</strong>ers, while timevarying<br />

phenomena (information to transmit, channel behaviour...) are<br />

simulated by filtering the output of built-in pseudo-random generators.<br />

Running the simulation and measuring the outputs of <strong>de</strong>tection and estimation<br />

blocks give an insight into the performance of the whole system<br />

<strong>de</strong>pending on the specified operating conditions. The validity of the results<br />

produced by this m<strong>et</strong>hod is guaranteed within a certain confi<strong>de</strong>nce<br />

interval provi<strong>de</strong>d that some conditions are respected as regards the number<br />

of simulations [91]. Measuring a Bit Error Rate (BER) level requests,<br />

for instance, that a significant number of errors have been observed before<br />

its measure can be accepted within the corresponding confi<strong>de</strong>nce interval.<br />

Performance <strong>de</strong>gradation of <strong>de</strong>tection due to estimation errors<br />

Another possible simplification comes from the subject of the study itself.<br />

Instead of <strong>de</strong>riving the performance of the estimation structures, some authors<br />

rather study the influence of param<strong>et</strong>er estimation errors on the <strong>de</strong>tection<br />

process. There has been an overwhelming number of contributions<br />

in this field. The references mentioned in the following paragraphs address<br />

this question in spread-spectrum systems.<br />

One of the major issues in the receiver is to ensure synchronisation in the<br />

broad sense, and, more particularly, in co<strong>de</strong> tracking. The sensitivity of<br />

the linear <strong>de</strong>correlating <strong>de</strong>tector [14, 39] is investigated in both [40, 92]<br />

in presence of a vari<strong>et</strong>y of synchronisation errors (timing, phase and frequency)<br />

in AWGN channel. A Gaussian distribution of propagation <strong>de</strong>lay<br />

estimates is assumed in [40] while synchronisation errors are regar<strong>de</strong>d as<br />

uniformly distributed in [92]. The performance <strong>de</strong>gradation in terms of<br />

bit error probability, asymptotic efficiency, and Near-Far resistance is computed<br />

in [93] and compared to those of the conventional <strong>de</strong>tector.<br />

Moving to frequency-selective channels, the performance <strong>de</strong>gradation in<br />

terms of BER, pack<strong>et</strong> throughput, and <strong>de</strong>lay due to the bias of a non-


42 State of the art<br />

coherent early-late correlator with half-chip spacing is studied in [94]. This<br />

tracking loop is nearly optimal in AWGN channels but its behaviour is<br />

severely distorted by ISI in case of fast fading. However, as long as the<br />

chip duration is shorter than the <strong>de</strong>lay spread, ISI mitigation appears at<br />

the output of the receiver thanks to the co<strong>de</strong> correlation properties. As<br />

regards MAI, its influence can be mo<strong>de</strong>lled in a similar way to ISI by substituting<br />

Signal-to-Multipath Ratio (SMR) to Signal-to-Interference Ratio<br />

(SIR). Moreover, tracking errors due to MAI appear to be negligible with<br />

respect to ISI.<br />

Sticking to synchronisation in the broad sense, the inci<strong>de</strong>nce of channel<br />

estimation errors on the Joint D<strong>et</strong>ection (JD) receiver of a synchronous<br />

CDMA system is given in [68] in terms of Mean Square Error (MSE) by<br />

linearly adding an error estimation noise term to each estimated tap of<br />

the channel impulse responses. This estimation noise is uncorrelated with<br />

either the channel noise or with the estimation noise affecting other users.<br />

The Signal-to-Noise Ratio (SNR) <strong>de</strong>gradation due to noisy channel estimation<br />

is illustrated for different estimators as a function of the length of<br />

the midamble training co<strong>de</strong>. An appropriate choice of midamble co<strong>de</strong>s<br />

appears to limit the SNR <strong>de</strong>gradation even with suboptimal channel estimation.<br />

Besi<strong>de</strong>s co<strong>de</strong> tracking, tight power control is requested to afford using conventional<br />

correlating receivers in strong interfering environments (Near-<br />

Far effect). Power control is analysed in [95] without introducing a Gaussian<br />

approximation of the MAI. The <strong>de</strong>gradation of BER and capacity is<br />

measured with respect to the system load if power control is imperfect.<br />

Rigorous estimation performance study<br />

The present work does not aim at mo<strong>de</strong>lling the MAI as a Gaussian noise,<br />

nor at relying on Monte-Carlo simulations. Nor is it concerned with the<br />

influence of estimation errors on the <strong>de</strong>tector. The objective of this work<br />

is to <strong>de</strong>rive, as far as possible, performance expressions of ML phase estimators<br />

in DS-CDMA communication systems. Such approach has not<br />

attracted much interest. A contribution [96] addressing this question appeared<br />

only recently. The pdf of the phase estimate produced by a DD<br />

(in<strong>de</strong>ed DA, since perfect <strong>de</strong>cisions are assumed) first-or<strong>de</strong>r PLL is <strong>de</strong>rived<br />

in the presence of AWGN, phase noise, and multiuser interference<br />

in coherent asynchronous DS-CDMA, with the help of the Fokker-Planck


2.4 Conclusions 43<br />

m<strong>et</strong>hod.<br />

2.4 Conclusions<br />

This chapter had several purposes. First, it has <strong>de</strong>scribed and scanned the<br />

current and future applications of the multiple access scheme consi<strong>de</strong>red<br />

in this thesis, viz. DS-CDMA. Limiting the scope of the present thesis to<br />

mobile communication systems, the stress has then been laid on the <strong>de</strong>sign<br />

of multiuser receivers. The work performed so far for symbol <strong>de</strong>tection as<br />

well as for param<strong>et</strong>er estimation has been reviewed. Finally, the last section<br />

of this chapter has focused on the estimation of the phase which is the<br />

param<strong>et</strong>er at the centre of the present work.<br />

The next chapter will <strong>de</strong>tail the communication system un<strong>de</strong>r investigation.<br />

It will also introduce the analytical foundations required for the<br />

performance study to be lead in Chapters 4 and 5.


Chapter 3<br />

Tools<br />

3.1 System <strong>de</strong>scription<br />

3.1.1 System un<strong>de</strong>r investigation<br />

Consi<strong>de</strong>r the uplink of a coherent CDMA communication system accommodating<br />

Nu users (Figure 3.1). The low-pass equivalent signal tk(t) transmitted<br />

by user k writes:<br />

tk(t) = 2Ek<br />

+<br />

m=<br />

I m k dk(t mT ). (3.1)<br />

Im k = am k + jbm k are the modulated data symbols. Angular modulation<br />

M-PSK will be consi<strong>de</strong>red in the following sections, with variance σ2 Ik .<br />

Ekσ2 Ik is thus the emitted energy per symbol Es,k = Eb,k log2 M of user k,<br />

as <strong>de</strong>tailed in Section 3.1.2. T stands for the symbol duration and dk(t) is<br />

the spreading waveform for user k<br />

dk(t) =<br />

Nc 1<br />

n=0<br />

v n k u (t nTc) (3.2)<br />

where vn k is the pseudo-random spreading co<strong>de</strong>, Tc is the chip duration,<br />

Nc = T is the processing gain and u(t) is a rectangular pulse of duration<br />

Tc<br />

Tc. This means that signal is not band-limited, which is not a reasonable<br />

assumption for real systems constrained to work in pre-assigned bands.


46 Tools<br />

I m 1<br />

I m 2<br />

I m Nu<br />

Spreading Power<br />

Matched<br />

Symbol<br />

control<br />

filtering sampling<br />

d1 (t)<br />

d2 (t)<br />

dNu (t)<br />

Ô 2 E1<br />

Ô 2 E2<br />

2 ENu<br />

e jφ1<br />

e jφ2<br />

e jφNu<br />

c2 (t)<br />

c1 (t)<br />

cNu (t)<br />

h⋆ 1 ( t)<br />

h ⋆ 2 ( t)<br />

h⋆ ( t) Nu<br />

t = mT<br />

t = mT<br />

t = mT<br />

Figure 3.1: Uplink of a coherent CDMA communication system<br />

y m 1<br />

y m 2<br />

y m Nu<br />

Coherent<br />

<strong>de</strong>modulation<br />

e j ˆ φ m 1<br />

e j ˆ φ m 2<br />

e j ˆ φ m Nu<br />

e j ˆ φ m 1 y m 1<br />

e j ˆ φ m 2 y m 2<br />

e j ˆ φ m Nu y m Nu


3.1 System <strong>de</strong>scription 47<br />

The spreading waveform dk(t) is normalised so that<br />

T<br />

0<br />

dk(t) 2 dt =1. (3.3)<br />

Signals tk (t) are transmitted through channels having impulse responses<br />

ck (t). Defining<br />

hk(t) =dk(t) ª ck(t), (3.4)<br />

the low-pass equivalent received signal at the BS, sum of the contributions<br />

of the Nu active users, may be written as<br />

Nu <br />

r(t) = 2Ek e jφk<br />

k=1<br />

+<br />

m=<br />

I m k hk(t mT )+n(t). (3.5)<br />

At the receiving end, the bandpass signal is downconverted to baseband<br />

using a local oscillator with correct frequency but arbitrary phase. Thus,<br />

φk, the param<strong>et</strong>er of interest in this thesis, appears in the expression (3.5)<br />

of the low-pass equivalent received signal r (t) as the carrier phase difference<br />

b<strong>et</strong>ween transmitter’s and receiver’s oscillators. Finally, n(t) is the<br />

low-pass equivalent of an AWGN with two-si<strong>de</strong>d power spectral <strong>de</strong>nsity<br />

N0<br />

2 .<br />

The low-pass equivalent received signal r (t) is applied to a bank of filters<br />

matched to the compl<strong>et</strong>e impulse responses hk (t). In Chapter 5, when<br />

<strong>de</strong>aling with DD estimation structures, hard <strong>de</strong>cisions will be taken from<br />

the phase-corrected outputs of these matched-filters. This is <strong>de</strong>finitely not<br />

the optimal <strong>de</strong>tector. Such <strong>de</strong>tector is only suited for synchronous transmissions<br />

of DS-CDMA signals using perfectly orthogonal co<strong>de</strong>s in AWGN<br />

environments. Out of these i<strong>de</strong>al conditions, its <strong>de</strong>cisions are plagued by<br />

ISI and MAI. This <strong>de</strong>ficiency might be corrected using a more appropriate<br />

<strong>de</strong>tector (See Section 2.2.1). However, it will not be the case here, since<br />

the aim of this thesis is to <strong>de</strong>monstrate that MAI contributions are informative<br />

and that they may be exploited to improve the performance of the<br />

receiver. Nevertheless, the front-end shown in Figure 3.1 fits all <strong>de</strong>tectors,<br />

wh<strong>et</strong>her optimal or not, as it provi<strong>de</strong>s sufficient statistics for reception.


48 Tools<br />

The outputs of these channel-matched filters are sampled at symbol rate.<br />

stands for the normalised matched filter output<br />

y p<br />

k<br />

y p<br />

k =<br />

1<br />

Ô 2EkT<br />

= e jφk<br />

+<br />

q=<br />

+ Nu <br />

l=1<br />

l=k<br />

+ν p<br />

k<br />

+<br />

e jφl<br />

h ⋆ k<br />

(t pT ) r (t) dt (3.6)<br />

I q q<br />

kxp k,k<br />

El<br />

Ek<br />

+<br />

q=<br />

I q q<br />

l<br />

xp<br />

k,l<br />

Useful term<br />

+ ISI<br />

MAI<br />

Additive noise<br />

(3.7)<br />

p q<br />

where xk,l represents the normalised channel correlation coefficient b<strong>et</strong>ween<br />

users k and l for a time shift (p q) T<br />

p q 1<br />

xk,l =<br />

T<br />

+<br />

h ⋆ k (t pT ) hl (t qT) dt (3.8)<br />

and ν p<br />

k are zero-mean complex samples of the noise filtered by the matched<br />

filter<br />

ν p<br />

k =<br />

1<br />

Ô 2EkT<br />

+<br />

h ⋆ k<br />

(t pT ) n (t) dt. (3.9)<br />

Noise samples produced by different matched filters at distant time instants<br />

might be correlated up to the value of the normalised channel cor-<br />

p q p q<br />

q<br />

relation coefficient xk,l . ρk,l and ρp<br />

k,l measure the correlation b<strong>et</strong>ween<br />

Rice components of the filtered noise:<br />

p q<br />

ρk,l = E ν p<br />

q<br />

p<br />

q<br />

k νl = E νk νl = 1<br />

p q<br />

N0 (xk,l )<br />

Ô<br />

2 EkElT<br />

p q<br />

ρk,l = E ν p<br />

q<br />

p<br />

q<br />

p q<br />

N0 1 (xk,l )<br />

k νl = E νk νl = Ô<br />

2 EkElT .<br />

(3.10)<br />

The first term of (3.7) inclu<strong>de</strong>s the useful symbol I p<br />

k as well as the interfering<br />

ones through ISI. The following term represents the MAI contribution<br />

found at the output of a filter matched to the channel response of user k.<br />

It results from the activity of interfering users within the same frequency


3.1 System <strong>de</strong>scription 49<br />

band at the same time. In an i<strong>de</strong>al case, this interference would be cancelled<br />

thanks to the correlation properties of the spreading co<strong>de</strong>s v n k .<br />

Unfortunately, a compl<strong>et</strong>e cancellation cannot be achieved in most cases<br />

because this would require orthogonal co<strong>de</strong>s and synchronous transmissions<br />

over non dispersive channels. Since these conditions are not fulfilled<br />

most of the time, co<strong>de</strong>s are chosen so as to produce as few MAI as possible<br />

[10].<br />

In the following sections, the co<strong>de</strong> sequences v n k and the channel responses<br />

ck(t) will be supposed to be perfectly known at the BS. Timing<br />

will be regar<strong>de</strong>d as perfectly recovered1 . Moreover, the channel responses<br />

will be assumed to be static2 with power <strong>de</strong>lay profiles <strong>de</strong>fined in accordance<br />

with COST 207 mo<strong>de</strong>ls [97]. Finally, the phases φk will be assumed<br />

to be constant.<br />

3.1.2 Definition of Energy-to-Noise ratios<br />

The <strong>de</strong>nominator of the ratio Eb,k<br />

N0<br />

being known from the previous section,<br />

Eb,k, the energy conveyed by each bit transmitted by user k, has now to<br />

be calculated. To do so, the variance of baseband symbols sent by user k<br />

is to be <strong>de</strong>rived. Multiplying this variance by the symbol length T will<br />

<strong>de</strong>fine the baseband symbol energy which is two times the bandpass symbol<br />

energy Es,k. Finally, consi<strong>de</strong>ring angular modulation with M states,<br />

Eb,k will come out of the division of the bandpass symbol energy Es,k by<br />

log 2 M.<br />

Eb,k = Es,k<br />

log 2 M =<br />

<br />

1 1<br />

baseband symbol variance T . (3.11)<br />

log2 M 2<br />

The first step is thus to calculate the variance of baseband symbols. Since<br />

the transmitted signal tk (t) is cyclostationary, the mathematical expectation<br />

operation is combined with stationarisation using a random <strong>de</strong>lay<br />

T0 uniformly distributed over [0,T]. Consi<strong>de</strong>ring in<strong>de</strong>pen<strong>de</strong>nt i<strong>de</strong>ntically<br />

1<br />

Sampling at the output of the matched filter occurs at the peak of the auto-correlation<br />

function of the shaping pulse.<br />

2<br />

A channel impulse responses can be regar<strong>de</strong>d to be static when its coherence time is<br />

greater than the symbol rate [7, p. 709]. In the frequency domain, it means that the Doppler<br />

spread of the channel is smaller than the loop bandwidth of the estimator.


50 Tools<br />

distributed data symbols I m k<br />

EI,T0 [tk (t T0) t ⋆ k (t T0)]<br />

= 2Ek<br />

T<br />

⎡<br />

⎣<br />

+ +<br />

I m k (In k )⋆<br />

EI<br />

= 2Ek<br />

T<br />

+<br />

m=<br />

m=<br />

= 2Ekσ 2 Ik<br />

T<br />

= 2Ekσ 2 Ik<br />

T<br />

n=<br />

+<br />

n=<br />

+<br />

m=<br />

+<br />

EI [I m k (In k )⋆ ]<br />

<br />

δ (m n)<br />

and using <strong>de</strong>finition (3.8), the variance writes<br />

0<br />

T<br />

0<br />

<br />

0<br />

T<br />

T<br />

hk (t mT T0) h ⋆ k (t nT T0) dT0⎦<br />

hk (t mT T0) h ⋆ k (t nT T0) dT0<br />

hk (t mT T0) h ⋆ k (t nT T0) dT0<br />

⎤<br />

(3.12)<br />

(3.13)<br />

hk (t τ) 2 dτ (3.14)<br />

= 2Ekσ 2 Ik x0 k,k (3.15)<br />

where δ (m) is a Kronecker <strong>de</strong>lta function.<br />

Introducing (3.15) in(3.11) gives Es,k and Eb,k<br />

Es,k = EkTx 0 k,k σ2 Ik (3.16)<br />

Eb,k = EkTx 0 k,k<br />

σ2 Ik . (3.17)<br />

log2 M<br />

Making use of Eb,k<br />

, the variance of the Rice components of the noise samples<br />

N0<br />

(3.9) writes<br />

σ 2 (νk) = σ2 (νk) = σ2 νk<br />

2<br />

for M-PSK modulation.<br />

= 1<br />

2<br />

N0 x 0 k,k<br />

EkT<br />

<br />

= 1<br />

⎡<br />

⎢<br />

σ<br />

⎣<br />

2<br />

2 Ik<br />

<br />

x0 2 k,k<br />

log 2 M<br />

Eb,k<br />

N0<br />

⎤<br />

1<br />

⎥<br />

⎦<br />

(3.18)


3.2 Maximum-Likelihood estimation 51<br />

3.2 Maximum-Likelihood estimation<br />

3.2.1 Maximum A Posteriori and Maximum-Likelihood<br />

Param<strong>et</strong>er estimation theory [59, 98] classifies estimation m<strong>et</strong>hods in two<br />

main domains, distinguishing the Bayes approach and the classic approach<br />

(also called Fisher approach).<br />

In the Bayes approach, the vector param<strong>et</strong>er is regar<strong>de</strong>d to be a random<br />

variable whose behaviour is <strong>de</strong>scribed by the pdf Tθ (θ) mo<strong>de</strong>lling its distribution<br />

over the span of possible values. The information provi<strong>de</strong>d by<br />

this pdf is exploited by the estimation process. Maximum A Posteriori<br />

(MAP) is the most famous algorithm operating un<strong>de</strong>r the Bayes umbrella.<br />

This algorithm searches for the vector param<strong>et</strong>er θ which maximises the a<br />

posteriori pdf.<br />

ˆθMAP =argmax<br />

θ T θr (θ r) = arg max<br />

θ<br />

T rθ (r θ) Tθ (θ)<br />

Tr (r)<br />

<br />

. (3.19)<br />

If the observations r and the vector param<strong>et</strong>er θ are jointly Gaussian, MAP<br />

and MMSE m<strong>et</strong>hods produce the same estimate [59, p. 485].<br />

However, the characterisation of the behaviour of the param<strong>et</strong>er through<br />

its pdf is not always available. Estimation m<strong>et</strong>hods of the classic approach<br />

have been <strong>de</strong>signed to solve this issue in as much as they do not call upon<br />

the a priori pdf of the param<strong>et</strong>er. This is ma<strong>de</strong> possible by assuming that<br />

the param<strong>et</strong>er is uniformly distributed. In<strong>de</strong>ed, the a priori pdf provi<strong>de</strong>s<br />

then no extra information and MAP becomes ML 3 . The ML estimation<br />

process tries to maximise the likelihood function T rθ (r θ), which is the<br />

probability of the observation r assuming the vector param<strong>et</strong>er θ.<br />

ˆθML =argmax<br />

θ T rθ (r θ) . (3.20)<br />

This estimation m<strong>et</strong>hod presents the interesting feature that, being conditioned<br />

on the param<strong>et</strong>ers to estimate, the likelihood function to maximise<br />

only <strong>de</strong>pends on the noise distribution. In the case of Gaussian noise ML<br />

and Least-Squares (LS) m<strong>et</strong>hods produce the same estimate [59, p. 483].<br />

The present work has been done in keeping with the classic approach.<br />

3 The rea<strong>de</strong>r ought to notice that some authors introduce classic estimation as an approach<br />

based on the assumption that the param<strong>et</strong>er to be estimated is a <strong>de</strong>terministic constant<br />

[7, 59] rather than as a particular case of the Bayes approach for uniform pdf.


52 Tools<br />

3.2.2 Likelihood function<br />

In the classic approach optimum results are asymptotically obtained by<br />

applying ML estimation. The estimate is obtained from the maximisation<br />

of the likelihood function Λ(r θ). This function has thus to be <strong>de</strong>rived. To<br />

do so, a s<strong>et</strong> of observable samples of the received signal (3.5) is required.<br />

Such a s<strong>et</strong> should be sufficient statistics, which means that all the information<br />

of the time-continuous received signal should be contained in this<br />

s<strong>et</strong> of samples. Several approaches can be consi<strong>de</strong>red to build such a s<strong>et</strong>.<br />

In the next sections, the general m<strong>et</strong>hod of Karhunen-Loève series expansion<br />

will first be presented. Next, it will be shown that the prefiltering and<br />

oversampling of the received signal implemented in nowadays digital receivers<br />

[85, p. 227] can be regar<strong>de</strong>d as a particular case of it. Finally, the<br />

averaging of the likelihood-function over the pdf of the data symbols for<br />

NDA estimation will be consi<strong>de</strong>red.<br />

Karhunen-Loève series expansion L<strong>et</strong> fi (t) be a compl<strong>et</strong>e orthonormal<br />

s<strong>et</strong> of NKL functions over the observation interval T0 [98, section 3.2]:<br />

<br />

(t) dt = δ (k l) . (3.21)<br />

T0<br />

fk (t) f ⋆ l<br />

Defining ri as the s<strong>et</strong> of the projections of the received signal r (t) (3.5)<br />

on the s<strong>et</strong> of the orthonormal functions fi (t)<br />

<br />

ri = r (t) f ⋆ i (t) dt (3.22)<br />

=<br />

T0<br />

Nu <br />

k=1<br />

2Ek e jφk<br />

m=<br />

+<br />

I m k<br />

<br />

T0<br />

hk(t mT )f ⋆ <br />

i (t) dt +<br />

T0<br />

n(t)f ⋆ i (t) dt<br />

(3.23)<br />

the Karhunen-Loève series expansion states that the s<strong>et</strong> of mutually uncorrelated<br />

coefficients ri can represent r (t) in the limit as NKL + [7,<br />

appendix 4A].<br />

r (t) = lim<br />

NKL+<br />

NKL <br />

i=1<br />

rifi (t) . (3.24)<br />

The ri do not only represent r (t), but also help to build the likelihood<br />

function. In the case un<strong>de</strong>r study, they form a NKL-dimensional vector.


3.2 Maximum-Likelihood estimation 53<br />

Provi<strong>de</strong>d the compl<strong>et</strong>e s<strong>et</strong> of orthonormal functions fi (t) is ma<strong>de</strong> of the<br />

eigenfunctions of the auto-correlation function of the received signal r (t),<br />

the ri components conditioned on the vector param<strong>et</strong>er Φ are complexvalued<br />

statistically in<strong>de</strong>pen<strong>de</strong>nt Gaussian random variables with mean<br />

⎡<br />

Nu <br />

E (ri Φ) = E ⎣ 2Ek e jφk<br />

and variance<br />

k=1<br />

σ 2 riΦ<br />

+<br />

m=<br />

⎡<br />

<br />

<br />

= E ⎣<br />

<br />

<br />

T0<br />

I m k<br />

<br />

T0<br />

hk(t mT )f ⋆ i<br />

<br />

<br />

<br />

(t) dt<br />

<br />

Φ<br />

⎤<br />

⎦ (3.25)<br />

n(t)f ⋆ <br />

2<br />

<br />

i (t) dt<br />

<br />

<br />

⎤<br />

<br />

<br />

Φ⎦<br />

. (3.26)<br />

Assuming the data symbols I m k and channel impulse responses hk (t)<br />

to be known, the NKL-dimensional joint pdf assuming Φ relies then only<br />

on the noise distribution. Since n (t) is statistically in<strong>de</strong>pen<strong>de</strong>nt Gaussian<br />

noise, its components in the Karhunen-Loève series expansion are statistically<br />

in<strong>de</strong>pen<strong>de</strong>nt and jointly Gaussian, so that the NKL-dimensional<br />

joint pdf writes<br />

T (rNKL Φ)<br />

rNKLΦ = NKL <br />

i=1<br />

1<br />

Ô 2πσriΦ<br />

exp 1<br />

2σ 2 riΦ<br />

<br />

<br />

<br />

<br />

<br />

<br />

<br />

<br />

ri Nu <br />

k=1<br />

Ô 2Ek e jφk<br />

+<br />

m=<br />

I m k<br />

<br />

T0<br />

hk(t mT )f ⋆ i<br />

(t) dt<br />

<br />

<br />

<br />

<br />

<br />

<br />

<br />

<br />

2<br />

.<br />

(3.27)<br />

By analogy to (3.24), the likelihood function Λ(r Φ) comes from (3.27) in<br />

the limit as NKL + .<br />

Λ(r Φ) = lim<br />

NKL+ T rN KLΦ<br />

⎡<br />

= Cst exp ⎣<br />

(rNKL Φ) (3.28)<br />

<br />

1<br />

r (t) s (t Φ)<br />

2N0<br />

2 ⎤<br />

dt⎦<br />

. (3.29)<br />

T0


54 Tools<br />

The ML estimate is the value of the param<strong>et</strong>er that maximises the likelihood<br />

function. From (3.29), it can be interpr<strong>et</strong>ed as the value of the param<strong>et</strong>er<br />

which minimises the distance b<strong>et</strong>ween the received signal r (t) and<br />

the noiseless signal s (t Φ) assuming Φ.<br />

Since it is assumed that an infinite sequence of data symbols is transmitted<br />

(3.5), but that the observation interval is finite and of length T0, it<br />

is more convenient to change integration limits in (3.29) from[0,T0] to<br />

[ , + ] and simultaneously, to change the summation limits in (3.5)<br />

from [ , + ] to [1,N] so that NT = T0 [83, p. 93]. Practical estimators<br />

built on this modified likelihood function are called pseudo-ML estimators<br />

in [99]. This approximation causes end effects [83, p. 93] traced back<br />

in [81, p. 82]. They generate a noise-in<strong>de</strong>pen<strong>de</strong>nt jitter component whose<br />

inci<strong>de</strong>nce <strong>de</strong>pends mainly on the width of the observation interval [81,<br />

p. 82] and on the type of modulation [99, p. 1126]. These end effects will<br />

be neglected in the present thesis.<br />

Using this approximation, relation (3.29) then becomes<br />

Λ(r Φ)<br />

⎡<br />

exp ⎣<br />

exp<br />

<br />

⎡<br />

⎢<br />

exp ⎣<br />

<br />

1<br />

2N0<br />

Nu <br />

T0<br />

EkT<br />

l=1<br />

l=k<br />

⎤<br />

r (t) 2 dt⎦<br />

exp<br />

N<br />

+<br />

<br />

<br />

Nu<br />

<br />

k=1<br />

(I m k )⋆I n n<br />

k xm<br />

k,k<br />

k=1<br />

N0<br />

m=1 n=<br />

Nu Nu <br />

Ô<br />

EkEl T<br />

N0<br />

k=1<br />

e j(φk<br />

N<br />

φl)<br />

m=1 n=<br />

2EkT<br />

<br />

N0<br />

+<br />

e jφk<br />

(I m l )⋆ I n k<br />

N<br />

(I m k )⋆y m <br />

k<br />

m=1<br />

n<br />

xm<br />

l,k<br />

⎤<br />

⎥<br />

⎦ . (3.30)<br />

Moving from (3.29) to(3.30) has led to split the distance r (t) s (t Φ) 2<br />

into its components, the energy of the received signal r (t) 2 , the double<br />

product 2 [r (t) s⋆ (t Φ)], and the energy of the noiseless signal assuming<br />

Φ, s (t Φ) 2 . The energy of the received signal stands explicitely in (3.30),<br />

while the double product turns into the second term, involving matched<br />

filter outputs ym k . Finally, the last two terms come from the expansion of<br />

s (t Φ) 2 .<br />

Here lies the major innovation of ML phase estimation in a multiuser context.<br />

So far, in most carrier recovery structures, the distance in (3.29) is


3.2 Maximum-Likelihood estimation 55<br />

expan<strong>de</strong>d so as to keep only the double product, the correlation of the<br />

received signal, and the conditioned one [r (t) s (t Φ)]. Square terms<br />

are disregar<strong>de</strong>d as not <strong>de</strong>pending on the param<strong>et</strong>er to estimate. This is<br />

no longer the case in multiuser spread-spectrum systems. The energy of<br />

the conditioned signal s (t Φ) 2 <strong>de</strong>pends on the param<strong>et</strong>ers to estimate<br />

through differences φk φl. It can thus not be disregar<strong>de</strong>d in the estimation<br />

process.<br />

Searching for ML estimate is not easy due to the exponential function in<br />

(3.30). However, since the logarithm is a monotonic function of its argument,<br />

the value which maximises f (Φ) also maximises log [f (Φ)]. So,<br />

usually, instead of <strong>de</strong>aling with the exponential function appearing in the<br />

likelihood function, one prefers to use its logarithm. Taking the logarithm<br />

and <strong>de</strong>veloping s (t Φ) by using (3.7) and (3.8) finally gives<br />

ΛL(r Φ)<br />

<br />

Nu 2EkT<br />

= Cst + e jφk<br />

Nu <br />

k=1<br />

EkT<br />

N0<br />

k=1<br />

Nu Nu <br />

k=1<br />

l=1<br />

l=k<br />

N0<br />

N<br />

+<br />

m=1 n=<br />

Ô EkElT<br />

N0<br />

N<br />

(I m k )⋆y m <br />

k<br />

m=1<br />

(I m k )⋆I n n<br />

k xm<br />

k,k<br />

e j(φk φl)<br />

N<br />

+<br />

m=1 n=<br />

(I m l )⋆I n n<br />

k xm<br />

l,k . (3.31)<br />

Sampling in<strong>de</strong>pen<strong>de</strong>ntly of the transmitter clock The log-likelihood<br />

function <strong>de</strong>rived by the Karhunen-Loève series expansion <strong>de</strong>pends on the<br />

matched filter outputs ym 1<br />

k produced at rate T . These are sufficient statistics<br />

for phase estimation (not for timing [85, p. 257]). Implicitly, it was assumed<br />

that the sampling of the matched filter output occurred at the right<br />

instants, in synchronisation with the transmitter clock. Instead of working<br />

at 1<br />

T ,[85, chapter 4] establishes conditions un<strong>de</strong>r which collecting samples<br />

at a higher rate compl<strong>et</strong>ely in<strong>de</strong>pen<strong>de</strong>nt of transmitter clock 1 1 > Ts T can<br />

produce sufficient statistics. This is specially useful for digital implementations<br />

since this reduces the number of information flows b<strong>et</strong>ween analog<br />

stages and digital stages. Samples can be produced by a free-running clock<br />

at rate 1 . By interpolation and <strong>de</strong>cimation of the samples another s<strong>et</strong> of<br />

Ts<br />

and at the right sampling instants.<br />

samples is produced at rate 1<br />

T


56 Tools<br />

On the one hand, this sampling provi<strong>de</strong>s samples r (mTs) which are sufficient<br />

statistics to represent r (t) provi<strong>de</strong>d a generalised anti-aliasing filter<br />

fulfilling conditions [85, p. 243] has been used. This is the sampling the-<br />

orem which can be un<strong>de</strong>rstood as a special case of the Karhunen-Loève<br />

series expansion using Whitakker basis functions<br />

gonal functions fi (t).<br />

sin x<br />

x as a s<strong>et</strong> of ortho-<br />

On the other hand, noise samples are complex-valued zero-mean Gaussian<br />

random variables. These samples might be separated into their Rice<br />

components nm I and nm N0<br />

Q , each of them exhibiting a variance of . The Ts<br />

joint pdf of N complex-valued noise samples writes<br />

TnI,nQ (nI, nQ)<br />

<br />

= Cst exp<br />

= Cst exp<br />

<br />

Ts<br />

N<br />

2N0 m=1<br />

<br />

N<br />

Ts<br />

2N0<br />

m=1<br />

n m I 2 +<br />

N <br />

m<br />

n <br />

Q<br />

2<br />

<br />

m=1<br />

r m I sm I 2 +<br />

m=1<br />

(3.32)<br />

N <br />

m<br />

rQ s m <br />

<br />

Q<br />

2<br />

<br />

. (3.33)<br />

In the limit as N + , TnI,nQ (nI, nQ) becomes Λ(r Φ) (3.30).<br />

Low SNR approximation for NDA estimation As explained in Section<br />

2.2.2, it might be <strong>de</strong>sirable un<strong>de</strong>r certain circumstances to estimate param<strong>et</strong>ers<br />

without relying on either training sequences (DA case) or on <strong>de</strong>cisions<br />

(DD case). From the point of view of ML, the maximisation of a<br />

modified likelihood function in which the influence of the data symbols<br />

has been cleared is a solution to this problem. This is the NDA approach<br />

which substitutes a log-averaged likelihood function to the usual likelihood<br />

function (3.30) to be maximised. The average is performed with<br />

respect to the data symbols. Note that the average operation is applied<br />

before taking the logarithm, since it is not valid to perform the average<br />

through a non-linearity (the logarithm here) [83, p. 227].<br />

Most of the time, the result of the averaging step is a non-linear function<br />

of the sufficient statistics and the Eb ratio [99]. Instead of <strong>de</strong>aling with<br />

N0<br />

these rather complicated equations, approximations at low and high SNRs<br />

are preferred [83, pp. 226-250]. While previous references present specific<br />

results for each case consi<strong>de</strong>red, the whole approach is formalised in<br />

[100] for low SNR. It suggests expanding the exponential function of (3.30)


3.2 Maximum-Likelihood estimation 57<br />

into a Taylor series and to apply the average operation on each of its term<br />

with respect to the data symbols. Only the remaining terms which are<br />

still function of the param<strong>et</strong>ers to estimate are kept. In the case of the low<br />

SNR limit, this <strong>de</strong>composition is further limited to the lowest power of Eb<br />

N0 .<br />

Consi<strong>de</strong>ring the averaged likelihood function produced by this m<strong>et</strong>hod in<br />

the case of M-PSK modulations, one can notice that a similar expression<br />

would be obtained by applying a non-linear function to data symbols in<br />

or<strong>de</strong>r to cancel the influence of the modulation.<br />

However, the higher the or<strong>de</strong>r of the constellation is, the higher the power<br />

to be consi<strong>de</strong>red in the Taylor series expansion g<strong>et</strong>s. Moreover, taking the<br />

Mth power of M-PSK modulated samples introduces a M-fold ambiguity<br />

in the estimation process. It has been <strong>de</strong>monstrated in [82] that it is not<br />

necessary to call upon a non-linearity of or<strong>de</strong>r proportional to the dimension<br />

of the constellation. With the help of the polar representation of the<br />

sufficient statistics, applying a non-linearity of or<strong>de</strong>r n


58 Tools<br />

respect to the unknown vector param<strong>et</strong>er Φ and s<strong>et</strong>ting the result equal to<br />

zero [98].<br />

<br />

∂ΛL(r Φ) <br />

=0. (3.35)<br />

∂Φ<br />

Φ= ˆ Φ<br />

Due to the multiuser context, (3.35) produces as many equations as phase<br />

param<strong>et</strong>ers to be estimated (Nu in the present case). Obviously, this leads<br />

to the search of an optimum in a Nu-dimensional space. Estimation m<strong>et</strong>hods<br />

mentioned in Section 4 lead such search. On the other hand, authors<br />

have proposed suboptimum m<strong>et</strong>hods aimed at producing reliable estimates<br />

without having to lead such a time-consuming search. Without forg<strong>et</strong>ting<br />

the practical difficulties of the search for the optimum, the present<br />

study will stick to the equations <strong>de</strong>fining the optimum point in or<strong>de</strong>r to g<strong>et</strong><br />

a b<strong>et</strong>ter insight into the performance of the multiuser estimation process.<br />

3.3 Optimal estimator performance<br />

Since estimators are built upon samples of observed signals embed<strong>de</strong>d<br />

with random noise, they exhibit themselves a random behaviour. In this<br />

perspective, it is natural to quantify their performance by the <strong>de</strong>rivation<br />

of their moments. This <strong>de</strong>rivation is most often limited to the first- and<br />

second-or<strong>de</strong>r moments, the mean and the variance, since these are the only<br />

statistics required to compl<strong>et</strong>ely characterise a stochastic process<br />

<br />

from<br />

<br />

a<br />

Gaussian perspective. An estimator whose estimation error E θ ˆθ is<br />

null will be called unbiased. The optimal<br />

<br />

estimator should not only be un-<br />

ˆθ <br />

2<br />

biased but also have a variance E E ˆθ as small as possible [59,<br />

chapter 2].<br />

3.3.1 Cramér-Rao Lower Bound<br />

Checking that the estimator is unbiased is a pr<strong>et</strong>ty easy operation. The<br />

reference value is clear: a null bias. On the other hand, when moving<br />

to second-or<strong>de</strong>r statistics performance, the need of a benchmark appears,<br />

against which the variance of a prospective estimator can be tested. On<br />

that account, the Cramér-Rao Lower Bound (CRLB) is precious in estimation<br />

problems, since it provi<strong>de</strong>s a lower bound on the variance of any<br />

unbiased estimator.


3.3 Optimal estimator performance 59<br />

The CRLB can be <strong>de</strong>rived from the diagonal elements of the inverse of<br />

the Fisher information matrix. In the case of the estimation of a vector<br />

param<strong>et</strong>er θ, the elements of the Fisher information matrix I (θ) k,l are the<br />

second <strong>de</strong>rivatives of the log-likelihood function with respect to the param<strong>et</strong>ers:<br />

<br />

∂2ΛL (r θ)<br />

I (θ) k,l = E<br />

(3.36)<br />

∂θk ∂θl<br />

<br />

∂ΛL (r Φ) ∂ΛL (r θ)<br />

= E<br />

. (3.37)<br />

∂θk ∂θl<br />

The diagonal elements of the Fisher information matrix (3.36) can be interpr<strong>et</strong>ed<br />

as a measure of the curvature of the log-likelihood function. The<br />

sharper the function is, the greater the curvature, the lower the variance,<br />

and the b<strong>et</strong>ter the estimator are [59, p. 29]. Another interpr<strong>et</strong>ation of the<br />

CRLB is given in [101], where the bound is <strong>de</strong>rived from the param<strong>et</strong>er<br />

power spectral <strong>de</strong>nsity (psd). This spectrum is <strong>de</strong>rived from the psd of<br />

the transmitted signal tk (t) filtered by the channel ck (t), so that it represents<br />

the localisation of the available information about the param<strong>et</strong>er in<br />

the frequency domain.<br />

Since the CRLB only <strong>de</strong>pends on the Fisher information matrix which itself<br />

in turn only relies on the likelihood function, it is a global benchmark.<br />

The CRLB is <strong>de</strong>rived for a specific problem but irrespective of the estimator<br />

to be tested against that benchmark. It provi<strong>de</strong>s a fundamental lower<br />

limit on the variance of any estimator of all the param<strong>et</strong>ers of the problem.<br />

However, some of them are som<strong>et</strong>imes not to be estimated. Called unwanted<br />

param<strong>et</strong>ers [98, section 2.5], they have y<strong>et</strong> to be taken into account<br />

in the <strong>de</strong>rivation of the CRLB. This can then appear quite intricate. Some<br />

modified bounds easier to <strong>de</strong>rive have been introduced to tackle this issue.<br />

First, the Modified CRLB (MCRB) was introduced in [102] for cases where<br />

wanted θ and unwanted u param<strong>et</strong>ers coexist. Instead of working as usual<br />

with the pdf T rθ (r θ) of the received signal assuming the wanted param<strong>et</strong>ers<br />

θ, MCRB <strong>de</strong>als with the pdf T rθ,u (r θ, u) assuming both wanted<br />

and unwanted u param<strong>et</strong>ers. It ends in a looser variance bound than the<br />

CRLB. Nevertheless, approximate equality of CRLB and MCRB is found<br />

to occur in several cases among which phase recovery when all other param<strong>et</strong>ers<br />

and data are known [102, section IV], which will be the case in DA<br />

estimation (Chapter 4).


60 Tools<br />

Besi<strong>de</strong>s MCRB, Asymptotic CRLB (ACRB) [103] is another means of g<strong>et</strong>ting<br />

a lower bound on the variance while avoiding heavy computations.<br />

ACRB is <strong>de</strong>rived as the high SNR asymptote of the CRLB. It has been<br />

shown in [103] that it equals CRLB when the Fisher information matrix<br />

does not <strong>de</strong>pend on unwanted param<strong>et</strong>ers. As far as the MCRB is concerned,<br />

the ACRB lies above it. However, ACRB and MCRB join when<br />

the unwanted param<strong>et</strong>ers are discr<strong>et</strong>e or when they are continuous but<br />

<strong>de</strong>coupled from the wanted param<strong>et</strong>er(s).<br />

3.3.2 ML performance<br />

Now that the CRLB has been introduced as the variance benchmark for<br />

any unbiased estimator the reasons of the optimality of ML estimation can<br />

be listed.<br />

ML estimation provi<strong>de</strong>s the optimal estimator in the classic approach because<br />

its estimator is asymptotically consistent and asymptotically efficient.<br />

Consistency means that the estimator is unbiased, while efficiency<br />

relates to the fact that its variance reaches the CRLB. However, these properties<br />

are only valid asymptotically in the case of the ML estimator, that is<br />

to say, in the limit when the number of observed samples N + .<br />

Bearing in mind the Fisher information matrix I (θ), these two properties<br />

can be summed into one statement stating that the ML estimator is asymptotically<br />

distributed according to a Gaussian distribution Æ θ, I 1 (θ) .<br />

3.3.3 CRLB for multiuser phase estimation<br />

Before <strong>de</strong>riving the CRLB in the case of multiuser phase estimation, a regularity<br />

condition has to be fulfilled [98]<br />

<br />

∂ΛL (r Φ)<br />

E<br />

=0 k [1,Nu] . (3.38)<br />

∂φk<br />

For the system un<strong>de</strong>r investigation, this condition writes


3.3 Optimal estimator performance 61<br />

<br />

∂ΛL (r Φ)<br />

E<br />

∂φu<br />

since data symbols Im u<br />

m n<br />

= <br />

= <br />

<br />

2EuT<br />

N0<br />

<br />

2EuT<br />

N0<br />

N<br />

+<br />

m=1 n=<br />

N<br />

+<br />

m=1 n=<br />

E [I n u (I m u ) ⋆ m n<br />

] xu,u σ 2 m n<br />

I δ (m n) xu,u <br />

<br />

(3.39)<br />

(3.40)<br />

= 0 (3.41)<br />

are in<strong>de</strong>pen<strong>de</strong>nt i<strong>de</strong>ntically distributed random<br />

variables and xu,u δ (m n) =x0u,u is real.<br />

Having checked the regularity condition, the Fisher information matrix<br />

can be calculated. It will be diagonal since off-diagonal elements<br />

I (Φ) u,v<br />

v=u<br />

<br />

∂2ΛL (r Φ)<br />

= E<br />

∂φu ∂φv<br />

= <br />

<br />

2 Ô EuEv T<br />

N0<br />

j(φv φu)<br />

e<br />

N<br />

+<br />

m=1 n=<br />

E [I n v (I m u ) ⋆ m n<br />

] xu,v <br />

(3.42)<br />

(3.43)<br />

= 0 (3.44)<br />

are null due to the fact that data symbols I m u and In v<br />

are uncorrelated. Diagonal<br />

elements only remain which, with the help of (3.17), write<br />

<br />

∂2 <br />

ΛL (r Φ)<br />

I (Φ) u,u = E<br />

(3.45)<br />

= 2N EuTx0 u,uσ2 Iu<br />

(3.46)<br />

N0<br />

= 2N Es,u<br />

. (3.47)<br />

N0<br />

The CRLB is <strong>de</strong>fined as the inverse of the Fisher information matrix. Since<br />

the non-zero elements of the latter lie on the diagonal, the former is given<br />

by the inverse of the corresponding diagonal elements. In a feedforward<br />

implementation using a N-sample window, the CRLB for user u writes [85,<br />

p. 331]<br />

CRLBFF,u = 1<br />

2N<br />

∂φ 2 u<br />

Es,u<br />

N0<br />

1<br />

. (3.48)


62 Tools<br />

If a feedback implementation with closed-loop frequency response Hu (z)<br />

is preferred to a feedforward structure, the one-si<strong>de</strong>d loop bandwidth<br />

BN,u<br />

2BN,u =<br />

1<br />

2T<br />

<br />

1<br />

2T<br />

<br />

<br />

Hu(e 2jπfT <br />

<br />

)<br />

2<br />

df (3.49)<br />

is substituted for the size of the observation window N in relation (3.48)<br />

un<strong>de</strong>r the condition that either the loop noise is white or the one-si<strong>de</strong>d<br />

loop bandwidth BN,u is narrow [85, p. 349]. The CRLB then becomes<br />

<br />

Es,u<br />

CRLBFB,u = BN,uT<br />

N0<br />

1<br />

. (3.50)<br />

In line with the frequency domain interpr<strong>et</strong>ation of [101], it is noticed<br />

in [83, chapter 4] that these bounds are the same as those valid for phase<br />

recovery in the case of a pure unmodulated carrier. Things happen as if<br />

the optimal estimator aiming at achieving the CRLB compacted the signal<br />

power into a spectral line to be tracked by a PLL.<br />

3.4 FF estimation<br />

This section will <strong>de</strong>al with two aspects of the study of FF estimators, namely<br />

the <strong>de</strong>rivation of a closed-form suitable for performance evaluation and<br />

the calculation of the variance.<br />

3.4.1 Closed form of the estimator<br />

In the search for a closed-form expression of the ML FF estimator, the loglikelihood<br />

function (3.31) can be handled in two different ways in or<strong>de</strong>r to<br />

extract the param<strong>et</strong>er of interest. The first possibility consists in estimating<br />

the phase param<strong>et</strong>er φ or real functions of it (cos φ, sin φ). The second one<br />

would rather <strong>de</strong>al with the complex phasor e jφ . Both possibilities will be<br />

investigated in the following paragraphs.<br />

Estimation of the phase param<strong>et</strong>er<br />

Applying (3.35) to(3.31) leads to the following expression<br />

ˆφu = tan<br />

1 (Cu)<br />

(Cu)<br />

(3.51)


3.4 FF estimation 63<br />

where<br />

Cu =<br />

m=1<br />

N<br />

(I m u )⋆ y m u<br />

Nu <br />

Ek<br />

k=1<br />

k=u<br />

Eu<br />

e j ˆ φk<br />

N<br />

+<br />

m=1 n=<br />

(I m u )⋆ I n k<br />

n<br />

xm<br />

u,k . (3.52)<br />

Somehow, this expression of the ML FF phase estimator is similar to classic<br />

ones [7, p. 326]. In<strong>de</strong>ed, the estimator takes the argument of a complex<br />

number Cu partly built from the product b<strong>et</strong>ween the matched filter outputs<br />

y m u and the data symbols I m u . Y<strong>et</strong>, due to the multiuser context, (3.52)<br />

exhibits an additional contribution. The comparison of (3.7) with (3.52)<br />

shows that this additional contribution tends to cancel the influence of the<br />

interfering users coming from the matched filter outputs.<br />

However, relation (3.52) is not appropriate for performance evaluation,<br />

since the param<strong>et</strong>er estimate of one user is an implicit function of the<br />

param<strong>et</strong>er estimates of the interfering users. In or<strong>de</strong>r to be able to solve<br />

these equations with respect to the phase param<strong>et</strong>er, a linear relationship<br />

b<strong>et</strong>ween phase estimates has been searched by applying linearisation to<br />

different factors.<br />

Linearisation of the first <strong>de</strong>rivative about optimum The first linearisation<br />

attempt is a truncated Taylor series expansion of the first <strong>de</strong>rivative of<br />

the log-likelihood function around the optimal value ˆ Φ [85, pp. 343-344].<br />

This is to exploit the fact that the first <strong>de</strong>rivative equals zero at this point.<br />

<br />

<br />

∂ΛL (r)<br />

∂ΛL (r)<br />

∂2ΛL (r)<br />

= 0 =<br />

+<br />

∂Φ Φ=ˆΦ<br />

∂Φ<br />

∂Φ Φ=Φ0<br />

2<br />

<br />

ˆΦ Φ0<br />

Φ=Φ0<br />

(3.53)<br />

where Φ0 is the true value. Solving (3.53) for Φ gives<br />

ˆΦ Φ0 =<br />

∂ 2 ΛL (r)<br />

∂Φ 2<br />

1<br />

Φ=Φ0<br />

<br />

∂ΛL (r)<br />

∂Φ<br />

Φ=Φ0<br />

. (3.54)<br />

Consi<strong>de</strong>ring that the observation window is large enough, the Fisher information<br />

matrix can be substituted for the second <strong>de</strong>rivative of the loglikelihood<br />

function in (3.54)<br />

ˆΦ Φ0 = I (Φ) 1<br />

<br />

∂ΛL (r)<br />

. (3.55)<br />

∂Φ<br />

Φ=Φ0


64 Tools<br />

Unfortunately, (3.55) does not produce the wished linear relationship. In<strong>de</strong>ed,<br />

each phase estimate finally writes as a quotient of functions of the<br />

complex phasors. Consi<strong>de</strong>ring that such a result is not suited for performance<br />

evaluation, it has been disregar<strong>de</strong>d.<br />

Linearisation of complex exponential A difficulty in <strong>de</strong>aling with phase<br />

param<strong>et</strong>ers of interfering users in (3.52) is that they appear as arguments of<br />

a non-linear function, namely the exponential function. Instead of linearising<br />

the first-<strong>de</strong>rivative of the log-likelihood function, a second attempt<br />

involves linearising this exponential function. The Taylor series expansion<br />

is limited to the first or<strong>de</strong>r un<strong>de</strong>r the hypothesis of small estimation error.<br />

e j ˆ <br />

φk = jφk e + ˆφk φk je jφk (3.56)<br />

= e jφk<br />

<br />

1+j ˆφk φk . (3.57)<br />

Applying this linearisation to (3.31) leads to the following condition<br />

⎧<br />

e<br />

⎪⎨<br />

<br />

jφu<br />

<br />

1 j ˆφu φu<br />

⎧<br />

N<br />

(I<br />

⎪⎨<br />

m=1<br />

m u )⋆ ym u<br />

Nu <br />

<br />

1+j ˆφk φk<br />

⎫<br />

⎫<br />

⎪⎬<br />

⎪⎬ =0. (3.58)<br />

⎪⎩<br />

⎪⎩<br />

Solving (3.58) gives<br />

where<br />

Du =<br />

N<br />

m=1<br />

Nu<br />

<br />

k=1<br />

k=1<br />

N<br />

m=1 n=<br />

(I m u ) ⋆ y m u<br />

Ek<br />

Eu ejφk<br />

+<br />

Ek<br />

Eu ejφk<br />

(I m u ) ⋆ I n k<br />

xm n<br />

u,k<br />

ˆφu φu = e jφu <br />

Du<br />

(e jφu Du)<br />

N<br />

1+j ˆφk φk<br />

⎪⎭<br />

+<br />

m=1 n=<br />

⎪⎭<br />

(I m u )⋆ I n k<br />

(3.59)<br />

xm n<br />

u,k .<br />

(3.60)<br />

Thanks to the linearisation, the tan 1 non-linearity in (3.52) has disappeared.<br />

By regarding some terms of (3.52) as negligible (See Section 4.2.2),


3.4 FF estimation 65<br />

the phase estimation error ˆ φu φu can become linearly <strong>de</strong>pen<strong>de</strong>nt of other<br />

phase estimation errors. As it will be shown in the next chapter, this paves<br />

the way for the performance analysis of the ML FF estimator.<br />

Rectangular representation Still estimating the real phase param<strong>et</strong>er φ,<br />

another strategy might be to try to recover cos φ and sin φ, since φ always<br />

appears as argument of a complex exponential e jφ =cosφ + j sin φ. This<br />

strategy might be applied in two different ways, either by directly estimating<br />

cos φ and sin φ, or by estimating φ and implementing the estimator so<br />

as to track cos φ and sin φ [83, pp. 216-226]. However, as far as the former<br />

case is concerned, <strong>de</strong>rivating with respect to φ or to (cos φ, sin φ) does not<br />

bring out a significantly new estimator, since all these param<strong>et</strong>ers are tied<br />

tog<strong>et</strong>her as follows<br />

This leads to<br />

dΛ =<br />

∂Λ<br />

∂φ =<br />

=<br />

∂Λ<br />

∂φ =0<br />

∂Λ<br />

∂Λ<br />

d cos φ + d sin φ<br />

∂ cos φ ∂ sin φ<br />

(3.61)<br />

∂Λ ∂ cos φ ∂Λ ∂ sin φ<br />

+<br />

∂ cos φ ∂φ ∂ sin φ ∂φ<br />

(3.62)<br />

∂Λ<br />

∂Λ<br />

sin φ + cos φ.<br />

∂ cos φ ∂ sin φ<br />

(3.63)<br />

∂Λ ∂Λ<br />

=<br />

∂ sin φ ∂ cos φ<br />

tan φ. (3.64)<br />

The latter is a question of implementation rather than a way to obtain a<br />

new expression of the estimator. It does not modify the analytical performance<br />

evaluation to be presented in the following chapters, since the<br />

estimator is still <strong>de</strong>rived from (3.52). This is the reason why it will not be<br />

studied here.<br />

Estimation of the phasor<br />

Having consi<strong>de</strong>red the phase estimation in the real space, through the<br />

phase param<strong>et</strong>er φ as well as through functions of it (cos φ, sin φ), it is<br />

time to move to the complex space.<br />

Planar filtering In (3.52), the phase estimate ˆ φu appears to be obtained<br />

as the argument of the complex number Cu. Instead of tracking the argument,<br />

another possible approach is to track the phasor itself. This tech-


66 Tools<br />

nique is called planar filtering [85, p. 312]. However, it is rather a postprocessing<br />

technique [104], in the sense that the optimum is still <strong>de</strong>fined<br />

with respect to the phase. Only the tracking takes the complex specificity<br />

of the phasor into account. Planar filtering improves the tracking [85,<br />

p. 414] but, as far as estimation is concerned, it does not <strong>de</strong>fine another<br />

estimate than the one obtained through phase estimation.<br />

Complex <strong>de</strong>rivation Regarding planar filtering as a post-processing improvement<br />

of a structure based on phase estimation, one could try to directly<br />

estimate the complex exponential e jφ . In the ML approach, the first<br />

<strong>de</strong>rivative to s<strong>et</strong> to zero is then taken with respect to a complex number,<br />

that is to say that the first <strong>de</strong>rivative of a real function with respect to<br />

a complex variable is to be computed. For this <strong>de</strong>rivative to exist, the<br />

Cauchy-Rieman conditions are to be fulfilled.<br />

In the most general case, s<strong>et</strong>ting ejφ = x + jywhere x and y are respectively<br />

the real and imaginary parts of ejφ , these conditions state that<br />

∂ [f (x, y)]<br />

=<br />

∂x<br />

∂ [f (x, y)]<br />

(3.65)<br />

∂y<br />

∂ [f (x, y)]<br />

∂x<br />

= ∂ [f (x, y)]<br />

. (3.66)<br />

∂y<br />

However, the likelihood function Λ is a real function of (x, y)<br />

Λ(x, y) =<br />

<br />

Ae jφ<br />

which does not fulfill Cauchy-Rieman conditions<br />

(3.67)<br />

= [(Ax + jAy)(x + jy)] (3.68)<br />

= Ax x Ay y (3.69)<br />

∂[f(x,y)]<br />

∂[f(x,y)]<br />

∂x = Ax; ∂y = 0<br />

∂[f(x,y)]<br />

∂x = 0; ∂[f(x,y)]<br />

(3.70)<br />

∂y = Ay.<br />

Derivating the likelihood function with respect to the phasor is thus not<br />

possible.<br />

3.4.2 Variance approximation<br />

The closed-form expressions of the ML FF estimator involve a quotient.<br />

This mathematical relationship is not easy to handle at the time of computing<br />

the variance. Two options appeared in the literature to circumvent


3.5 Performance evaluation of DD estimators 67<br />

this difficulty.<br />

From (3.54), the rea<strong>de</strong>r can notice that the phase estimation error is equal<br />

to a ratio b<strong>et</strong>ween first- and second-<strong>de</strong>rivative of the log-likelihood function.<br />

In [105, 106], the statistical fluctuations of the second <strong>de</strong>rivative are<br />

assumed to be small with respect to its mean value, so as to substitute the<br />

second <strong>de</strong>rivative for its mean. In terms of variance, this finally gives<br />

σ 2<br />

Φ ˆΦ =<br />

<br />

∂2ΛL (r)<br />

E<br />

∂Φ2 <br />

Φ=Φ0<br />

2<br />

E<br />

∂ΛL (r)<br />

∂Φ<br />

2<br />

Φ=Φ0<br />

<br />

. (3.71)<br />

On the other hand, starting from (3.52), the expectation of the square of<br />

the argument may turn into [104]<br />

un<strong>de</strong>r the hypotheses that<br />

<br />

E [arg (Cu)] 2 <br />

E [ (Cu)]<br />

=<br />

2<br />

E [ (Cu)] 2<br />

(3.72)<br />

E [ (Cu)] = 0 (3.73)<br />

<br />

E (Cu) E [ (Cu)] 2<br />

E [ (Cu)] 2<br />

(3.74)<br />

<br />

E (Cu) E [ (Cu)] 2<br />

E [ (Cu)] 2 . (3.75)<br />

Only the first option (3.71) will be applied in the next chapters. However,<br />

it was worth mentioning the second one (3.72) for review purposes.<br />

3.5 Performance evaluation of DD estimators<br />

The performance of DD estimators is not an easy issue. It is a coupled<br />

problem, since <strong>de</strong>cision errors are eager to cause estimation errors which<br />

in turn will affect the <strong>de</strong>cision process and so on. Few contributions really<br />

tackle the problem. Most of the time, <strong>de</strong>cisions are assumed to be correct,<br />

which, as far as performance is concerned, brings back to DA analysis. A<br />

less ru<strong>de</strong> approach to take into account the possible faulty outcome of the<br />

<strong>de</strong>cision process is to weight the performance <strong>de</strong>rived in the case of DA<br />

structures by the probability of error PE [84, p. II-7].


68 Tools<br />

No such approximation is ma<strong>de</strong> in [87]. The performance of carrier phase<br />

recovery systems for single-user transmissions over AWGN channels is<br />

calculated using<br />

<br />

analytical<br />

<br />

expressions of products b<strong>et</strong>ween data symbols<br />

⋆<br />

and <strong>de</strong>cisions E Îm k Im <br />

k . Generalising this approach to multiuser systems<br />

working over non i<strong>de</strong>al channels is pr<strong>et</strong>ty intricate. A look at (3.31)<br />

reveals first that the expectations <strong>de</strong>rived in [87], which involve symbols<br />

and <strong>de</strong>cisions from the same user at the same time instant, are now to span<br />

over all users, as a result of the non-orthogonality b<strong>et</strong>ween users, and over<br />

the whole observation window due the time<br />

<br />

dispersiveness<br />

<br />

of the chan-<br />

⋆<br />

nel. In the most general case, expectations E Îm k In <br />

l will be computed<br />

for any pair (k, l) and (m, n).<br />

Moreover, (3.31) contains contributions whose positive effect is to mitigate<br />

interference. These terms involve products of <strong>de</strong>cisions, possibly related<br />

to different users or different time instants, which leads to the computation<br />

of<br />

⋆Î <br />

m n<br />

E Îk l<br />

<br />

= E (â m k ânl )+E<br />

<br />

ˆb mˆn k bl + j E â m k ˆb n <br />

l<br />

<br />

E ˆb m<br />

k â n <br />

l . (3.76)<br />

The <strong>de</strong>rivation of such expectations is the subject of the current section.<br />

Consi<strong>de</strong>ring M-PSK constellations with unit variance (σ 2 I<br />

tions hereafter will be used<br />

p q<br />

Rk,l p q<br />

Ik,l =<br />

=<br />

⎧<br />

⎨<br />

⎩<br />

⎧<br />

⎨<br />

⎩<br />

<br />

El<br />

Ek <br />

<br />

Ô <br />

2 El<br />

2 Ek <br />

<br />

El<br />

Ek <br />

<br />

Ô <br />

2 El<br />

2 Ek <br />

e j(φl ˆ <br />

φk) p q<br />

xk,l <br />

e j(φl ˆ <br />

φk) p q<br />

xk,l e j(φl ˆ <br />

φk) p q<br />

xk,l <br />

e j(φl ˆ <br />

φk) p q<br />

xk,l Similarly, for the noise samples<br />

ν m k =<br />

<br />

e j ˆ ν<br />

φk m<br />

νk m k =<br />

<br />

e j ˆ φk m<br />

νk <br />

(BPSK)<br />

(QPSK)<br />

(BPSK)<br />

(QPSK).<br />

=1), the nota-<br />

(3.77)<br />

(3.78)<br />

(3.79)<br />

<br />

. (3.80)<br />

For the sake of simplicity, the following <strong>de</strong>velopments will be limited to<br />

conventional hard <strong>de</strong>cisions, although optimal and suboptimal <strong>de</strong>tection


3.5 Performance evaluation of DD estimators 69<br />

strategies have been mentioned in Section 2.2.1.<br />

where<br />

â m k<br />

ˆ b m k<br />

=<br />

=<br />

A p<br />

k<br />

= Āp<br />

k<br />

=<br />

<br />

sign (Am k ) = sign Ām k + νm <br />

Ô k<br />

2<br />

2 sign (Am Ô<br />

2<br />

k ) = 2 sign Ām k + νm <br />

k<br />

<br />

sign (Bm k ) = sign B¯ m<br />

k + νm <br />

Ô k<br />

2<br />

2 sign (Bm Ô<br />

2<br />

k ) = 2 sign B¯ m<br />

k + νm <br />

k<br />

+ νp<br />

k<br />

+ <br />

q=<br />

B p<br />

k<br />

= ¯ B p<br />

=<br />

k + νp<br />

k<br />

+ <br />

q=<br />

a q p<br />

kRq k,k bq<br />

<br />

p<br />

kIq k,k<br />

a q q<br />

kIp k,k<br />

<br />

q<br />

+ bq<br />

kRp k,k<br />

Nu <br />

+<br />

l=1<br />

l=k<br />

Nu <br />

+<br />

l=1<br />

l=k<br />

+<br />

q=<br />

+<br />

q=<br />

<br />

<br />

a q p<br />

l<br />

Rq<br />

a q q<br />

l<br />

Ip<br />

k,l<br />

k,l b q<br />

l<br />

3.5.1 Direct space - Gaussian probability integral<br />

(BPSK)<br />

(QPSK)<br />

(BPSK)<br />

(QPSK)<br />

(3.81)<br />

(3.82)<br />

<br />

p<br />

Iq<br />

k,l + ν p<br />

k<br />

(3.83)<br />

<br />

q<br />

+ bq<br />

l<br />

Rp<br />

k,l + ν p<br />

k .<br />

(3.84)<br />

In direct space, the mathematical expectations of (3.76) turns into a double<br />

integral over the noise samples alleviating the arguments (3.83) and (3.84).<br />

In the most general case, the calculation of these expectations ends in a<br />

non separable double integral, since their limits are functions of the same<br />

data symbols. For instance, here is the expression of the expectation of the<br />

product b<strong>et</strong>ween the <strong>de</strong>cision on the I-branch for user u at instant m and<br />

the <strong>de</strong>cision on the Q-branch for user v at instant n, for QPSK-modulated


70 Tools<br />

symbols since there is no information on the Q-branch in BPSK<br />

<br />

E â m u ˆb n <br />

v<br />

= 1<br />

2 EI,ν<br />

m<br />

sign Āu + ν m <br />

u sign ¯B n<br />

v + ν n v<br />

(3.85)<br />

= 1<br />

2 EI<br />

⎧<br />

⎨<br />

+ +<br />

sign<br />

⎩<br />

Ām u + νm <br />

u sign ¯B n<br />

v + νn <br />

v<br />

Tνm u ,νn v (νm u , νn v ) dνm u dνn ⎫<br />

⎬<br />

⎭<br />

v<br />

(3.86)<br />

⎧<br />

⎪⎨<br />

+<br />

= EI<br />

⎪⎩ +<br />

+<br />

Tν m u ,ν n v (νm u , ν n v ) dν m u dν n v<br />

Āmu ¯ Bn v<br />

Ām u<br />

¯ Bn v<br />

Tνm u ,νn v (νm u , νn v ) dνm u dνn v<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

1<br />

2 .<br />

(3.87)<br />

The <strong>de</strong>cisions are the results of sign functions whose arguments are the<br />

sum of a linear combination of random variables related to the data and<br />

of noise samples. Developing the expectation over the noise brings from<br />

relation (3.85) to(3.86). However, integrating the product of these sign<br />

functions over the plane (ν m u , ν n v ) is equivalent to integrate it over subdomains<br />

where it takes values ¦1 (Figure 3.2).<br />

¯ B n v<br />

ν n v<br />

Ām u<br />

ν m u<br />

Figure 3.2: Sub-domains in the plane (ν m u , ν n v )


3.5 Performance evaluation of DD estimators 71<br />

Exploiting the Gaussian characteristic of the noise samples, one finally<br />

g<strong>et</strong>s relation (3.87). Similar relations can be written for the other expectations<br />

of (3.76).<br />

The fact that the integration limits in (3.87) <strong>de</strong>pend on data symbols complicates<br />

the averaging operation. Recently, an alternative form of the Gaussian<br />

probability integral has been presented [107] and applied to the <strong>de</strong>rivation<br />

of the error probability for various communication systems [108].<br />

Preliminary studies show that this powerful transformation seems very<br />

promising for performance evaluation of estimators. Unfortunately, time<br />

lacked to lead a thorough analysis of this new m<strong>et</strong>hod and to apply it to<br />

the problem un<strong>de</strong>r investigation in the present work.<br />

Coming back to the subject of relation (3.87), the rea<strong>de</strong>r notices that it is<br />

suited for numerical computation. However, if the goal is to g<strong>et</strong> an analytical<br />

solution, moving to the reciprocal space and using the characteristic<br />

function has appeared to be an elegant move.<br />

3.5.2 Reciprocal space - Characteristic function<br />

Several m<strong>et</strong>hods relying on the characteristic function have been proposed<br />

in or<strong>de</strong>r to compute the error probability in presence of noise and interference.<br />

The first to be mentioned here is <strong>de</strong>veloped in [89, 109]. The Gaussian<br />

probability integral of the noise samples, whose limits are function of the<br />

interference like in (3.87), is solved assuming the interference. As a result,<br />

the solution has the form of the Gaussian probability function Q (x)<br />

Q (x) = 1<br />

2 erfc<br />

<br />

xÔ2<br />

= 1<br />

<br />

xÔ2<br />

1 erf<br />

2<br />

= 1<br />

2π<br />

+<br />

x<br />

<br />

exp<br />

u2<br />

2<br />

(3.88)<br />

<br />

du (3.89)<br />

whose argument <strong>de</strong>pends on the structure of the interference. The calculation<br />

of the error probability would then require to average the obtained<br />

result over the interference pdf. At first sight, this is a very consuming<br />

operation, since it involves a number of computations which is exponential<br />

in the number of symbols contributing to the interference. However,


72 Tools<br />

<strong>de</strong>veloping the error function into its Fourier series expansion introduces<br />

a s<strong>et</strong> of basis exponential functions which, combined to the interference<br />

pdf and the integration, makes appear the characteristic function of the<br />

interference. The interesting point is that taking into account the interference<br />

through its characteristic function requires a number of computations<br />

which is only linear in the number of involved symbols.<br />

Another m<strong>et</strong>hod for <strong>de</strong>riving the error probability is introduced in [110]<br />

for systems plagued with ISI. A generalisation to systems suffering from<br />

different kinds of interference is presented in [111]. This m<strong>et</strong>hod embeds<br />

all contributions, signal, interference, and noise into one random variable.<br />

The moment-generating function of this global random variable is<br />

obtained by Fourier transform. It is well-known that the inverse Fourier<br />

transform of the moment-generating function gives the pdf of the random<br />

variable. Computing the error probability turns into a <strong>de</strong>finite integral<br />

of this pdf. As in [89, 109], the use of the characteristic function avoids<br />

exponential computations in favour of linear complexity computations.<br />

Moreover, switching the probability integral and the inverse Fourier transform<br />

integral brings another simplification. This m<strong>et</strong>hod which relies on<br />

the Fourier transform has also been applied using the Laplace transform<br />

[112].<br />

The main advantage of these techniques is the possibility to perform the<br />

averaging operation over the interference with a computational effort linear<br />

in the number of inclu<strong>de</strong>d symbols. This has been a motivation for<br />

applying the second one to the analysis of DD FB estimators. The <strong>de</strong>tails<br />

of the <strong>de</strong>rivation are presented in Appendix E and their exploitation for<br />

the <strong>de</strong>rivation of the performance of DD FB structures will be shown in<br />

Chapter 5.<br />

3.6 Conclusions<br />

The uplink segment of the mobile DS-CDMA communication system un<strong>de</strong>r<br />

investigation has been presented in this chapter. Aiming at performing<br />

coherent reception, the carrier phase has to be recovered. Un<strong>de</strong>r the<br />

assumption of uniformly distributed random variables, ML has been introduced<br />

as the optimal estimation m<strong>et</strong>hod. ML estimators are known to<br />

be asymptotically unbiased and their variance is boun<strong>de</strong>d by the CRLB,<br />

which has been <strong>de</strong>rived.


3.6 Conclusions 73<br />

On the other hand, some analytical issues have been <strong>de</strong>alt with in this<br />

chapter. First, the means to <strong>de</strong>rive a closed-form expression of a ML FF estimate<br />

and to compute its performance have been presented. Next, several<br />

m<strong>et</strong>hods for studying the working of DD structures have been explained.<br />

Using the material exposed in the first two chapters, ML phase estimators<br />

will be <strong>de</strong>rived in the following chapters, first in a DA mo<strong>de</strong> (Chapter 4),<br />

then in a DD mo<strong>de</strong> (Chapter 5).


Chapter 4<br />

Data-Ai<strong>de</strong>d<br />

This chapter <strong>de</strong>als with ML estimation of phase param<strong>et</strong>ers in DA context.<br />

In such situation, the receiver has a perfect knowledge of the symbols I p<br />

k<br />

transmitted by user k. This happens during acquisition sessions on the<br />

link b<strong>et</strong>ween the transmitter and the receiver, when the transmitter emits<br />

pre<strong>de</strong>fined symbol sequences used at the receiver to estimate the param<strong>et</strong>ers<br />

of the link.<br />

As mentioned in Section 3.2.3, a necessary but not sufficient condition for<br />

<strong>de</strong>riving the ML estimate is to s<strong>et</strong> to zero the first <strong>de</strong>rivative of the loglikelihood<br />

function with respect to the param<strong>et</strong>er of interest. Calculating<br />

the first <strong>de</strong>rivative of (3.31) with respect to Φ and s<strong>et</strong>ting the result equal<br />

to zero leads to a s<strong>et</strong> of Nu conditions of the type<br />

<br />

∂ΛL(Φ) <br />

<br />

∂φu<br />

<br />

Φ=ˆΦ<br />

⎡<br />

N<br />

⎤<br />

= 2EuT<br />

N0<br />

⎢<br />

e<br />

⎢<br />

⎢<br />

⎣<br />

j ˆ φu<br />

Nu <br />

k=1<br />

k=u<br />

m=1<br />

(I m u )⋆ y m u<br />

Ek<br />

Eu ej( ˆ φk ˆ φu) N<br />

+<br />

m=1 n=<br />

(Im u )⋆I n n<br />

k<br />

xm<br />

u,k<br />

= 0. (4.1)<br />

The fact that this estimation process works in DA mo<strong>de</strong> appears through<br />

the <strong>de</strong>pen<strong>de</strong>ncy upon true data symbols I m u and not upon estimates Îm u .<br />

Relation (4.1) can <strong>de</strong>scribe feedback as well as feedforward phase recovery<br />

implementations. The former is built as a locked loop tracking the phase<br />

⎥<br />


76 Data-Ai<strong>de</strong>d<br />

according to an error signal u m u,DA<br />

<br />

∂ΛL(Φ) <br />

<br />

∂φu<br />

Φ= ˆ Φ<br />

= 2EuT<br />

N0<br />

N<br />

m=1<br />

u m u,DA<br />

=0 (4.2)<br />

while the latter explicitly computes a closed-form estimate of the phase<br />

param<strong>et</strong>er ˆ φu. Both implementations will be studied in the following sections.<br />

4.1 Feedback<br />

The subject of the present section is the Multiuser (MU) DA ML FB phase<br />

estimator that can be <strong>de</strong>rived from (4.1). This section will <strong>de</strong>al with feedback<br />

structures working in tracking mo<strong>de</strong>. In this mo<strong>de</strong>, the recovery loop<br />

is tracking the variations of the param<strong>et</strong>er, starting from a rough estimate<br />

obtained during the acquisition mo<strong>de</strong>. Provi<strong>de</strong>d a proper <strong>de</strong>sign of the<br />

loop, the estimate in tracking loop exhibits small fluctuations around the<br />

true value of the param<strong>et</strong>er.<br />

The multiuser recovery loop is shown in Figure 4.1 for a 2-user case. Without<br />

the signal flows b<strong>et</strong>ween the two main branches, it would appear as<br />

two Single-User (SU) recovery loops working in parallel. The exchange of<br />

information b<strong>et</strong>ween them turns the structure into an MU estimator. This<br />

loop is driven by the error signal u m u,DA<br />

u m ⎡<br />

e<br />

⎢<br />

u,DA = ⎣<br />

j ˆ φm u (Im u ) ⋆ ym u<br />

Nu <br />

<br />

Ek<br />

Eu ej( ˆ φm k ˆ φm u ) +<br />

n=<br />

k=1<br />

k=u<br />

(Im u ) ⋆ In n<br />

k<br />

xm<br />

u,k<br />

⎤<br />

⎥<br />

⎦ . (4.3)<br />

It results from two contributions. Besi<strong>de</strong>s the classic one, relying on matched<br />

filter outputs ym u [7, 83], the rea<strong>de</strong>r notices a second term which <strong>de</strong>pends<br />

only on interfering users. This term comes out from the fact that the<br />

log-likelihood function Λ(r Φ) takes into account the multiuser context of<br />

the transmission. The <strong>de</strong>rivation of a log-likelihood function missing this<br />

aspect leads to the SU DA ML FB estimator, only <strong>de</strong>pending on the term<br />

involving the output ym u of the filter matched to the equivalent channel<br />

mo<strong>de</strong>l of user u.


(t)<br />

h ⋆ u ( t)<br />

h ⋆ v ( t)<br />

y m u<br />

e j ˆ φ m u<br />

e j ˆ φ m v<br />

y m v<br />

(.) ⋆<br />

NCO<br />

NCO<br />

e j ˆ φ m u y m u<br />

<br />

<br />

e j ˆ φ m v y m v<br />

u m u<br />

u m v<br />

+<br />

-<br />

(.) ⋆<br />

Figure 4.1: 2-user DA phase recovery loop<br />

-<br />

+<br />

(.) ⋆<br />

<br />

(.) ⋆<br />

(.) ⋆<br />

I m u<br />

x m u,v<br />

I m v<br />

4.1 Feedback 77


78 Data-Ai<strong>de</strong>d<br />

The introduction of (3.7) into (4.3) gives a b<strong>et</strong>ter insight into the workings<br />

of the MU FB loop.<br />

u m ⎡<br />

⎢ e<br />

⎢<br />

u,DA = ⎢<br />

⎣<br />

j(φu ˆ φm u ) +<br />

n=<br />

+ Nu <br />

e<br />

k=1<br />

k=u<br />

j(φk ˆ φm u ) Ek<br />

Eu<br />

Nu <br />

e<br />

k=1<br />

k=u<br />

j( ˆ φm k ˆ φm u ) Ek<br />

Eu<br />

+ e j ˆ φm u Im u νm u,DA<br />

(Im u )⋆I n u xm<br />

n<br />

u,u<br />

n=<br />

+<br />

n=<br />

+<br />

(Im u ) ⋆ In n<br />

k<br />

xm<br />

u,k<br />

(Im u )⋆I n n<br />

k<br />

xm<br />

u,k<br />

⎤<br />

⎥ . (4.4)<br />

⎥<br />

⎦<br />

The second term in (4.4) is the MAI contribution which entered the loop<br />

through the matched filter output (3.7). This contribution <strong>de</strong>pends on the<br />

difference b<strong>et</strong>ween the phases of the interfering users φk and the current<br />

phase estimate for the user of interest ˆ φm u , as also noticed in [96]. That interference<br />

is counterbalanced by the third term, introducing a correction<br />

<strong>de</strong>rived from the log-likelihood function (3.31).<br />

The performance of recovery loops driven by the error signal u m u,DA (4.4)<br />

will be <strong>de</strong>scribed in the next sections. The jitter variance will serve as performance<br />

measure. Unlike what will be done for FF estimators in the next<br />

section, the pdf of the FB phase estimation error has not been <strong>de</strong>rived in<br />

the present work. As mentioned in Section 2.3, such pdf has been <strong>de</strong>rived<br />

in a single-user context in [86, p. 90] and in a multiuser context in [96].<br />

Notice, however, that the work presented in [96] is performed in the Bayes<br />

approach, while the present thesis follows the Fisher approach.<br />

4.1.1 Open-loop study<br />

Operating point<br />

As <strong>de</strong>scribed in Section 2.3.2, the first step in the study of a recovery loop<br />

in tracking mo<strong>de</strong> is to <strong>de</strong>termine the operating point of the loop, that is,<br />

the position for which the error signal driving the loop will be null in the<br />

mean. The expressions of the mean of the error signal to be established in<br />

the following will serve for this purpose. They will also help to build a<br />

linear version of the loop, which is more suited for closed-loop investigations.


4.1 Feedback 79<br />

BPSK modulation The data symbols I p<br />

k in (4.4) reduce to their real part.<br />

then writes<br />

U BPSK<br />

u,DA<br />

U BPSK<br />

u,DA<br />

<br />

= E u m u,DAˆΦ <br />

=0, Φ=∆<br />

<br />

= e j∆u<br />

+ <br />

E a m u anu <br />

ˆΦ =0, Φ=∆<br />

⎡<br />

⎢<br />

+ ⎣<br />

n=<br />

Nu<br />

<br />

Ek<br />

k=1<br />

k=u<br />

⎡<br />

⎢Nu<br />

⎣<br />

k=1<br />

k=u<br />

Eu<br />

Ek<br />

Eu<br />

e j(δk,u+∆u) + <br />

n=<br />

e j(δk,u+∆u ∆k) + <br />

n=<br />

m n<br />

xu,u <br />

<br />

E a m u ank <br />

ˆΦ =0, Φ=∆ x<br />

m n<br />

u,k<br />

<br />

E a m u ank <br />

ˆΦ =0, Φ=∆ x<br />

⎤<br />

⎥<br />

⎦<br />

(4.5)<br />

m n<br />

u,k<br />

+E (a m u ν m u ) (4.6)<br />

= K BPSK<br />

D,u sin ∆u (4.7)<br />

where ∆k = φk ˆ φk and δk,l = φk φl. K BPSK<br />

D,u = σ 2 Iu x0 u,u is the phase<br />

<strong>de</strong>tector gain.<br />

In (4.7), U BPSK<br />

u,DA <strong>de</strong>pends on ∆u through a sinusoidal function. This means<br />

that driving the error signal of the loop to zero (U BPSK<br />

u,DA =0) is equivalent<br />

to have ∆u =0, which is an unbiased operating point. Also worth noticing<br />

is the fact that the MAI contribution (second term) is cancelled by the<br />

correcting term (third term) as soon as the phase error is recovered on the<br />

interfering loops (∆k =0 k = u). In<strong>de</strong>ed, it is not surprising to find the<br />

same final result as for SU estimators. In the DA context, the MAI can be<br />

cancelled thanks to the perfect knowledge of interfering users’ messages.<br />

Multiuser relations are thus equivalent to their single-user counterparts.<br />

Finally, notice that DA estimation introduces no phase ambiguity. Thus<br />

exhibits a 2π-periodicity [83, p. 204].<br />

U BPSK<br />

u,DA<br />

U BP SK<br />

u,DA<br />

Drawing<br />

KBP SK with respect to ∆u produces a S-curve of sinusoidal shape.<br />

D,u<br />

Around the operating point, this sinusoidal curve presents an interesting<br />

linear area in that it produces a mean error signal Uu directly proportional<br />

to the phase estimation error ∆u. On the other hand, it is well-known that<br />

the slope of the S-curve at the operating point can be reduced due to <strong>de</strong>-<br />

⎤<br />

⎥<br />


80 Data-Ai<strong>de</strong>d<br />

cision errors [83, p. 207]. This will be illustrated in the next chapter, which<br />

<strong>de</strong>als with DD structures. In the DA case, however, there is no <strong>de</strong>cision<br />

error. The slope is thus equal to 1.<br />

QPSK modulation Moving to QPSK modulation, both real and imagin-<br />

QP SK<br />

ary parts of data symbols will be taken into account in (4.4). Uu,DA writes<br />

QP SK<br />

Uu,DA <br />

= E u m <br />

u,DAˆΦ =0, Φ=∆<br />

⎧<br />

⎨<br />

= <br />

⎩ ej∆u<br />

⎡<br />

+<br />

⎣<br />

n=<br />

E<br />

<br />

am u an <br />

u ˆΦ =0, Φ=∆<br />

<br />

+E bm u bnu ⎤ ⎫<br />

⎬<br />

⎦ m n<br />

xu,u ˆΦ =0, Φ=∆ ⎭<br />

⎧<br />

⎨<br />

+<br />

⎩ ej∆u<br />

⎡<br />

+<br />

⎣<br />

n=<br />

E<br />

<br />

am u bnu <br />

ˆΦ =0, Φ=∆<br />

<br />

+E bm u an ⎤ ⎫<br />

⎬<br />

⎦ m n<br />

xu,u u ˆΦ =0, Φ=∆ ⎭<br />

⎧<br />

⎫<br />

⎪⎨<br />

+<br />

⎪⎩<br />

⎧<br />

⎪⎨<br />

+<br />

⎪⎩<br />

⎧<br />

⎪⎨<br />

<br />

⎪⎩<br />

⎧<br />

⎪⎨<br />

<br />

⎪⎩<br />

Nu <br />

k=1<br />

k=u<br />

n=<br />

+<br />

Nu <br />

k=1<br />

k=u<br />

n=<br />

+<br />

Nu <br />

k=1<br />

k=u<br />

n=<br />

+<br />

Nu <br />

k=1<br />

k=u<br />

n=<br />

+<br />

Ek<br />

Eu ej(δk,u+∆u)<br />

⎡<br />

⎣ E<br />

<br />

am u an k <br />

ˆΦ =0, Φ=∆<br />

<br />

+E bm u bn k <br />

ˆΦ =0, Φ=∆<br />

Ek<br />

Eu ej(δk,u+∆u)<br />

⎡<br />

⎣ E<br />

<br />

am u bn k <br />

ˆΦ =0, Φ=∆<br />

<br />

+E bm u an k <br />

ˆΦ =0, Φ=∆<br />

Ek<br />

Eu ej(δk,u+∆u ∆k)<br />

⎡<br />

⎣ E<br />

<br />

am u an k <br />

ˆΦ =0, Φ=∆<br />

<br />

+E bm u bn k <br />

ˆΦ =0, Φ=∆<br />

Ek<br />

Eu ej(δk,u+∆u ∆k)<br />

⎡<br />

⎣ E<br />

<br />

am u bn k <br />

ˆΦ =0, Φ=∆<br />

<br />

+E bm u an k <br />

ˆΦ =0, Φ=∆<br />

⎤<br />

⎦ m n<br />

xu,k ⎤<br />

⎦ m n<br />

xu,k ⎤<br />

⎦ m n<br />

xu,k ⎤<br />

⎦ m n<br />

xu,k ⎪⎬<br />

⎪⎭<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

(4.8)


4.1 Feedback 81<br />

QP SK<br />

where K<br />

D,u = σ 2 Iu x0 u,u<br />

+E (a m u νm u ) E (bmu νm u ) (4.9)<br />

=<br />

QP SK<br />

KD,u sin ∆u (4.10)<br />

. The operating point, the S-curve shape, and<br />

periodicity are thus i<strong>de</strong>ntical to those <strong>de</strong>rived for BPSK-modulated symbols.<br />

Loop noise<br />

The <strong>de</strong>rivation of the mean of the error signal u m u,DA<br />

is a first step towards<br />

the building of a linearised version of the loop. This linearised loop is used<br />

to perform the closed-loop performance study in the presence of noise and<br />

interference. The variance of the phase jitter is evaluated as the variance<br />

of the loop noise filtered by the linearised loop [79, section 3.1]. The loop<br />

noise embeds additive noise, self-noise (due to the random nature of the<br />

signal ¢ signal terms related to a single user), and cross-noise (due to the<br />

random nature of the signal ¢ signal terms related to a pair of interfering<br />

users). Its characterisation is the subject of the following <strong>de</strong>velopments.<br />

BPSK modulation Using (4.6) and (4.7), um u,DA can be split into its mean<br />

value U BPSK<br />

u,DA and the loop noise ξm u which is the sum of the additive noise<br />

and the self- and cross-noise.<br />

⎡<br />

⎢ e<br />

⎢<br />

⎢<br />

⎣<br />

j(φu ˆ φm u ) +<br />

n=<br />

+ Nu <br />

e<br />

k=1<br />

k=u<br />

j(φk ˆ φm u ) + Ek<br />

Eu<br />

n=<br />

Nu <br />

e j( ˆ φm k ˆ φm u ) + Ek<br />

Eu<br />

k=1<br />

k=u<br />

Im u In u xm n<br />

u,u<br />

<br />

e j ˆ φm u I m u νm <br />

u,DA<br />

n=<br />

Im u In n<br />

k<br />

xm<br />

u,k<br />

Im u In n<br />

k<br />

xm<br />

u,k<br />

⎤<br />

⎥<br />

⎦<br />

K BPSK<br />

D,u<br />

(4.11)<br />

<br />

sin φu ˆ φ m <br />

u .<br />

(4.12)<br />

The psd of the loop noise Sξu (z,∆) is given by the z-transform of its auto-


82 Data-Ai<strong>de</strong>d<br />

correlation function C m u,u (∆).<br />

Sξu<br />

(z,∆) =<br />

m=<br />

+<br />

C m u,u (∆) z m . (4.13)<br />

Both psd and auto-correlation function <strong>de</strong>pend on the phase estimation<br />

error ∆. However, in the following paragraphs, the <strong>de</strong>velopments will be<br />

limited to the study at equilibrium (∆ =0). The auto-correlation function<br />

of the loop noise is then obtained from the general expression given in<br />

Appendix A by s<strong>et</strong>ting u = v, n =0and ∆=0<br />

⎧ +<br />

⎫<br />

C m ⎪⎨<br />

u,u (0) = δ(m)<br />

⎪⎩<br />

p=<br />

[ (x p u,u)] 2<br />

+ N0x0 u,u<br />

2EuT<br />

+ Nu Ek<br />

Eu<br />

k=1 p=<br />

k=u<br />

Nu Ek<br />

Eu<br />

k=1 p=<br />

k=u<br />

+<br />

+<br />

2 ejδk,ux p<br />

u,k<br />

2 ejδk,ux p<br />

u,k<br />

⎪⎬<br />

⎪⎭<br />

x m 2 u,u .<br />

(4.14)<br />

The third and fourth terms of (4.14) are i<strong>de</strong>ntical, but their cancellation is<br />

only obtained in the case of the MU estimator. The SU estimator does not<br />

exhibit the fourth term. It can thus not compensate the effect of the MAI<br />

present in the matched filter output, which gives rise to the third term of<br />

(4.14).<br />

Introducing (4.14) in(4.13) leads to the expression of Sξu (z,0). For an SU<br />

estimator working in presence of MAI, this psd is ma<strong>de</strong> of three contributions:<br />

additive noise, self-, and cross-noise. They are shown in Figure<br />

4.2. The additive noise is the translation in the frequency domain of the<br />

second term of (4.14). It mainly <strong>de</strong>pends on the ratio Eu<br />

. The combination<br />

N0<br />

of the first and fifth terms gives birth to the self-noise which is shaped by<br />

the auto-correlation function of user u’s channel impulse response hu (t)<br />

(xm u,u factors). Finally, the cross-noise comes from the third term, due to<br />

MAI. The structure of this term is given by the cross-correlation function<br />

b<strong>et</strong>ween channel impulse responses of users u and k (xm u,k factors) with<br />

k = u. Notice however that there is no cross-noise contribution to be<br />

found in Sξu (z,0) in the case of an MU estimator, thanks to the cancellation<br />

mentioned here above. From (4.14), it appears that the additive noise


4.1 Feedback 83<br />

Power Spectral Density<br />

10 −2<br />

10 −3<br />

10 −4<br />

2−user system − 31−chip Gold co<strong>de</strong>s − BPSK modulation − TU channel − E s /N 0 = 20 dB<br />

Additive Noise<br />

Self−Noise<br />

Cross−Noise<br />

10<br />

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5<br />

−5<br />

f<br />

Figure 4.2: Power spectral <strong>de</strong>nsity of Additive Noise, Self- and Cross-<br />

Noise<br />

and the cross-noise contribute to the auto-correlation function only at zero<br />

time-shift (m =0). This leads to flat power spectral <strong>de</strong>nsities. On the other<br />

hand, the self-noise contributes to the whole auto-correlation function, up<br />

to the value of the normalised channel correlation coefficient x m u,u. In the<br />

frequency domain, the spectrum of the self-noise vanishes at f =0.[84,<br />

p. II-5] explains this high-pass behaviour as a result of the even symm<strong>et</strong>ry<br />

of the pulse at the matched filter output.<br />

QPSK modulation Similarly, um QP SK<br />

u,DA can be split into its expectation Uu,DA and the loop noise ξm u . As mentioned previously, the loop noise is at least<br />

the sum of the additive noise and the self-noise, but it might also inclu<strong>de</strong>


84 Data-Ai<strong>de</strong>d<br />

a cross-noise contribution when <strong>de</strong>aling with an SU estimator facing MAI.<br />

u m u,DA<br />

QP SK<br />

= K<br />

D,u<br />

⎪⎩<br />

<br />

sin φu ˆ φm <br />

u<br />

<br />

+ (Im u )⋆ νm <br />

u,DA<br />

⎧ ⎡<br />

⎢ e<br />

⎢<br />

⎪⎨ ⎢<br />

⎢<br />

+ ⎢<br />

⎣<br />

j(φu ˆ φm u ) +<br />

p=<br />

+ Nu <br />

<br />

Ek<br />

Eu<br />

k=1<br />

k=u<br />

ej(φk ˆ φm u ) +<br />

p=<br />

Nu <br />

<br />

Ek<br />

Eu ej( ˆ φm k<br />

K<br />

k=1<br />

k=u<br />

QP SK<br />

D,u<br />

sin<br />

<br />

φu ˆ φm <br />

u<br />

(Im u ) ⋆ I p m p<br />

uxu,u ˆ φm u ) +<br />

p=<br />

(Im u )⋆ I p p<br />

kxm u,k<br />

(Im u ) ⋆ I p p<br />

kxm u,k<br />

⎤ ⎫<br />

⎥ ⎪⎬ ⎥<br />

⎦<br />

⎪⎭<br />

QP SK<br />

Uu,DA Additive<br />

noise<br />

Self-<br />

+ Cross-<br />

Noise<br />

(4.15)<br />

Following the same procedure than with BPSK-modulated data symbols,<br />

the auto-correlation of the loop noise at equilibrium is <strong>de</strong>rived from (A.2).<br />

Cm u,u (∆ = 0) writes<br />

C m 1<br />

u,u (0) =<br />

2<br />

⎧<br />

⎡<br />

⎢<br />

⎪⎨<br />

⎢<br />

δ (m) ⎢<br />

⎣<br />

⎪⎩<br />

+<br />

p=<br />

x p u,u 2<br />

+ N0x0 u,u<br />

EuT<br />

+ Nu Ek<br />

Eu<br />

k=1 p=<br />

k=u<br />

Nu <br />

k=1<br />

k=u<br />

Ek<br />

Eu<br />

+<br />

+<br />

p=<br />

<br />

<br />

x p<br />

<br />

<br />

x p<br />

<br />

<br />

u,k<br />

2<br />

<br />

<br />

u,k<br />

2<br />

⎤<br />

⎥<br />

⎦<br />

<br />

<br />

x m <br />

u,u<br />

2<br />

⎫<br />

⎪⎬<br />

.<br />

⎪⎭<br />

(4.16)<br />

As far as the third and fourth terms are concerned, the same remark as in<br />

the previous paragraph can be ma<strong>de</strong> with respect to (4.16). The psd of the<br />

loop-noise is finally obtained by introducing (4.16) into (4.13).<br />

4.1.2 Closed-loop study<br />

In the tracking mo<strong>de</strong>, the estimate exhibits small fluctuations around the<br />

true value of the param<strong>et</strong>er. In the linear portion of the S-curve, these


4.1 Feedback 85<br />

small fluctuations translate into proportional fluctuations of the error signal.<br />

This enables us to <strong>de</strong>sign a linearised mo<strong>de</strong>l of the recovery loop at<br />

equilibrium. Using this linear mo<strong>de</strong>l, the jitter variance is obtained as the<br />

variance of the loop noise filtered by the linearised closed-loop transfer<br />

function.<br />

Linear mo<strong>de</strong>l of the recovery loop<br />

In the general case, the error signal u m l<br />

driving the loop is function of the<br />

phase estimation error ∆ through its mean Ul and through the loop noise.<br />

The working equation of the closed loop writes<br />

ˆφ m+1<br />

k = ˆ φ m k + K0,kFk(z) u m k ,k [1,Nu] (4.17)<br />

where K0,k and Fk(z) represent respectively the gain of NCO and the filter<br />

applied to the error signal u m k in the loop updating ˆ φk.<br />

In the linearised closed-loop <strong>de</strong>rived at equilibrium, Uk is replaced by the<br />

linear term of its Taylor-series expansion. This introduces a linear <strong>de</strong>pen<strong>de</strong>ncy<br />

of the error signal with respect to the phase estimation error whose<br />

proportionality coefficient is the slope of the S-curve at equilibrium:<br />

u m k = Uk + ξ m k = ∂Uk<br />

∂∆<br />

<br />

<br />

<br />

∆=0<br />

∆+ξ m k ,k [1,Nu] . (4.18)<br />

BPSK modulation U BPSK<br />

u,DA is replaced by its linear expansion around the<br />

operating point ∆=0<br />

⎡<br />

U<br />

⎢<br />

⎣<br />

BPSK<br />

1,DA<br />

U BPSK<br />

2,DA<br />

...<br />

U BPSK<br />

⎤<br />

⎥<br />

⎦<br />

Nu,DA<br />

⎡ <br />

∂UBP SK <br />

1,DA <br />

⎢ ∂∆1 <br />

⎢ ∆=0<br />

⎢ ∂UBP SK <br />

⎢ 2,DA <br />

= ⎢ ∂∆1 <br />

⎢<br />

∆=0<br />

⎢ ... <br />

⎣ ∂UBP SK <br />

<br />

<br />

∂UBP SK <br />

1,DA <br />

∂∆2 <br />

∆=0<br />

∂UBP SK <br />

2,DA <br />

∂∆2 <br />

∆=0<br />

... <br />

∂UBP SK <br />

<br />

...<br />

...<br />

...<br />

...<br />

<br />

∂UBP SK <br />

1,DA <br />

∂∆Nu <br />

∆=0<br />

∂UBP SK <br />

2,DA <br />

∂∆Nu <br />

∆=0<br />

... <br />

∂UBP SK <br />

Nu,DA <br />

⎤<br />

⎥ ⎡<br />

⎥ ∆1 ⎥ ⎢ ∆2 ⎥ ⎢<br />

⎥ ⎣<br />

⎥ ...<br />

⎥ ∆Nu ⎦<br />

⎤<br />

⎥<br />

⎦ .<br />

Nu,DA<br />

∂∆1<br />

∆=0<br />

Nu,DA<br />

∂∆2<br />

∆=0<br />

∂∆Nu<br />

∆=0<br />

(4.19)


86 Data-Ai<strong>de</strong>d<br />

In (4.19), the Nu ¢ Nu square matrix of the first <strong>de</strong>rivative of U BPSK<br />

Nu,DA is<br />

the Fisher information matrix, if not some multiplicative terms related to<br />

Es and the loop bandwidth. From that point of view, the off-diagonal<br />

N0<br />

elements characterise the coupling due to the MAI. However, examination<br />

of (4.7) reveals that they are null.<br />

⎡<br />

⎢<br />

⎣<br />

⎤<br />

⎥<br />

⎦ =<br />

⎡<br />

KD,1<br />

⎢ 0<br />

⎣ ...<br />

...<br />

KD,2<br />

...<br />

...<br />

...<br />

...<br />

0<br />

0<br />

...<br />

⎤ ⎡<br />

⎥ ⎢<br />

⎥ ⎢<br />

⎦ ⎣<br />

∆1<br />

∆2<br />

...<br />

⎤<br />

⎥<br />

⎦ . (4.20)<br />

0 0 ... KD,Nu ∆Nu<br />

U BPSK<br />

1,DA<br />

U BPSK<br />

2,DA<br />

...<br />

U BPSK<br />

Nu,DA<br />

There is thus no coupling b<strong>et</strong>ween recovery processes of different users.<br />

This is a benefit of the DA estimation process.<br />

Using (4.20), (4.17) becomes a s<strong>et</strong> of Nu equations of the type<br />

ˆφ m+1<br />

k = ˆ φ m k<br />

+ KkFk(z)<br />

<br />

φk ˆ φ m k<br />

<br />

+ K0,kFk(z) ξ m k ,k [1,Nu] (4.21)<br />

where Kk = KD,kK0,k is the loop gain. This is the equation of the loop<br />

shown at Figure 4.3. It illustrates the <strong>de</strong>coupling b<strong>et</strong>ween phase recovery<br />

φ m u<br />

φ m v<br />

+<br />

+<br />

ˆφ m u<br />

ˆφ m v<br />

-<br />

-<br />

∆ m u<br />

∆ m v<br />

KD,u<br />

0<br />

K0,u (z 1) 1<br />

KD,v<br />

0<br />

K0,v (z 1) 1<br />

ξ m u<br />

ξ m v<br />

Figure 4.3: DA BPSK PLL<br />

Fu (z)<br />

Fv (z)<br />

processes thanks to the knowledge of interfering users’ symbol sequences<br />

(DA context).


4.1 Feedback 87<br />

QP SK<br />

QPSK modulation Since U<br />

u,DA (4.10) is similar to U BPSK<br />

u,DA (4.7), the linearised<br />

mo<strong>de</strong>l built in the case of QPSK-modulated data symbols does<br />

not differ significantly from the one obtained with BPSK-modulated data<br />

symbols.<br />

Jitter variance<br />

It was mentioned earlier that the jitter variance σ 2 ˆ φu<br />

would serve as per-<br />

formance measure. Equation (4.21) enables to <strong>de</strong>rive it as the variance of<br />

the loop noise ξm u filtered by the closed loop [85]<br />

σ 2 ˆ φu =<br />

T<br />

2 Uu,DA∆=0 1<br />

2T<br />

<br />

1<br />

2T<br />

S ˆ φu<br />

<br />

e 2jπfT<br />

df (4.22)<br />

where S ˆ φu (z,∆) is the spectral <strong>de</strong>nsity of the filtered loop noise.<br />

<br />

<br />

Ku<br />

Sφu ˆ (z,∆) = <br />

z<br />

1 Fu,u(z)<br />

<br />

1<br />

Ku<br />

z 1 Fu,u(z)<br />

<br />

1 2<br />

Sξu (z,∆) . (4.23)<br />

Consi<strong>de</strong>ring a narrow noise bandwidth BN,u and a first-or<strong>de</strong>r loop, the<br />

phase jitter variance σ2 φu ˆ can be given by a Taylor-series expansion in the<br />

variable 2 BN,uT of the general variance expression (4.22) at equilibrium<br />

[84, p. II-4].<br />

σ 2 ˆ φu <br />

2 BN,uT<br />

Sξu<br />

∂Uu,DA 2<br />

∂∆u ∆=0<br />

(1, 0) 2<br />

(2 BN,uT ) 2<br />

<br />

<br />

+<br />

2 ∂Uu,DA<br />

m=<br />

∂∆u ∆=0<br />

m C m u,u (0) .<br />

(4.24)<br />

This expansion is limited to the second or<strong>de</strong>r. The quadratic term has to<br />

be taken into account due to the shape of the self-noise spectrum. Since it<br />

vanishes at f =0, it does not contribute to σ2 φu ˆ through the first term which<br />

is linear in BN,uT . The self-noise contribution to the variance comes thus<br />

from the term quadratic in σ2 φu ˆ . Figure 4.4 shows the jitter variance computed<br />

with and without self-noise in the case of BPSK-modulated data<br />

symbols. The importance of the quadratic term of (4.24) appears on the<br />

variance of the MU estimator. Its inci<strong>de</strong>nce on the SU estimator is less visible,<br />

the performance of this estimator being already <strong>de</strong>gra<strong>de</strong>d by MAI.


88 Data-Ai<strong>de</strong>d<br />

Regardless of the context, notice that the variance of the estimate is lowerand<br />

upper-boun<strong>de</strong>d. The lower bound on the variance is the CRLB, introduced<br />

in Section 3.3.1. Its upper-bound is the variance of an uniformlydistributed<br />

random variable whose span would be the same than the param<strong>et</strong>er<br />

un<strong>de</strong>r investigation, in the present case [0, 2π].<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

2−user system − 31−chip Gold co<strong>de</strong>s − BPSK modulation − TU channel<br />

Single−user<br />

Multiuser<br />

Without Self−Noise<br />

With Self−Noise<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 4.4: Inci<strong>de</strong>nce of the quadratic term of the Taylor-series expansion<br />

at equilibrium of the variance expression<br />

Using (4.24), the jitter variance can now be computed consi<strong>de</strong>ring successively<br />

BPSK- and QPSK-modulated data symbols. The computational<br />

results will be presented in the following figures. Each of these will show<br />

the jitter variance of a single phase estimate, namely the phase estimate of<br />

user 1, exhibited in a different scenario by SU and MU estimators. Such<br />

results have always been obtained by averaging the computations over<br />

1,000 iterations, each one corresponding to a specific snapshot scenario so<br />

as to g<strong>et</strong> results that are in<strong>de</strong>pen<strong>de</strong>nt of the param<strong>et</strong>ers of the scenario<br />

(co<strong>de</strong> sequences, channel responses, <strong>et</strong>c.).<br />

BPSK modulation Figure 4.5 illustrates the influence of the correlation<br />

properties of the co<strong>de</strong>s and of the load of the system in an AWGN channel,<br />

that is to say, in a situation where the MAI is the only interference. With


4.1 Feedback 89<br />

orthogonal Hadamard co<strong>de</strong>s, there is no MAI. The variance of both SU and<br />

MU estimators are thus equal. Moving to quasi-orthogonal Gold co<strong>de</strong>s,<br />

an irreducible variance floor appears on the SU curve. This floor rises<br />

along with the load. A similar <strong>de</strong>gradation has been shown in [46] which<br />

established the performance of an SU ML chip synchroniser in DS-CDMA<br />

communication systems. However, the MU estimator presented in the<br />

current work goes a step further, in that it mitigates the effect of the MAI<br />

so that the variance sticks to the CRLB even with quasi-orthogonal co<strong>de</strong>s<br />

or with a high system load.<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

Hadamard<br />

Gold<br />

Single−user<br />

Multiuser<br />

N = 2<br />

u<br />

N = 20<br />

u<br />

BPSK modulation − AWGN channel − 2 B N T = 0.1<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 4.5: Variance of DA FB estimators in AWGN channel (BPSK)<br />

Consi<strong>de</strong>ring a 2-user system over an AWGN channel, the variance curves<br />

of both SU and MU estimators are drawn in Figure 4.6 for different values<br />

of the Near-Far ratio. The performance of the SU estimator being <strong>de</strong>gra<strong>de</strong>d<br />

by the MAI, it is not surprising to see that the irreducible variance<br />

floor rises as the Near-Far ratio grows. On the other hand, the MU estimator<br />

benefits from the MAI mitigation. Its variance still reaches the CRLB,<br />

whichever Near-Far ratios are consi<strong>de</strong>red.<br />

Finally, the effect of the frequency selectivity of the channel is illustrated<br />

in Figure 4.7. The variances have been computed for two different baud


90 Data-Ai<strong>de</strong>d<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

2−user system − 31−chip Gold co<strong>de</strong>s − BPSK modulation − AWGN channel − 2 B N T = 0.1<br />

Single−user<br />

Multiuser<br />

Near−Far= 0 dB<br />

Near−Far= 3 dB<br />

Near−Far= 6 dB<br />

Near−Far= 9 dB<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 4.6: Near-Far effect on DA FB estimators (BPSK)<br />

rates in an Hilly Terrain (HT) channel 1 . Both SU and MU estimators suffer<br />

from ISI. The variance of the MU estimator is lower than the one of the SU<br />

because the latter also suffers from MAI. For both estimators, the lower<br />

the baud rate is, the longer the symbol becomes. Thus the inci<strong>de</strong>nce of the<br />

ISI is also lower. In<strong>de</strong>ed, neither the SU nor the MU estimator have been<br />

<strong>de</strong>signed to face ISI. While in <strong>de</strong>tection studies ISI and MAI are <strong>de</strong>alt with<br />

simultaneously by regarding MAI as a time-varying version of ISI [38], the<br />

phase estimation structure handles them separately. Inspection of (3.31)<br />

reveals that the ISI influence vanishes when taking the first <strong>de</strong>rivative of<br />

the log-likelihood function with respect to the phase param<strong>et</strong>er since it<br />

does not <strong>de</strong>pend on this param<strong>et</strong>er. The inci<strong>de</strong>nce of ISI on recovery loops<br />

has been studied in [113, 114]. Estimation structures taking into account<br />

ISI have been presented in [113, 115]. Without such <strong>de</strong>sign, computing the<br />

phase jitter variance in dispersive environments makes it clear that, while<br />

the SU estimator is the only one affected by MAI, both suffer from ISI so<br />

that none of them ever reaches CRLB at high Es<br />

N0 .<br />

1 See Appendix G


4.1 Feedback 91<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

2−user system − 31−chip Gold co<strong>de</strong>s − BPSK modulation − HT channel − 2 B N T = 0.1<br />

Single−user<br />

Multiuser<br />

R = 1e4 Bauds<br />

R = 1e5 Bauds<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 4.7: Variance of DA FB estimators in dispersive channels (BSPK)<br />

QPSK modulation Following the same procedure as in the BPSK case,<br />

(4.16) can be used to <strong>de</strong>rive first the psd of the loop noise (4.13) and then,<br />

the jitter variance (4.24) of the phase recovery loop operating on QPSKmodulated<br />

symbols.<br />

The conclusions drawn in the previous paragraph with BPSK-modulated<br />

data symbols are still valid using QPSK modulation. In an AWGN channel,<br />

the MU estimator has a variance which reaches the CRLB, with orthogonal<br />

as well as with quasi-orthogonal co<strong>de</strong>s, irrespective of the load<br />

enabled by the resolution of the co<strong>de</strong> thanks to the MAI mitigation (Figure<br />

4.8). On the other hand, the SU estimator exhibits an irreducible variance<br />

floor as soon as the orthogonality of the co<strong>de</strong>s is lost. The level of this<br />

floor <strong>de</strong>pends on the load of the system. Facing a Near-Far effect (Figure<br />

4.9), the MU estimator appears to be Near-Far resistant, while the variance<br />

floor limiting the performance of the SU estimator rises with the Near-Far<br />

ratio. Finally, in dispersive channels, the fact that neither the SU nor the<br />

MU estimators take into account the inci<strong>de</strong>nce of ISI leads to an irreducible<br />

variance floor on both of them which <strong>de</strong>pends on the level of ISI<br />

<strong>de</strong>grading their performance.


92 Data-Ai<strong>de</strong>d<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

Hadamard<br />

Gold<br />

Single−user<br />

Multiuser<br />

N u = 2<br />

N u = 20<br />

QPSK modulation − AWGN channel − 2 B N T = 0.1<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 4.8: Variance of DA FB estimators in AWGN channel (QPSK)<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

2−user system − 31−chip Gold co<strong>de</strong>s − QPSK modulation − AWGN channel − 2 B N T = 0.1<br />

Single−user<br />

Multiuser<br />

Near−Far= 0 dB<br />

Near−Far= 3 dB<br />

Near−Far= 6 dB<br />

Near−Far= 9 dB<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 4.9: Near-Far effect on DA FB estimators (QPSK)


4.1 Feedback 93<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

2−user system − 31−chip Gold co<strong>de</strong>s − QPSK modulation − HT channel − 2 B N T = 0.1<br />

Single−user<br />

Multiuser<br />

R = 1e4 Bauds<br />

R = 1e5 Bauds<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 4.10: Variance of DA FB estimators in dispersive channels (QSPK)


94 Data-Ai<strong>de</strong>d<br />

4.2 Feedforward<br />

Besi<strong>de</strong> the FB estimator, solving (4.1) can also lead to the FF estimator<br />

given by relation (3.52)<br />

where<br />

Cu =<br />

m=1<br />

N<br />

(I m u )⋆ y m u<br />

ˆφu = tan<br />

Nu <br />

Ek<br />

k=1<br />

k=u<br />

1 (Cu)<br />

(Cu) = arg (Cu) (4.25)<br />

Eu<br />

e j ˆ φk<br />

N<br />

+<br />

m=1 n=<br />

(I m u )⋆ I n n<br />

k xm<br />

u,k<br />

(4.26)<br />

as already mentioned in Section 3.4.1. This is the expression of the MU<br />

DA ML FF phase estimator. Before studying the performance of its simplified<br />

version (3.59) obtained by linearising the complex exponential, the<br />

<strong>de</strong>gradation of the performance of the SU estimator working in a multiuser<br />

context will be established in the next section.<br />

4.2.1 Pdf of an SU estimator in a multiuser context<br />

The SU DA ML FF phase estimator is given by [7, p. 326]<br />

<br />

N<br />

ˆφu = arg (I m u )⋆ y m <br />

u . (4.27)<br />

m=1<br />

It is the same kind of estimator (argument of a complex number) as the MU<br />

one (4.26) but it lacks the MAI mitigation term. It should thus be stressed<br />

that the SU estimator (4.27) is not the optimal one for either frequencyselective<br />

and/or multiuser contexts due to the presence of interference,<br />

ISI, and/or MAI. In<strong>de</strong>ed, this estimator has been <strong>de</strong>rived from a log-likelihood<br />

function which does not take into account any interference at all. As<br />

a result, its performance will be <strong>de</strong>gra<strong>de</strong>d by interference. The purpose of<br />

the following calculations is to quantify this <strong>de</strong>gradation on the variance<br />

of the estimation error ∆u = φu ˆ φu. The variance will be calculated from<br />

the pdf of ∆u.<br />

Analytical <strong>de</strong>rivation of the pdf<br />

To <strong>de</strong>rive the pdf of the phase estimation error ∆u, the complex number<br />

whose argument is ∆u is consi<strong>de</strong>red:


4.2 Feedforward 95<br />

ˆrue j∆u<br />

= ˆxu + j ˆyu (4.28)<br />

=<br />

N +<br />

(Im u )⋆ In u xm<br />

n<br />

u,u<br />

<br />

Ek<br />

Eu<br />

Useful term<br />

+ ISI<br />

ej(φk φu) N +<br />

(Im u ) ⋆ In n<br />

k<br />

xm<br />

u,k MAI<br />

m=1 n=<br />

+ Nu <br />

k=1<br />

k=u<br />

+ e jφu<br />

N<br />

m=1<br />

(I m u )⋆ ν m u,DA<br />

m=1 n=<br />

Additive noise<br />

(4.29)<br />

The characteristic function ψˆxu,ˆyu (ωr,ωi) can be calculated, so that its inverse<br />

Fourier transform gives the joint pdf T ˆxu,ˆyu (ˆxu, ˆyu). Then, a change<br />

of variable from cartesian (x, y) to polar (r, ∆) coordinates and an integration<br />

of Tˆru,∆u (ˆru, ∆u) over the range of ru yields the pdf of the phase<br />

estimation error ∆u. The calculation in the case of BPSK-modulated data<br />

symbols, using a single-tap averaging window (N =1), is <strong>de</strong>tailed in Appendix<br />

B. The pdf finally writes (B.18)<br />

T∆u(∆u)<br />

1<br />

=<br />

2 (NuSx) π⎧<br />

2 (NuSx 2)<br />

<br />

k=1<br />

⎪⎨<br />

⎪⎩<br />

<br />

exp<br />

<br />

1<br />

+<br />

with Sx, cu, f ¦ u and g¦ u<br />

Computational results<br />

<br />

exp<br />

<br />

1+<br />

<br />

g u<br />

4cu<br />

<br />

π f u<br />

cu 2 exp<br />

2 <br />

(f u ) f u 1 erf<br />

4cu<br />

2 Ô <br />

cu<br />

<br />

<br />

g + u<br />

4cu<br />

<br />

π f<br />

cu<br />

+ u<br />

2 exp<br />

<br />

+ 2<br />

(f u )<br />

4cu<br />

<br />

1 erf<br />

<strong>de</strong>fined in (B.5), (B.9) and (B.14-B.17).<br />

f + u<br />

2 Ô <br />

cu<br />

<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

(4.30)<br />

Relation (4.30) has been computed in different scenarii. They might appear<br />

simplistic but this is to avoid the computational complexity problem<br />

mentioned in Section B.2. For every scenario, the pdf has been averaged


96 Data-Ai<strong>de</strong>d<br />

over 1,000 computations, each one being characterised by a specific random<br />

choice of the users’ phases and co<strong>de</strong>s, and of the channel impulse<br />

responses. This strategy has been chosen in or<strong>de</strong>r to produce results not<br />

sensitive to the choice of a s<strong>et</strong> of simulation param<strong>et</strong>ers. Such an averaging<br />

operation is avoi<strong>de</strong>d in [96] at the price of the use of stochastic mo<strong>de</strong>ls for<br />

the interference.<br />

Figure 4.11 presents the pdf obtained in a 2-user system using 7-chip Gold<br />

co<strong>de</strong>s in channels whose <strong>de</strong>lay profiles are given by COST 207 Rural Area<br />

(RA) mo<strong>de</strong>l 2 . Thanks to the DA structure, even an SU ML estimator remains<br />

unbiased in interfering contexts. However, this interference is a<br />

source of <strong>de</strong>gradation. This appears when looking at the variance of the<br />

estimator. Obviously, with respect to the dotted pdf which is obtained in<br />

T ( Δ )<br />

Δ1 1<br />

1.5<br />

1<br />

0.5<br />

2−user system − 7−chip Gold co<strong>de</strong>s − R = 1e5 Bauds − N = 1<br />

AWGN, N u = 1<br />

RA, N u = 2<br />

E s /N 0 = 0 dB<br />

E s /N 0 = 4 dB<br />

E s /N 0 = 8 dB<br />

0<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

1<br />

1 2 3 4<br />

Figure 4.11: Pdf of the SU DA ML FF phase estimate in a 2-user, RA channel<br />

context<br />

an SU scenario with AWGN channel, the consi<strong>de</strong>red pdf exhibits a larger<br />

variance as a result of two interfering effects. First, the use of quasiorthogonal<br />

co<strong>de</strong>s introduces MAI. Second, the fact that the channel is<br />

frequency-selective causes ISI to also <strong>de</strong>gra<strong>de</strong> the performance of the estimator.<br />

2 See Appendix G


4.2 Feedforward 97<br />

These influences are illustrated more clearly in Figure 4.12 which shows<br />

the variance of an SU DA ML FF estimator as a function of Es in several<br />

N0<br />

contexts. Due to the heavy computations it would have led to, the variance<br />

has not been computed from (4.30), but rather measured from the pdf <strong>de</strong>rived<br />

from this equation and illustrated in Figure 4.11. The shortcomings<br />

of this m<strong>et</strong>hod appear in Figure 4.12 at low Es<br />

ratios, where the variance<br />

N0<br />

does not exactly match the CRLB. Nevertheless, the rea<strong>de</strong>r distinguishes<br />

the inci<strong>de</strong>nce of MAI and of ISI on the variance curves obtained in 1- and<br />

2-user systems consi<strong>de</strong>ring either AWGN or HT channel.<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

2−user system − 7−chip Gold co<strong>de</strong>s − BPSK modulation − R = 1e5 Bauds − N = 1<br />

AWGN<br />

HT<br />

N u = 1<br />

N u = 2<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30<br />

−4<br />

E /N [dB]<br />

s 0<br />

Figure 4.12: Variances of the SU DA ML FF phase estimation error as a<br />

function of the number of user Nu and of the channel type<br />

In a situation where there is neither MAI (Nu =1) nor ISI (AWGN channel),<br />

the variance of the pdf shown in Figure 4.12 equals the CRLB (but<br />

at low Es<br />

ratios, as explained here above). However, due to the interfe-<br />

N0<br />

rence, either MAI or ISI, a variance floor appears which is not <strong>de</strong>pending<br />

on Es<br />

, causing the variance curve to rise away from the CRLB. The level of<br />

N0<br />

this floor <strong>de</strong>pends on how much interference enters the system and thus<br />

on the interfering conditions. The more users there are, the more MAI<br />

plagues the system. The more dispersive the channel is, the more ISI there<br />

is.<br />

MAI<br />

ISI


98 Data-Ai<strong>de</strong>d<br />

Another aspect should be taken into account, namely the Near-Far effect.<br />

The study of its influence on the variance of the DA ML FF phase estimator<br />

is <strong>de</strong>ferred to the next section, where the variance of the pdf computed in<br />

the present section will serve as a benchmark for the variance <strong>de</strong>rived from<br />

the closed-form of the estimate.<br />

4.2.2 Linearised multiuser estimator in 2-user system<br />

The previous section was concerned with the <strong>de</strong>rivation of the pdf of an<br />

SU DA ML FF phase estimator in a multiuser context. It was shown that<br />

the performance were not optimal due to the fact that the inci<strong>de</strong>nce of<br />

interference was disregar<strong>de</strong>d. The present section will take MAI into account.<br />

In Section 3.4.1, a modified version of the ML FF phase estimator given<br />

by equation (3.52) has been introduced after linearisation of the complex<br />

exponentials. The motivation of this linearisation is to ease the following<br />

<strong>de</strong>rivation of a closed-form expression of the DA ML FF phase estimator<br />

which is suitable for performance evaluation.<br />

Closed-form expression of the phase estimation error<br />

Multiuser case In or<strong>de</strong>r to evaluate the performance of the MU DA ML<br />

FF phase estimator, the matched filter outputs y m u are expan<strong>de</strong>d in the analytical<br />

expression (3.59) of the <strong>de</strong>rived estimator. Limiting the <strong>de</strong>velopments<br />

to a 2-user case, Du becomes<br />

Du<br />

= ejφu ⎡<br />

⎣ N<br />

Im u 2 x0 u,u + N<br />

m=1<br />

m=1<br />

<br />

Ev<br />

+j ∆v Eu ejφv<br />

N +<br />

m=1 n=<br />

+ N<br />

m=1<br />

(I m u ) ⋆ ν m u,DA<br />

+<br />

n=<br />

n=m<br />

(I m u ) ⋆ I n u x<br />

(Im u ) ⋆ In v xm n<br />

u,v<br />

m n<br />

u,u<br />

⎤<br />

⎦<br />

Useful term<br />

+ ISI<br />

MAI<br />

Additive noise<br />

(4.31)<br />

The MAI introduced by the matched filter output is partly cancelled by the<br />

mitigation term. Only a small MAI contribution is left in (4.31), weighted


4.2 Feedforward 99<br />

by j∆v. This weighting term reduces the influence of the MAI proportionally<br />

to the reduction of the phase estimation error on user v. In the case<br />

of an SU estimator there is no such weighting. In this case, the MAI fully<br />

disturbs the estimation process.<br />

Using (4.31) in(3.59) enables to solve the system for the phase estimation<br />

error. Willing to reach a closed form solution, MAI and noise contributions<br />

in the <strong>de</strong>nominator of (3.59) are regar<strong>de</strong>d as negligible with respect to the<br />

direct (x0 u,u) and ISI (xm n<br />

u,u ,m = n and n [ , + ]) terms. Furthermore,<br />

noticing that direct terms are real, these contributions vanish from<br />

the numerator of (3.59). The MU phase estimation error finally writes<br />

∆u <br />

(ISIu + Noiseu) (Directv + ISIv)<br />

(MAIv,u) (ISIv + Noisev)<br />

= (4.32)<br />

(Directu + ISIu) (Directv + ISIv)<br />

where<br />

MAIu,v =<br />

(MAIv,u) (MAIu,v)<br />

Directu =<br />

ISIu =<br />

Ev<br />

Eu<br />

N<br />

m=1<br />

N<br />

I m u 2 x 0 u,u<br />

m=1<br />

+<br />

n=<br />

n=m<br />

j(φv φu)<br />

e<br />

Noiseu = e jφu<br />

(I m u )⋆ I n n<br />

u xm u,u<br />

N<br />

+<br />

m=1 n=<br />

N<br />

m=1<br />

(I m u )⋆ I n n<br />

v xm u,v<br />

(4.33)<br />

(4.34)<br />

(4.35)<br />

(I m u )⋆ ν m u,DA . (4.36)<br />

Single-user case In the same context with the same hypotheses, the SU<br />

phase estimation error writes<br />

∆u = (MAIu,v + ISIu + Noiseu)<br />

. (4.37)<br />

(Directu + ISIu)


100 Data-Ai<strong>de</strong>d<br />

Mean<br />

Comparing (4.32) and (4.37), it appears that in a noiseless (Noisek 0 k)<br />

and non dispersive (ISIk 0 k) environment, that is to say in a situation<br />

where MAI is the only interference, the numerator of (4.32) is driven<br />

to zero. By itself, without any more processing, the estimation error of the<br />

MU estimator is small. On the contrary, the corresponding performance of<br />

the SU estimator is plagued by an MAI contribution which requires filtering<br />

(averaging) to disappear. However, if the channel becomes dispersive<br />

(ISIk = 0 k), both estimators exhibit a sensitivity to the ISI contribution.<br />

Nevertheless, it is clear from (4.34) that its inci<strong>de</strong>nce can be reduced<br />

by appropriate filtering (enlarging the width N of the averaging window).<br />

From a statistical point of view, an approximation of the expectations of<br />

these quotients can be performed if the <strong>de</strong>nominator is substituted by its<br />

mathematical expectation, as suggested in Section 3.4.2. Then, as already<br />

shown in Figure 4.11, both SU and MU estimators are unbiased since the<br />

expectation operator s<strong>et</strong> their numerators equal to zero thanks to the in<strong>de</strong>pen<strong>de</strong>nce<br />

b<strong>et</strong>ween data symbols and noise, and b<strong>et</strong>ween data symbols<br />

from different users (MAI issue) or from the same user but taken at different<br />

time instants (ISI issue).<br />

Variance<br />

The expressions of the variance of several DA ML FF estimators are given<br />

in Appendix C, <strong>de</strong>pending on the structure of the estimator (MU vs SU)<br />

and on the modulation of the data symbols (BPSK vs QPSK).<br />

The rea<strong>de</strong>r can notice that all variances present a contribution linear in Es<br />

N0<br />

and a contribution in<strong>de</strong>pen<strong>de</strong>nt of this ratio. The former comes from signal<br />

¢ noise terms while the latter is produced by signal ¢ signal terms [84,<br />

p. II-3]. However, the origin of the latter differs according to the type of<br />

estimator.<br />

In the case of MU estimators, variance expressions (C.4 and C.8) involve<br />

a term linear in Es<br />

, which is not surprising for DA estimators relying on<br />

N0<br />

a perfect knowledge of the channel behaviour [105, 106]. Besi<strong>de</strong> this term<br />

comes a contribution not <strong>de</strong>pending on the Es ratio. Its presence is due to<br />

N0<br />

the dispersiveness of the channel since it reflects the inci<strong>de</strong>nce of ISI. Interestingly,<br />

its structure, <strong>de</strong>tailed in (C.6) for the BPSK case, shows that the


4.2 Feedforward 101<br />

variance contribution due to ISI is a function of the mismatching b<strong>et</strong>ween<br />

the total spread of coefficients xn m<br />

u,u and the width N of the averaging<br />

window:<br />

⎛<br />

<br />

N<br />

BPSK 2 ⎜<br />

σISIu = f ⎝<br />

+ <br />

n m<br />

x <br />

u,u<br />

2<br />

N N <br />

n m<br />

x <br />

u,u<br />

2<br />

⎞<br />

⎟<br />

⎠ (4.38)<br />

m=1<br />

n=<br />

n=m<br />

m=1<br />

n=1<br />

n=m<br />

When this window becomes wi<strong>de</strong>r, the second term of (4.38) mitigates the<br />

first one. The ISI contribution to the variance then reduces, asymptotically<br />

becoming null when N + . It will be shown in the following computational<br />

results that the ISI contribution leads to an irreducible variance<br />

floor plaguing MU estimators in the case of dispersive channels.<br />

On the other hand, besi<strong>de</strong> the contribution linearly <strong>de</strong>pen<strong>de</strong>nt on the Es<br />

N0<br />

ratio, variances of SU estimators (C.12 and C.13) also exhibit an irreducible<br />

variance floor. However, contrary to the case of MU estimators, it results<br />

not only from ISI but also from MAI. Analytically, the inci<strong>de</strong>nce of the<br />

MAI appears through the third term of the SU variance expressions (C.12)<br />

and (C.13). Graphically, its effect is illustrated in Figure 4.13. Since these<br />

curves are obtained in an AWGN channel, there is no ISI. As a result, the<br />

variance floor, which obviously only appears on the variance curves of the<br />

SU estimator, is due to MAI. Figure 4.13 shows that the variance floor is<br />

reduced by improving the orthogonality properties of the co<strong>de</strong>s, moving<br />

from 7-chip to 31-chip quasi-orthogonal Gold co<strong>de</strong>s, then to 8-chip orthogonal<br />

Hadamard co<strong>de</strong>s. In<strong>de</strong>ed, choosing b<strong>et</strong>ter co<strong>de</strong>s reduces the MAI,<br />

and thus leads to lower variance floors. Of course, another way to reduce<br />

the variance is to enlarge the span N of the averaging window, as illustrated<br />

in Figure 4.13 where a window of size N =10is used instead of<br />

one of size N =1.<br />

A perfect power control scenario has been assumed so far. Consi<strong>de</strong>ring<br />

now power imbalance b<strong>et</strong>ween users, the behaviour of the estimators in<br />

the presence of a Near-Far effect is clear from both expressions <strong>de</strong>tailed in<br />

Appendix C and from Figure 4.14. In a non-dispersive environment, relations<br />

(C.4) and (C.8) are not explicitly <strong>de</strong>pen<strong>de</strong>nt of power ratios b<strong>et</strong>ween<br />

users, which indicates Near-Far resistance. Figure 4.14 confirms this statement:<br />

the variance curves of the MU estimator are superimposed, showing<br />

no sensitivity to the Near-Far ratio. On the other hand, the SU estimator<br />

appears to be sensitive to the level of power imbalance, as its variance


102 Data-Ai<strong>de</strong>d<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

2−user system − BPSK modulation − R = 1e5 Bauds − AWGN channel<br />

Single−user<br />

Multiuser<br />

N = 1<br />

N = 10<br />

8−chip Hadamard<br />

7−chip Gold<br />

31−chip Gold<br />

MAI reduction<br />

Uniform distribution<br />

MAI reduction<br />

CRLB, N=1<br />

CRLB, N=10<br />

10<br />

0 5 10 15 20 25 30<br />

−5<br />

E /N [dB]<br />

s 0<br />

Figure 4.13: Variance of DA FF estimators in an AWGN channel<br />

floor rises when the Near-Far ratio increases.<br />

Figure 4.14 is also an opportunity to check the match b<strong>et</strong>ween the variance<br />

of the SU estimator calculated from the pdf of Section 4.2.1 and<br />

the expressions shown in Appendix C. There is a close match in a perfect<br />

power control scenario (Near-Far ratio = 0 dB). When the Near-Far ratio<br />

increases, the variance <strong>de</strong>rived from linearisation slightly un<strong>de</strong>restimates<br />

the true variance given by the pdf. This un<strong>de</strong>restimation might be due to<br />

the simplification performed in or<strong>de</strong>r to reach closed-form expressions of<br />

the estimation error (MAI neglected in front of direct contributions in the<br />

<strong>de</strong>nominator of (3.59)).<br />

Finally, the inci<strong>de</strong>nce of dispersive channels is illustrated in Figure 4.15. It<br />

shows the results obtained in an i<strong>de</strong>al AWGN channel and in a frequencyselective<br />

HT channel. As far as the MU estimator is concerned, the rea<strong>de</strong>r<br />

notices the expected variance floor due to ISI. Consi<strong>de</strong>ring now the SU<br />

estimator, it appears that its variance floor encompasses both MAI and ISI<br />

effects. In<strong>de</strong>ed, the variance floor already plagues the SU estimator in the<br />

AWGN channel, due to MAI. In the case of the dispersive channel, ISI adds<br />

upon MAI and raises the variance floor. Finally, a slight un<strong>de</strong>restimation


4.3 Feedback-Feedforward correspon<strong>de</strong>nce 103<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

2−user system − 7−chip Gold co<strong>de</strong>s − BPSK modulation − R = 1e4 Bauds − AWGN channel − N = 1<br />

pdf<br />

FF<br />

Single−user<br />

Multiuser<br />

Near−Far = 0 dB<br />

Near−Far = 3 dB<br />

Near−Far = 6 dB<br />

Near−Far = 9 dB<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30<br />

−4<br />

E /N [dB]<br />

s 0<br />

Figure 4.14: Near-Far effect on DA FF estimators<br />

of the variance is noticed again comparing lienarised and true curves.<br />

Similar curves can be drawn in the case of QPSK-modulated data symbols.<br />

Since they do not bring significant new conclusions, and since it would not<br />

have been possible to check them with results from Section 4.2.1 for the<br />

pdf has been <strong>de</strong>rived in the case of BPSK-modulated data symbols, they<br />

are not shown here.<br />

4.3 Feedback-Feedforward correspon<strong>de</strong>nce<br />

Having <strong>de</strong>rived analytical expressions for the variance of several DA estimators,<br />

it is worthwhile to check the performance correspon<strong>de</strong>nce b<strong>et</strong>ween<br />

FB and FF implementations. It is well known that they are related<br />

in as much as their bandwidth correspond [85, p. 349]. This is illustrated<br />

by their respective CRLB expressions (3.50) and (3.48) from which the correspon<strong>de</strong>nce<br />

condition is <strong>de</strong>rived :<br />

CRLBu = BN,uT<br />

1<br />

Es,u<br />

N0<br />

= 1<br />

2N<br />

1<br />

Es,u<br />

N0<br />

2 BN,uT = 1<br />

. (4.39)<br />

N


104 Data-Ai<strong>de</strong>d<br />

Variance [rad 2 ]<br />

10 1<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

2−user system − 7−chip Gold co<strong>de</strong>s − R = 1e5 Bauds − BPSK modulation − N = 1<br />

pdf<br />

FF<br />

Single−user<br />

Multiuser<br />

AWGN<br />

HT<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30<br />

−4<br />

E /N [dB]<br />

s 0<br />

Figure 4.15: Inci<strong>de</strong>nce of ISI on DA FF estimators<br />

The correspon<strong>de</strong>nce can be checked b<strong>et</strong>ween the pairs of FB/FF estimators<br />

<strong>de</strong>rived in this chapter, either SU or MU, for either BPSK- or QPSKmodulated<br />

data symbols. This study was performed un<strong>de</strong>r the hypothesis<br />

of small loop bandwidth (2 BN,uT 0, N + ).<br />

In the case of FB estimators, (4.24) then reduces to<br />

where<br />

<br />

∂Uu,DA 2<br />

∂∆u ∆=0<br />

<br />

σ 2 ˆ φu <br />

2 BN,uT<br />

Sξu<br />

∂Uu,DA 2<br />

∂∆u ∆=0<br />

(1, 0) (4.40)<br />

= KD,u = σ2 Iux0u,u and Sξu (1, 0) is given by (4.13)<br />

Sξu<br />

(1, 0) =<br />

m=<br />

+<br />

C m u,u (0) (4.41)<br />

with C m u,u (0) written as (4.14) or(4.16) <strong>de</strong>pending on the modulation.<br />

Thanks to the infinite sum in (4.41) the ISI contribution from (4.14) and<br />

(4.16) vanishes. Only noise and MAI terms are thus left in the simplified<br />

expressions listed in Table 4.1. These expressions are <strong>de</strong>rived assuming a<br />

2-user case in or<strong>de</strong>r to enable comparison with the relations obtained in


4.4 Conclusions 105<br />

Section 4.2 for FB estimators in such a scenario.<br />

Looking now at FF variance expressions in the case of small loop bandwidths<br />

(N + ), ISI contributions asymptotically disappear in relations<br />

presented in Appendix C, as explained in Section 4.2.2. Among the<br />

remaining terms, only the highest powers of N will be consi<strong>de</strong>red. Using<br />

the fact that<br />

<br />

n m<br />

x <br />

u,v<br />

2 <br />

<br />

= e jδv,u <br />

n m<br />

xu,v 2<br />

<br />

= e jδv,u 2 <br />

n m<br />

xu,v + e jδv,u 2 n m<br />

xu,v = e jδv,u <br />

2 <br />

n m<br />

x +2 e jδv,u 2 n m<br />

x (4.42)<br />

u,v<br />

simplified expressions are obtained. They are gathered in Table 4.1. Applying<br />

the bandwidth equivalence expression (4.39), there is in<strong>de</strong>ed correspon<strong>de</strong>nce<br />

b<strong>et</strong>ween FB and FF relations. Moreover, the rea<strong>de</strong>r can verify<br />

that, in the absence of MAI, SU estimators exhibit a MAI contribution to<br />

their variance, while the variance of MU estimators only <strong>de</strong>pend on the<br />

Es<br />

N0 ratio.<br />

A graphical validation of the correspon<strong>de</strong>nce b<strong>et</strong>ween FB and FF estimators<br />

is provi<strong>de</strong>d in Figure 4.16. It is a remin<strong>de</strong>r of Figure 4.14, compl<strong>et</strong>ed<br />

with the variances of FB estimators operating in the same scenario. Figure<br />

4.16 shows a close matching b<strong>et</strong>ween curves of FB and FF estimators.<br />

4.4 Conclusions<br />

Two different implementations of DA estimators, namely FB and FF, have<br />

been consi<strong>de</strong>red in this chapter. Their performance in terms of jitter variance<br />

have been <strong>de</strong>rived analytically for MU as well as for SU estimators<br />

and computed in several scenarii. These results have been cross-checked<br />

by comparing asymptotical FB and FF variance expressions. The latter<br />

expression has also been compared to the variance computed using the<br />

analytical expression of the pdf of the SU estimate.<br />

In FB implementations, the main advantage of DA estimation has appeared<br />

to be the <strong>de</strong>coupling b<strong>et</strong>ween recovery loops. From the point of<br />

view of the jitter variance, conclusions were the same for FB as well as for<br />

u,v


106 Data-Ai<strong>de</strong>d<br />

BPSK SU FB (4.24) with C m u,u (0)<br />

given by (4.14)<br />

Original expression Asymptotical expression<br />

FF (C.12) 1<br />

2 N<br />

MU FB (4.24) with C m u,u (0)<br />

given by (4.14)<br />

FF (C.4) 1<br />

2 N<br />

QPSK SU FB (4.24) with C m u,u (0)<br />

given by (4.16)<br />

FF (C.13) 1<br />

2 N<br />

MU FB (4.24) with C m u,u (0)<br />

given by (4.16)<br />

FF (C.8) 1<br />

2N<br />

1<br />

Es,u<br />

BN,uT<br />

N0<br />

1<br />

Es,u<br />

N0<br />

1<br />

Es,u<br />

BN,uT<br />

N0<br />

1<br />

Es,u<br />

N0<br />

1<br />

Es,u<br />

BN,uT<br />

N0<br />

1<br />

Es,u<br />

N0<br />

1<br />

Es,u<br />

BN,uT<br />

N0<br />

1<br />

Es,u<br />

N0<br />

+ 2 BN,uT<br />

(x0 u,u) 2<br />

Ev<br />

Eu<br />

+<br />

+<br />

p=<br />

+<br />

1<br />

N(x0 u,u) 2<br />

Ev<br />

Eu<br />

p=<br />

+ BN,uT<br />

(x0 u,u) 2<br />

Ev<br />

Eu<br />

+<br />

p=<br />

+<br />

+<br />

1<br />

2 N(x0 u,u) 2<br />

Ev<br />

Eu<br />

p=<br />

<br />

jδv,u p 2 e xu,v <br />

jδv,u p 2 e xu,v x p u,v 2<br />

x p u,v 2<br />

Table 4.1: Asymptotical variance expressions of DA estimators in a 2-user case


4.4 Conclusions 107<br />

Variance [rad 2 ]<br />

10 1<br />

2−user system − 7−chip Gold co<strong>de</strong>s − BPSK modulation − R = 1e4 Bauds − AWGN channel − 2 B T = N = 1<br />

N<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

FB<br />

FF<br />

pdf<br />

Single−user<br />

Multiuser<br />

Near−Far = 0 dB<br />

Near−Far = 3 dB<br />

Near−Far = 6 dB<br />

Near−Far = 9 dB<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30<br />

−4<br />

E /N [dB]<br />

s 0<br />

Figure 4.16: Correspon<strong>de</strong>nce b<strong>et</strong>ween DA FB and FF estimators<br />

FF estimators: the MU estimator is not affected by MAI while its SU counterpart<br />

exhibits a variance floor <strong>de</strong>pending on the level of MAI entering<br />

the system. However, a similar variance floor limits the performance of<br />

both estimators in dispersive environments as a result of ISI.<br />

The estimators studied in this chapter rely on the knowledge of the transmitted<br />

symbols. The next chapter will <strong>de</strong>al with structures using the fedback<br />

<strong>de</strong>cisions instead of the true symbols.


Chapter 5<br />

Decision Directed<br />

Similarly to what was done in the previous chapter, the first <strong>de</strong>rivative of<br />

the log-likelihood function (3.31) with respect to the phase param<strong>et</strong>er is<br />

used to <strong>de</strong>rive the ML estimator. In the DD context, it writes<br />

<br />

∂ΛL(Φ) <br />

<br />

∂φu<br />

<br />

Φ= Φˆ<br />

= 2EuT<br />

⎡<br />

⎢<br />

e<br />

⎢<br />

⎢<br />

N0 ⎣<br />

j ˆ N ⋆ φu Îm u y<br />

m=1<br />

m u<br />

Nu <br />

<br />

Ek<br />

Eu ej( ˆ φk ˆ φu) N<br />

⎤<br />

⎥<br />

+ <br />

⎥<br />

⋆Î ⎥<br />

Îm n m n<br />

u k x ⎦<br />

u,k<br />

.<br />

k=1<br />

k=u<br />

m=1 n=<br />

(5.1)<br />

The main difference b<strong>et</strong>ween (5.1) and its DA counterpart (4.1) is the use of<br />

Îm u instead of Im u . While, in DA structures, the data information used in the<br />

estimation process are obtained through the transmission of pre<strong>de</strong>fined<br />

training sequences, DD phase estimators g<strong>et</strong> this information from the <strong>de</strong>cision<br />

stage of the receiver. Thus, not only do DD estimators, like DA<br />

ones, exhibit coupling b<strong>et</strong>ween users in that the estimation of the phase<br />

param<strong>et</strong>er φu of user u also <strong>de</strong>pends on the estimation of φk, k = u, but<br />

the use of <strong>de</strong>cisions within the param<strong>et</strong>er estimation process introduces<br />

another kind of coupling: b<strong>et</strong>ween <strong>de</strong>tection and estimation stages.<br />

Again, s<strong>et</strong>ting (5.1) equal to zero is a necessary but not sufficient condition<br />

to <strong>de</strong>rive the ML estimate. It can give birth to two different kinds of estimators.<br />

On the one hand, (5.1) can be used as error signal um u,DD driving a


110 Decision Directed<br />

phase recovery loop<br />

<br />

∂ΛL(Φ) <br />

<br />

∂φu<br />

Φ= ˆ Φ<br />

= 2EuT<br />

N0<br />

N<br />

m=1<br />

u m u,DD<br />

=0. (5.2)<br />

On the other hand, solving (5.1) for Φ produces a DD ML FF phase estimator.<br />

Both implementations, FB and FF, will be <strong>de</strong>alt with in the following<br />

sections.<br />

5.1 Feedback<br />

The DD recovery loop, shown in Figure 5.1 in a 2-user case, is driven by<br />

the error signal u m u,DD<br />

u m u,DD<br />

⎡<br />

⎢<br />

= ⎢<br />

⎣<br />

e j ˆ φ m u<br />

Nu <br />

k=1<br />

k=u<br />

Î m u<br />

⋆<br />

y m u<br />

Ek<br />

Eu ej( ˆ φ m k ˆ φ m l ) +<br />

n=<br />

Î m u<br />

As mentioned earlier, the use of the <strong>de</strong>cisions Îm u<br />

⋆Î n m n<br />

k xu,k ⎤<br />

⎥<br />

⎦ . (5.3)<br />

to provi<strong>de</strong> the information<br />

requested by the estimation process introduces a coupling b<strong>et</strong>ween<br />

estimation and <strong>de</strong>cision stages which does not appear in the DA phase recovery<br />

loop since this one can rely on pre<strong>de</strong>fined training sequences.<br />

Expanding the matched filter output y m u in (5.3), the error signal <strong>de</strong>rived<br />

from the ML phase estimation of user u writes<br />

u m ⎡<br />

⎢ e<br />

⎢<br />

u,DD = ⎢<br />

⎣<br />

j(φu ˆ φm u ) + ⋆ Îm u I<br />

n=<br />

n u<br />

+ Nu <br />

e<br />

k=1<br />

k=u<br />

j(φv ˆ φm u ) + Ev<br />

Eu<br />

n=<br />

Nu <br />

e<br />

k=1<br />

k=u<br />

j( ˆ φm v ˆ φm u ) + Ev<br />

Eu<br />

n=<br />

+ e j ˆ φm ⋆ u Îm u νm u<br />

xm n<br />

u,u<br />

Î m u<br />

Î m u<br />

⋆<br />

I n v<br />

xm n<br />

u,v<br />

⋆Î n<br />

v xm n<br />

u,v<br />

⎤<br />

⎥ . (5.4)<br />

⎥<br />

⎦<br />

The contributions appearing in (5.4) have the same significance as in the<br />

DA case (4.4). As far as the MAI is concerned, the matched filter output<br />

introduces the interference (second term) and the MU estimator tries to


(t)<br />

h ⋆ u ( t)<br />

h ⋆ v ( t)<br />

y m u<br />

e j ˆ φ m u<br />

e j ˆ φ m v<br />

y m v<br />

(.) ⋆<br />

NCO<br />

NCO<br />

e j ˆ φ m u y m u<br />

<br />

<br />

e j ˆ φ m v y m v<br />

u m u<br />

u m v<br />

+<br />

-<br />

(.) ⋆<br />

Figure 5.1: 2-user DD phase recovery loop<br />

-<br />

+<br />

(.) ⋆<br />

<br />

(.) ⋆<br />

(.) ⋆<br />

Î m u<br />

Î m v<br />

x m u,v<br />

5.1 Feedback 111


112 Decision Directed<br />

mitigate it (third term). However, DD estimators exhibit two restrictions<br />

with respect to their DA counterparts.<br />

First of all, while in DA structures the mitigation term reduces the influence<br />

of the MAI as soon as the phase estimation error related to the<br />

interfering users is small, the success of the mitigation in DD structures<br />

requests also that the <strong>de</strong>tection stage provi<strong>de</strong>s correct <strong>de</strong>cisions. In<strong>de</strong>ed,<br />

the MAI disturbance term and the MAI mitigation term in (4.4) only differ<br />

in their phases since they both rely on pre<strong>de</strong>fined training sequences.<br />

This is no longer the case with DD estimators (5.4). In such structures, the<br />

mitigation term cancels the interference provi<strong>de</strong>d that two conditions are<br />

fulfilled, namely that the phase estimation error is small and that the <strong>de</strong>cisions<br />

are correct.<br />

The second difference b<strong>et</strong>ween DA and DD implementations regards the<br />

aforementioned <strong>de</strong>cisions. Using training sequences, DA estimators can<br />

exploit the entire transmitted sequence in or<strong>de</strong>r to perform param<strong>et</strong>er estimation.<br />

On the other hand, DD estimators rely on <strong>de</strong>cisions provi<strong>de</strong>d by<br />

the <strong>de</strong>tector. This imposes on them a causal working mo<strong>de</strong> in which the<br />

interference related to un<strong>de</strong>tected symbols cannot be mitigated. As a result,<br />

the MAI mitigation term in (5.4) can be built including, at most, past<br />

and present <strong>de</strong>tected symbols.<br />

In the following paragraphs, the study of DD ML FB estimators will be<br />

split into two main parts. The first part will assume that the <strong>de</strong>cisions are<br />

correct up to the present time. In<strong>de</strong>ed, it will lead to reinterpr<strong>et</strong> relations<br />

presented in the previous chapter for DA estimators in the light of causality.<br />

On the other hand, the second part will make no assumption regarding<br />

the quality of the <strong>de</strong>cisions, thus including possible faulty outcomes.<br />

A new and original open-loop study will be performed and some aspects<br />

related to the closed-loop performance study will also be introduced. Notice<br />

that, for the ease of treatment, the restriction of causality mentioned<br />

here above will be relaxed in the second part.<br />

5.1.1 Decisions assumed correct<br />

Firstly <strong>de</strong>cisions are assumed to be correct. In this respect, the results<br />

presented in the previous chapter in the case of DA estimators can be used<br />

here to illustrate the performance of DD structures, provi<strong>de</strong>d that the estimators<br />

are ma<strong>de</strong> causal. This treatment is to be presented hereafter for


5.1 Feedback 113<br />

BPSK- and QPSK-modulated data symbols.<br />

BPSK modulation<br />

Consi<strong>de</strong>ring first BPSK modulation, the <strong>de</strong>velopments r<strong>et</strong>urn at relation<br />

(4.14), which gives the auto-correlation function of the loop noise at equilibrium.<br />

Limiting the MAI mitigation term to causal contributions, (4.14)<br />

becomes<br />

⎧ +<br />

⎫<br />

C m ⎪⎨<br />

u,u (0) = δ(m)<br />

⎪⎩<br />

p=<br />

[ (x p u,u)] 2<br />

+ N0x0 u,u<br />

2EuT<br />

+ Nu Ek<br />

Eu<br />

k=1 p=<br />

k=u<br />

Nu Ek<br />

Eu<br />

k=1 p=<br />

k=u<br />

+<br />

0<br />

2 ejδk,ux p<br />

u,k<br />

2 ejδk,ux p<br />

u,k<br />

⎪⎬<br />

⎪⎭<br />

x m 2 u,u .<br />

(5.5)<br />

It has been stressed in Chapter 4 <strong>de</strong>aling with DA estimators that the advantage<br />

of MU estimators with respect to SU ones lies in the mitigation<br />

of the MAI entering the system (third term of (5.5)) by the last term of the<br />

relation. This advantage is partly lost in DD estimators, in as much as only<br />

the causal part of the MAI (p 0) is mitigated. Thus, the DD estimator is<br />

plagued by the anti-causal part of the MAI .<br />

As a result, the variance of the DD estimator is greater than, or at best<br />

equal to the one of the DA estimator. In<strong>de</strong>ed, un<strong>de</strong>r some conditions (nondispersive<br />

channels for instance), the MAI inci<strong>de</strong>nce is con<strong>de</strong>nsed in the<br />

x 0 u,v<br />

coefficient. Then, if this one is involved in the mitigation term for<br />

DD estimators as it is for DA ones, both estimators exhibit the same variance.<br />

On the other hand, strictly limiting the mitigation term to p


114 Decision Directed<br />

Variance [rad 2 ]<br />

10 1<br />

2−user system − 31−chip Gold co<strong>de</strong>s − BPSK modulation − R = 1e4 Bauds − HT channel − 2 B T = 0.1<br />

N<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

DA<br />

DD with<br />

DD without<br />

Single−user<br />

Multiuser<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 5.2: Variance of DD ML FB estimators in ISI-free scenario (BPSK)<br />

Nevertheless, in dispersive environments, the MAI inci<strong>de</strong>nce is spread<br />

over a span of coefficients x m u,v, m =0, 1,.... Those among them which<br />

contribute to the anti-causal part of the interference provoke an increase<br />

of the variance. This is illustrated in Figure 5.3, where the variances of<br />

DA and DD estimators are compared in a dispersive HT channel. Again,<br />

curves are shown for two DD estimators, one including the x 0 u,v contribution,<br />

the other not. The variance floor that limits the performance of the<br />

DA estimator as a result of ISI stands below the variance floor related to<br />

the DD estimator. The difference b<strong>et</strong>ween them is due to the imperfect<br />

MAI mitigation. However, the rea<strong>de</strong>r can notice that <strong>de</strong>spite missing the<br />

MAI due to the present symbol the second MU DD estimator (curve ”DD<br />

without”) still performs b<strong>et</strong>ter than the SU one thanks to the mitigation of<br />

MAI due to past symbols.<br />

QPSK modulation<br />

Moving to QPSK modulation, the auto-correlation function of the loop<br />

noise at equilibrium is given by (4.16). Again, limiting its MAI mitigation


5.1 Feedback 115<br />

Variance [rad 2 ]<br />

10 1<br />

2−user system − 31−chip Gold co<strong>de</strong>s − BPSK modulation − R = 1e5 Bauds − HT channel − 2 B T = 0.1<br />

N<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

DA<br />

DD with<br />

DD without<br />

Single−user<br />

Multiuser<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 5.3: Variance of DD ML FB estimators in presence of ISI (BPSK)<br />

term to causal contributions, it turns into<br />

C m 1<br />

u,u (0) =<br />

2<br />

⎧<br />

⎡<br />

⎢<br />

⎪⎨<br />

⎢<br />

δ (m) ⎢<br />

⎣<br />

⎪⎩<br />

+<br />

p=<br />

x p u,u 2<br />

+ N0x0 u,u<br />

EuT<br />

+ Nu Ek<br />

Eu<br />

k=1 p=<br />

k=u<br />

Nu Ek<br />

Eu<br />

k=1 p=<br />

k=u<br />

+<br />

0<br />

<br />

<br />

x p<br />

<br />

<br />

x p<br />

<br />

<br />

u,k<br />

2<br />

<br />

<br />

u,k<br />

2<br />

⎤<br />

⎥<br />

⎦<br />

⎫<br />

<br />

m<br />

x <br />

u,u<br />

2<br />

⎪⎬<br />

.<br />

⎪⎭<br />

(5.6)<br />

The same conclusions apply in the QPSK case as in the BPSK one: due to<br />

partial MAI mitigation, the variance of the DD estimator is greater than<br />

the DA one. In an ISI-free scenario (Figure 5.4), it can lead to an MU DD<br />

estimator exhibiting the same variance than its SU counterpart if the x 0 u,v<br />

contribution is not taken into account (curve ”DD without”). However, the<br />

performance of the DD estimator is b<strong>et</strong>ter than the SU one in dispersive<br />

channels (Figure 5.5) when MAI coming from past symbols is cancelled.


116 Decision Directed<br />

Variance [rad 2 ]<br />

10 1<br />

2−user system − 31−chip Gold co<strong>de</strong>s − QPSK modulation − R = 1e4 Bauds − HT channel − 2 B T = 0.1<br />

N<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

DA<br />

DD with<br />

DD without<br />

Single−user<br />

Multiuser<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 5.4: Variance of DD ML FB estimators in ISI-free scenario (QPSK)<br />

In the following paragraphs of this section <strong>de</strong>dicated to the FB implementation<br />

of an MU DD ML phase estimator, no assumption is to be ma<strong>de</strong> regarding<br />

the correctness of the <strong>de</strong>cisions used in the <strong>de</strong>tection process. As<br />

a result, new and original relations will be <strong>de</strong>rived in which the inci<strong>de</strong>nce<br />

of <strong>de</strong>tection errors will appear. Another difference with the current paragraph<br />

is that the causal restriction will be lifted.<br />

5.1.2 Actual <strong>de</strong>cisions - Open-loop study<br />

Similarly to the treatment presented in Chapter 4, the study of the recovery<br />

loop splits into open-loop and closed-loop studies. The present paragraph<br />

presents the first one.<br />

Direct-space - Brute-force <strong>de</strong>velopment in a simplified context<br />

The analytical expression of Uu,DD, the mean of the error signal u m u,DD ,is<br />

to be <strong>de</strong>rived as a function of the phase estimation error ∆. This expression<br />

illustrates the working of the multiuser phase estimator through the<br />

drawing of S-hypersurfaces. S-hypersurfaces are multi-dimensional extensions<br />

of S-curves. This multi-dimensional aspect comes from the fact<br />

that Uu,DD <strong>de</strong>pends not only on the phase estimation error of user u but


5.1 Feedback 117<br />

Variance [rad 2 ]<br />

10 1<br />

2−user system − 31−chip Gold co<strong>de</strong>s − QPSK modulation − R = 1e5 Bauds − HT channel − 2 B T = 0.1<br />

N<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

DA<br />

DD with<br />

DD without<br />

Single−user<br />

Multiuser<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 5.5: Variance of DD ML FB estimators in presence of ISI (QPSK)<br />

also on the estimation errors related to the interfering users.<br />

For reasons which will be explained later, the study will be limited to a<br />

simplified context, namely a 2-user non frequency-selective synchronous<br />

system. In this case, the only interference to be consi<strong>de</strong>red is the MAI<br />

due to the second user. Starting from (5.4), these simplifications will be<br />

introduced progressively along with the following <strong>de</strong>velopments. In this<br />

2-user system, S-hypersurfaces will <strong>de</strong>generate into S-surfaces.<br />

BPSK modulation Consi<strong>de</strong>ring BPSK-modulated data symbols, the mathematical<br />

expectation of (5.4) in open-loop conditions assuming Φ is given<br />

by<br />

U BPSK<br />

u,DD<br />

<br />

(∆) = E u m <br />

<br />

ˆΦ u,DD =0, Φ=∆<br />

(5.7)


118 Decision Directed<br />

U BPSK<br />

u,DD (∆)<br />

<br />

= e j∆u<br />

+<br />

<br />

+<br />

n=<br />

Ev<br />

Eu<br />

Ev<br />

Eu<br />

<br />

E â m u anu <br />

ˆΦ =0, Φ=∆<br />

e j(δv,u+∆u)<br />

n=<br />

+<br />

j(δv,u+∆u ∆v)<br />

e<br />

n=<br />

m n<br />

xu,u <br />

<br />

E â m u a n <br />

v ˆΦ =0, Φ=∆<br />

+<br />

m n<br />

xu,v <br />

E â m u â n <br />

v ˆΦ =0, Φ=∆<br />

<br />

m n<br />

xu,v +E (â m u ν m u ) (5.8)<br />

since the data symbols I p<br />

k and <strong>de</strong>cisions Îp<br />

k are real-only (BPSK modulation).<br />

Due to the signal constellation symm<strong>et</strong>ry, the additive noise contribution<br />

to (5.8) disappears [87]. As a result, the average error signal in the<br />

multiuser context is ma<strong>de</strong> of three main contributions, the first two from<br />

the expansion of the matched filter output, embedding a useful contribution<br />

(âm u am u ), the ISI (âm u an u,m = n), and the MAI (âm u an v ), and a last one<br />

being the MAI mitigation introduced by the multiuser estimation process<br />

(âm u ânv ).<br />

D<strong>et</strong>ailed expressions of the first or<strong>de</strong>r statistics used in (5.8) are presented<br />

in Appendix D, consi<strong>de</strong>ring synchronous transmissions over a non dispersive<br />

channel. These are the result of a study called ”brute-force”, in<br />

the sense that it has been performed in the direct space by averaging the<br />

performance over all possible realisations of the data symbols regar<strong>de</strong>d as<br />

random variables. Such exhaustive treatment explains the applied simplifications<br />

(2-user non frequency-selective synchronous system). Without<br />

them, the analytical study would have been unrealistic, at least in the direct<br />

space, due to the exponential complexity of the computations in the<br />

number of users and in the <strong>de</strong>lay spread.


5.1 Feedback 119<br />

QPSK modulation With information spread on I- and Q-branches, the<br />

mean of (5.4) expands into<br />

QP SK<br />

Uu,DD (∆)<br />

<br />

= E<br />

⎧<br />

⎨<br />

= <br />

u m u,DD<br />

⎩ ej∆u<br />

⎧<br />

⎨<br />

+<br />

⎪⎨<br />

+<br />

⎪⎩<br />

⎧<br />

⎪⎨<br />

+<br />

⎪⎩<br />

⎧<br />

⎩ ej∆u<br />

n=<br />

⎧ <br />

Ev<br />

Eu ej(δv,u+∆u)<br />

⎪⎨<br />

<br />

⎪⎩<br />

⎧<br />

⎪⎨<br />

<br />

⎪⎩<br />

<br />

<br />

ˆΦ =0, Φ=∆<br />

⎡<br />

+<br />

⎣<br />

n=<br />

(5.9)<br />

E<br />

<br />

âm u anu <br />

ˆΦ =0, Φ=∆<br />

<br />

+E ˆb m<br />

u bn <br />

<br />

u<br />

ˆΦ=0,<br />

⎤ ⎫<br />

⎬<br />

⎦ m n<br />

xu,u Φ=∆ ⎭<br />

⎡<br />

+<br />

⎣ E<br />

<br />

âm u bn <br />

u ˆΦ =0, Φ=∆<br />

<br />

+E ˆb m<br />

u an <br />

<br />

u<br />

ˆΦ=0,<br />

⎤ ⎫<br />

⎬<br />

⎦ m n<br />

xu,u Φ=∆ ⎭<br />

⎡<br />

+<br />

⎣ E<br />

<br />

âm u an <br />

v ˆΦ =0, Φ=∆<br />

<br />

+E ˆb m<br />

u bn <br />

<br />

v ˆΦ=0,<br />

⎫<br />

⎤ ⎪⎬<br />

⎦ xm n<br />

u,v ⎪⎭<br />

Φ=∆<br />

⎡<br />

+<br />

⎣ E<br />

<br />

âm u bnv <br />

ˆΦ =0, Φ=∆<br />

<br />

+E ˆb m<br />

u an <br />

<br />

v ˆΦ=0,<br />

⎫<br />

⎤ ⎪⎬<br />

⎦ xm n<br />

u,v ⎪⎭<br />

Φ=∆<br />

+<br />

⎣ E<br />

<br />

<br />

+E ˆb mˆ u bn v ˆΦ=0,<br />

⎫<br />

⎤ ⎪⎬<br />

⎦ xm n<br />

u,v ⎪⎭<br />

Φ=∆<br />

⎡<br />

+<br />

⎣ E<br />

<br />

âm u ˆb n <br />

<br />

v ˆΦ=0,<br />

<br />

Φ=∆<br />

<br />

+E ˆb m<br />

u ân <br />

<br />

v ˆΦ=0,<br />

⎫<br />

⎤ ⎪⎬<br />

⎦ xm n<br />

u,v ⎪⎭<br />

Φ=∆<br />

<br />

) E ˆb m<br />

u ν m <br />

u . (5.10)<br />

n=<br />

<br />

Ev<br />

Eu ej(δv,u+∆u)<br />

n=<br />

<br />

Ev<br />

∆v)<br />

ej(δv,u+∆u<br />

Eu ⎡ <br />

âm u ânv ˆΦ =0, Φ=∆<br />

n=<br />

<br />

Ev<br />

∆v)<br />

ej(δv,u+∆u<br />

Eu<br />

n=<br />

+E (â m u νm u<br />

As with BPSK-modulated data symbols, the noise contribution in (5.10)<br />

QP SK<br />

vanishes thanks to the constellation symm<strong>et</strong>ry. Uu,DD is then the result<br />

of three contributions which involve signals on the Q-branch and mixed<br />

product of signals from both I- and Q-branches. The first-or<strong>de</strong>r statistics<br />

used in (5.10) are also <strong>de</strong>tailed in Appendix D.


120 Decision Directed<br />

Computational results Introducing results of Appendix D into (5.8) and<br />

(5.10), the S-surfaces have been drawn in three different scenarii, differing<br />

from each other by the level of global coupling (cross-correlation value +<br />

Near-Far ratio) b<strong>et</strong>ween users.<br />

In an uncoupled context (xv,u =0), S-surfaces <strong>de</strong>generate into S-curves.<br />

Figure 5.6 compares the S-curves representing the mean Uu,DD of the error<br />

signal u m u,DD with respect to the phase estimation error ∆u, param<strong>et</strong>-<br />

rised on the modulation (BPSK or QPSK) and on the Es<br />

N0<br />

ratio in such an<br />

uncoupled context. Several remarks can be ma<strong>de</strong>. Firstly, these curves present<br />

a 2π<br />

M -periodicity due to the phase ambiguity inherent to the <strong>de</strong>cision<br />

process [83, p. 206]. In<strong>de</strong>ed, without any si<strong>de</strong> information, the receiver<br />

makes ambiguous <strong>de</strong>cisions up to a shift of a multiple of 2π<br />

M . Secondly,<br />

the slopes of both S-curves rise up to 1 with Es<br />

[84, p. II-16]. In the un-<br />

N0<br />

coupled situation, <strong>de</strong>cision errors are only due to the noisy environment.<br />

The higher the Es<br />

ratio is, the less numerous <strong>de</strong>cision errors are, and the<br />

N0<br />

closer the S-curve becomes to its DA counterpart which was shown to exhibit<br />

a unit slope. Moreover, the vulnerability of QPSK to <strong>de</strong>cision errors<br />

due to additive noise with respect to BPSK increasingly turns into a lower<br />

value of the slope at the same Es ratio. However, both BPSK and QPSK<br />

N0<br />

S-curves converge to the unit slope.<br />

Sticking to the uncoupled context, a broa<strong>de</strong>r view of the situation is gained<br />

by looking at Figure 5.7 which shows the S-surface Uu,DD (∆u, ∆v) for both<br />

BPSK and QPSK cases. Of course, since the users are consi<strong>de</strong>red to be orthogonal<br />

(xv,u =0), the S-surface exhibits no sensitivity to the interfering<br />

estimation process ∆v. This is obvious on Figures 5.9a and 5.9b which<br />

show the traces of the S-surface intersected by planes perpendicular to the<br />

∆u-axis. These traces are thus S-curves Uu,DD (∆v) ∆u function of ∆v and<br />

param<strong>et</strong>rised on ∆u. These S-curves are flat since there is no sensitivity to<br />

the interfering phase estimation error ∆v. On the other hand, their counterpart<br />

Uu,DD (∆u) ∆v (Figures 5.8a and 5.8b) illustrate the inci<strong>de</strong>nce of the<br />

useful phase estimation error on the mean of the error signal. In fact, the<br />

S-curves presented in Figure 5.6 are similar cuts ma<strong>de</strong> in the S-surface.<br />

Introducing some coupling in the system modifies the S-surface in a way<br />

that shows the influence of the interfering estimation process on the useful<br />

one. It appears on Figures 5.11a and 5.11b through a broa<strong>de</strong>ning of the<br />

conglomerate traces, while Figures 5.12a and 5.12b more clearly present a


5.1 Feedback 121<br />

U u,DD<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

−0.2<br />

−0.4<br />

−0.6<br />

−0.8<br />

0<br />

2−user system − x = 0 − AWGN channel<br />

v,u<br />

BPSK<br />

QPSK<br />

E s /N 0 = 0 dB<br />

E s /N 0 = 5 dB<br />

E s /N 0 = 10 dB<br />

E s /N 0 = 20 dB<br />

−1<br />

−2 −1.5 −1 −0.5 0<br />

Δ [rad]<br />

u<br />

0.5 1 1.5 2<br />

Figure 5.6: S-curves in a 2-user non-dispersive synchronous system, xv,u =<br />

0<br />

sensitivity to ∆v.<br />

Going one step further, Figure 5.13 shows the situation in presence of a<br />

Near-Far effect. It exacerbates the results observed with mo<strong>de</strong>rate coupling<br />

in Figure 5.10. Conglomerate S-curves (Figures 5.14a and 5.14b) present<br />

a wi<strong>de</strong> broa<strong>de</strong>ning while the traces obtained at ∆u constant (Figures<br />

5.15a and 5.15b) have now the shape of S-curves with respect to the phase<br />

estimation error ∆v of the interfering user. The point where both phase<br />

estimation errors ∆u and ∆v g<strong>et</strong> to zero is a stable operating point for the<br />

MU DD phase recovery loop.


122 Decision Directed<br />

Phase Error D<strong>et</strong>ector<br />

Phase Error D<strong>et</strong>ector<br />

1<br />

0.5<br />

0<br />

−0.5<br />

−1<br />

−4<br />

0.4<br />

0.3<br />

0.2<br />

0.1<br />

0<br />

−0.1<br />

−0.2<br />

−0.3<br />

−0.4<br />

−4<br />

0<br />

2−user system − x = 0 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−2<br />

−2<br />

0<br />

Δ u [rad]<br />

2<br />

4<br />

4<br />

2<br />

0<br />

Δ v [rad]<br />

(a) S-surface U BPSK<br />

u,DD (∆u, ∆v)<br />

0<br />

2−user system − x = 0 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

0<br />

Δ u [rad]<br />

2<br />

4<br />

4<br />

2<br />

0<br />

Δ v [rad]<br />

QP SK<br />

(b) S-surface Uu,DD (∆u, ∆v)<br />

Figure 5.7: S-surfaces of a 2-user non-dispersive synchronous system, uncoupled<br />

scenario (a: BPSK, b: QPSK)<br />

−2<br />

−2<br />

−4<br />

−4


5.1 Feedback 123<br />

Phase Error D<strong>et</strong>ector<br />

Phase Error D<strong>et</strong>ector<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

−0.2<br />

−0.4<br />

−0.6<br />

−0.8<br />

0<br />

2−user system − x = 0 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−1<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

u<br />

1 2 3 4<br />

0.4<br />

0.3<br />

0.2<br />

0.1<br />

0<br />

−0.1<br />

−0.2<br />

−0.3<br />

(a) S-curve U BPSK<br />

u,DD (∆u) ∆v<br />

0<br />

2−user system − x = 0 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−0.4<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

u<br />

1 2 3 4<br />

QP SK<br />

(b) S-curve U<br />

u,DD (∆u)<br />

<br />

<br />

<br />

∆v<br />

Figure 5.8: S-curves function of ∆u, param<strong>et</strong>rised on ∆v - 2-user nondispersive<br />

synchronous system, uncoupled scenario (a: BPSK, b: QPSK)


124 Decision Directed<br />

Phase Error D<strong>et</strong>ector<br />

Phase Error D<strong>et</strong>ector<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

−0.2<br />

−0.4<br />

−0.6<br />

−0.8<br />

0<br />

2−user system − x = 0 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−1<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

v<br />

1 2 3 4<br />

0.4<br />

0.3<br />

0.2<br />

0.1<br />

0<br />

−0.1<br />

−0.2<br />

−0.3<br />

(a) S-curve U BPSK<br />

u,DD (∆v) ∆u<br />

0<br />

2−user system − x = 0 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−0.4<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

v<br />

1 2 3 4<br />

QP SK<br />

(b) S-curve U<br />

u,DD (∆v)<br />

<br />

<br />

<br />

∆u<br />

Figure 5.9: S-curves function of ∆v, param<strong>et</strong>rised on ∆u - 2-user nondispersive<br />

synchronous system, uncoupled scenario (a: BPSK, b: QPSK)


5.1 Feedback 125<br />

Phase Error D<strong>et</strong>ector<br />

Phase Error D<strong>et</strong>ector<br />

1<br />

0.5<br />

0<br />

−0.5<br />

−1<br />

−4<br />

0.3<br />

0.2<br />

0.1<br />

0<br />

−0.1<br />

−0.2<br />

−0.3<br />

−0.4<br />

−4<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−2<br />

−2<br />

0<br />

Δ u [rad]<br />

2<br />

4<br />

4<br />

2<br />

0<br />

Δ v [rad]<br />

(a) S-surface U BPSK<br />

u,DD (∆u, ∆v)<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

0<br />

Δ u [rad]<br />

2<br />

4<br />

4<br />

2<br />

0<br />

Δ v [rad]<br />

QP SK<br />

(b) S-surface Uu,DD (∆u, ∆v)<br />

Figure 5.10: S-surfaces of a 2-user non-dispersive synchronous system,<br />

coupled scenario (a: BPSK, b: QPSK)<br />

−2<br />

−2<br />

−4<br />

−4


126 Decision Directed<br />

Phase Error D<strong>et</strong>ector<br />

Phase Error D<strong>et</strong>ector<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

−0.2<br />

−0.4<br />

−0.6<br />

−0.8<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−1<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

u<br />

1 2 3 4<br />

0.3<br />

0.2<br />

0.1<br />

0<br />

−0.1<br />

−0.2<br />

−0.3<br />

(a) S-curve U BPSK<br />

u,DD (∆u) ∆v<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−0.4<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

u<br />

1 2 3 4<br />

QP SK<br />

(b) S-curve U<br />

u,DD (∆u)<br />

<br />

<br />

<br />

∆v<br />

Figure 5.11: S-curves function of ∆u, param<strong>et</strong>rised on ∆v - 2-user nondispersive<br />

synchronous system, coupled scenario (a: BPSK, b: QPSK)


5.1 Feedback 127<br />

Phase Error D<strong>et</strong>ector<br />

Phase Error D<strong>et</strong>ector<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

−0.2<br />

−0.4<br />

−0.6<br />

−0.8<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−1<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

v<br />

1 2 3 4<br />

0.3<br />

0.2<br />

0.1<br />

0<br />

−0.1<br />

−0.2<br />

−0.3<br />

(a) S-curve U BPSK<br />

u,DD (∆v) ∆u<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 0 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−0.4<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

v<br />

1 2 3 4<br />

QP SK<br />

(b) S-curve U<br />

u,DD (∆v)<br />

<br />

<br />

<br />

∆u<br />

Figure 5.12: S-curves function of ∆v, param<strong>et</strong>rised on ∆u - 2-user nondispersive<br />

synchronous system, coupled scenario (a: BPSK, b: QPSK)


128 Decision Directed<br />

Phase Error D<strong>et</strong>ector<br />

Phase Error D<strong>et</strong>ector<br />

1<br />

0.5<br />

0<br />

−0.5<br />

−1<br />

−4<br />

0.3<br />

0.2<br />

0.1<br />

0<br />

−0.1<br />

−0.2<br />

−0.3<br />

−0.4<br />

−4<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 10 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−2<br />

−2<br />

0<br />

Δ u [rad]<br />

2<br />

4<br />

4<br />

2<br />

0<br />

Δ v [rad]<br />

(a) S-surface U BPSK<br />

u,DD (∆u, ∆v)<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 4 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

0<br />

Δ u [rad]<br />

2<br />

4<br />

4<br />

2<br />

0<br />

Δ v [rad]<br />

QP SK<br />

(b) S-surface Uu,DD (∆u, ∆v)<br />

Figure 5.13: S-surfaces of a 2-user non-dispersive synchronous system,<br />

Near-Far scenario (a: BPSK, b: QPSK)<br />

−2<br />

−2<br />

−4<br />

−4


5.1 Feedback 129<br />

Phase Error D<strong>et</strong>ector<br />

Phase Error D<strong>et</strong>ector<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

−0.2<br />

−0.4<br />

−0.6<br />

−0.8<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 10 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−1<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

u<br />

1 2 3 4<br />

0.25<br />

0.2<br />

0.15<br />

0.1<br />

0.05<br />

0<br />

−0.05<br />

−0.1<br />

−0.15<br />

−0.2<br />

(a) S-curve U BPSK<br />

u,DD (∆u) ∆v<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 4 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−0.25<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

u<br />

1 2 3 4<br />

QP SK<br />

(b) S-curve U<br />

u,DD (∆u)<br />

<br />

<br />

<br />

∆v<br />

Figure 5.14: S-curves function of ∆u, param<strong>et</strong>rised on ∆v - 2-user nondispersive<br />

synchronous system, Near-Far scenario (a: BPSK, b: QPSK)


130 Decision Directed<br />

Phase Error D<strong>et</strong>ector<br />

Phase Error D<strong>et</strong>ector<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

−0.2<br />

−0.4<br />

−0.6<br />

−0.8<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 10 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−1<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

v<br />

1 2 3 4<br />

0.25<br />

0.2<br />

0.15<br />

0.1<br />

0.05<br />

0<br />

−0.05<br />

−0.1<br />

−0.15<br />

−0.2<br />

(a) S-curve U BPSK<br />

u,DD (∆v) ∆u<br />

0<br />

2−user system − x = 0.2 − δv,u = 0.1745 rad − Near−Far ratio = 4 dB − E /N = 10 dB<br />

v,u<br />

s 0<br />

−0.25<br />

−4 −3 −2 −1 0<br />

Δ [rad]<br />

v<br />

1 2 3 4<br />

QP SK<br />

(b) S-curve U<br />

u,DD (∆v)<br />

<br />

<br />

<br />

∆u<br />

Figure 5.15: S-curves function of ∆v, param<strong>et</strong>rised on ∆u - 2-user nondispersive<br />

synchronous system, Near-Far scenario (a: BPSK, b: QPSK)


5.1 Feedback 131<br />

Reciprocal space - Characteristic function<br />

The appreciation of the <strong>de</strong>velopments in direct space as presented in the<br />

previous section, is mixed. To their advantage, one notices that they illustrate<br />

the specificity of the multiuser phase recovery loop in a 2-user system.<br />

In<strong>de</strong>ed, it shows the inci<strong>de</strong>nce of the interfering estimation process<br />

on the useful one. However, the fact that these <strong>de</strong>velopments are limited<br />

to a 2-user system due to the heavy <strong>de</strong>rivations that would be required<br />

in the case of more complex systems belongs to their shortcomings. Y<strong>et</strong>,<br />

would it be possible to <strong>de</strong>velop a performance mo<strong>de</strong>l that is valid for systems<br />

accommodating more users ?<br />

A way to answer this question is to apply more efficient calculation m<strong>et</strong>hods,<br />

such as the one <strong>de</strong>scribed in Section 3.5.2. The general expression of<br />

the mean of the error signal will be <strong>de</strong>rived in the following paragraphs<br />

by using this m<strong>et</strong>hod. This expression will be illustrated in several cases.<br />

General expressions The global expressions of the mean Uu,DD of the<br />

estimation error um u,DD have been <strong>de</strong>rived for BPSK- and QPSK-modulated<br />

data symbols.<br />

In the case of an MU DD phase recovery loop <strong>de</strong>aling with BPSK-modulated<br />

data symbols, the mean of the estimation error is given by relation<br />

(5.8). The expressions of the first-or<strong>de</strong>r statistics shown in Appendix E.1<br />

which have been <strong>de</strong>rived in reciprocal space with the help of the characteristic<br />

function, were used in (5.8) instead of their counterparts shown<br />

in Appendix D, and which were obtained in direct space. This led to the<br />

general expression of the mean of the phase estimation error in the case of<br />

BPSK modulation. It is given by relation (F.1).<br />

Similarly, moving to QPSK modulation, the use of the first-or<strong>de</strong>r statistics<br />

<strong>de</strong>tailed in Appendix E.2 turns the mean of the estimation error given by<br />

(5.10) into (F.3).<br />

Computational results in a simplified context In or<strong>de</strong>r to g<strong>et</strong> some insight<br />

into (F.1) and (F.3), these expressions have been <strong>de</strong>rived at equilibrium<br />

(∆ =0) in a 2-user i<strong>de</strong>al (neither AWGN nor ISI) system. The means<br />

have then been computed and plotted as a function of δv,u, the true phase<br />

difference b<strong>et</strong>ween users. On the other hand, the working of the open-loop


132 Decision Directed<br />

configuration has been simulated, enabling to compare analytical and simulation<br />

results. Moreover, they have also been compared to the value at<br />

equilibrium of the S-surfaces param<strong>et</strong>rised on δv,u obtained in direct space<br />

(Section 5.1.2).<br />

¯ BPSK modulation<br />

With the help of [116], (F.1) becomes at equilibrium in the i<strong>de</strong>al situation<br />

U BPSK<br />

u,DD (0)<br />

<br />

Nu <br />

=2<br />

N0 <br />

=0<br />

p q<br />

xk,l =0 p = q<br />

= 1<br />

2 I0 <br />

0<br />

u,v sign Ru,u + R 0 <br />

0<br />

u,v sign Ru,u R 0 <br />

u,v<br />

1<br />

2 I0 <br />

sign R0 u,u + R<br />

u,v<br />

0 <br />

u,v sign R0 v,u + R0 <br />

v,v<br />

+sign R0 u,u R0 <br />

u,v sign R0 v,u R0 <br />

.(5.11)<br />

v,v<br />

(5.11) was computed with respect to the phase difference b<strong>et</strong>ween<br />

users u and v, δv,u = φv φu and illustrated in Figure 5.16. In<strong>de</strong>ed,<br />

this figure is threefold. First, it illustrates the result of the computations<br />

(circles) with and without the mitigating term (second term<br />

of 5.11). Including the mitigating term simulates the MU estimator<br />

while not including it simulates the SU one. Second, it shows simulation<br />

results, drawn as continuous lines. Simulated values of<br />

U BPSK<br />

u,DD (0) were obtained through Monte-Carlo simulations of the<br />

open-loop configuration. The phase was supposed to be perfectly<br />

estimated so as to illustrate the inci<strong>de</strong>nce of MAI. Finally, analytical<br />

results <strong>de</strong>rived in direct space (Section 5.1.2) are shown as crosses.<br />

The rea<strong>de</strong>r can notice the match b<strong>et</strong>ween analytical results in direct<br />

and in reciprocal spaces, and also b<strong>et</strong>ween analytical and simulation<br />

results.<br />

Since Figure 5.16 shows the mean U BPSK<br />

u,DD (0) at equilibrium (∆ =0)<br />

as a function of δv,u, the i<strong>de</strong>al situation is to have this mean null<br />

everywhere. This is the case for the multiuser curve, not for the<br />

single-user one. In<strong>de</strong>ed, a bias affects periodically the SU estimation<br />

process around δv,u values such that arg x0 <br />

u,v + δv,u = kπ. This<br />

bias comes from the fact that the inci<strong>de</strong>nce of the MAI <strong>de</strong>pends on<br />

the phase difference b<strong>et</strong>ween users. This is <strong>de</strong>scribed analytically


5.1 Feedback 133<br />

BPSK<br />

U ( 0 )<br />

u,DD<br />

1<br />

0.5<br />

0<br />

−0.5<br />

−1<br />

−1.5<br />

0<br />

2−user system − x = 0.5 − Near−Far ratio = 10 dB − Es /N = 50 dB<br />

v,u<br />

0<br />

Single−user<br />

Multiuser<br />

−3 −2 −1 0<br />

δ [rad]<br />

v,u<br />

1 2 3<br />

Figure 5.16: U BPSK<br />

u,DD (0) as a function of δv,u (- simulation, ¢ computation<br />

direct space, Æ computation reciprocal space)<br />

by the first term of (5.11), which <strong>de</strong>pends on I 0 u,v. Bearing in mind<br />

that φl<br />

ˆ φk = δl,k +∆k, <strong>de</strong>finitions (3.77) and (3.78) explain how<br />

the phase difference b<strong>et</strong>ween users influence the performance of the<br />

single-user estimator.<br />

To explain this more physically, a <strong>de</strong>tection error occurs as soon as<br />

the interference phasor brings the resulting phasor of the matched<br />

filter output out of the right <strong>de</strong>tection zone (Figure 5.17). This occurs<br />

when the contribution of the interferer is stronger than the one of<br />

the user. In such a case, the output of the matched filter is driven by<br />

the interferer and any param<strong>et</strong>er estimation process relying only on<br />

this output goes wrong. This is the case of the SU phase estimator,<br />

leading thus to a bias. Nevertheless, this bias disappears in the multiuser<br />

structure thanks to the correcting term in (5.3) that mitigates<br />

the influence of the MAI on the matched filter output.<br />

¯ QPSK modulation<br />

In the 2-user non-dispersive synchronous system, (F.3) writes at equi-


134 Decision Directed<br />

<br />

arg x0 <br />

0<br />

v,u + δv,u x 0 u,u<br />

Ev<br />

<br />

Eu<br />

<br />

0 x <br />

v,u<br />

Figure 5.17: Phasor contributions of user u and interferer v to matched<br />

filter output y m u for BSPK-modulated data symbols<br />

librium<br />

<br />

Nu <br />

=2<br />

N0 <br />

=0<br />

p q<br />

xk,l =0 p = q<br />

= 1<br />

4 sign R 0 u,u + R 0 u,v + I 0 0<br />

u,v Ru,v I 0 <br />

u,v<br />

+ 1<br />

4 sign R 0 u,u + R 0 u,v I0 0<br />

u,v Ru,v + I 0 <br />

u,v<br />

+ 1<br />

4 sign R 0 u,u R 0 u,v + I0 <br />

0<br />

u,v Ru,v I 0 <br />

u,v<br />

1<br />

4 sign R 0 u,u R 0 u,v I 0 0<br />

u,v Ru,v + I 0 <br />

u,v<br />

+ 1<br />

8 I0 <br />

sign R0 u,u<br />

u,v<br />

R0 u,v + I0 <br />

u,v<br />

+sign R0 u,u R0 u,v I0 <br />

<br />

u,v<br />

sign R0 v,v R0 v,u + I0 <br />

v,u<br />

+sign R0 v,v R0 v,u I0 <br />

<br />

v,u<br />

1<br />

8 I0 <br />

sign R0 u,u +<br />

u,v<br />

R0 u,v + I0 <br />

u,v<br />

+sign R0 u,u + R0 u,v I0 <br />

<br />

u,v<br />

sign R0 v,v + R0 v,u + I0 <br />

v,u<br />

+sign R0 v,v + R0 v,u I0 <br />

<br />

v,u<br />

QP SK<br />

Uu,DD (0)


5.1 Feedback 135<br />

+ 1<br />

8 R0 u,v<br />

1<br />

8 R0 u,v<br />

<br />

sign R0 u,u + R0 u,v + I0 <br />

u,v<br />

+sign R0 u,u R0 u,v I0 <br />

<br />

u,v<br />

sign R0 v,v R0 v,u + I0 <br />

v,u<br />

+sign R0 v,v + R0 v,u I0 <br />

<br />

v,u<br />

<br />

sign R0 u,u + R0 u,v I0 <br />

u,v<br />

+sign R0 u,u R0 u,v + I0 <br />

<br />

u,v<br />

sign R0 v,v + R0 v,u + I0 <br />

v,u<br />

+sign R0 v,v R0 v,u I0 <br />

.<br />

v,u<br />

(5.12)<br />

QP SK<br />

The rea<strong>de</strong>r can notice that Uu,DD given by (5.12) becomes equal to<br />

U BPSK<br />

u,DD (5.11) if(5.12) is only ma<strong>de</strong> of the terms weighting I 0 v,u while<br />

<strong>de</strong>l<strong>et</strong>ing I 0 v,u in the argument of the sign functions. The reason for<br />

keeping only terms weighting I 0 v,u<br />

is that these are the ones present<br />

in BPSK since the error signal takes the imaginary part of the phasecorrected<br />

matched filter output. On the other hand, <strong>de</strong>l<strong>et</strong>ing I 0 v,u<br />

in sign functions comes from the fact that BPSK hard-<strong>de</strong>cisions rely<br />

only on real parts of the phase-corrected matched filter output.<br />

Figure 5.18 presents the result of the computation of (5.12) in the<br />

chosen context. As in Figure 5.16, continuous curves are the result of<br />

Monte-Carlo simulations of the open-loop process with perfect estimate<br />

of the phase. Circles show the computational results of the<br />

QP SK<br />

expression of Uu,DD <strong>de</strong>rived in reciprocal space, while the crosses<br />

illustrate the expression <strong>de</strong>rived in direct space. Again, the matching<br />

b<strong>et</strong>ween computations and simulations is good. However, the<br />

simulation results have som<strong>et</strong>imes slightly diverged from the computational<br />

ones due to numerical inaccuracies.<br />

QP SK<br />

Similarly to the BPSK case, a bias due to MAI affects Uu,DD . This<br />

bias appears when the MAI contribution in the matched filter output<br />

y p<br />

k is stronger than the one from the user of interest, so that the matched<br />

filter output is driven by MAI (Figure 5.19). Then the chosen<br />

<strong>de</strong>tection process produces wrong estimates of the data and an estimation<br />

process relying only on y p<br />

k is biased, as shown in Figure 5.18.<br />

This bias appears around values of δv,u such as arg x0 <br />

u,v +δv,u = k π<br />

2 .<br />

Nevertheless, thanks to the introduction of a correcting term, this<br />

bias is cancelled by the MU estimator.


136 Decision Directed<br />

QPSK<br />

U ( 0 )<br />

u,DD<br />

0.2<br />

0.1<br />

0<br />

−0.1<br />

−0.2<br />

−0.3<br />

0<br />

2−user system − x = 0.5 − Near−Far ratio = 4 dB − Es /N = 50 dB<br />

v,u<br />

0<br />

Single−user<br />

Multiuser<br />

−3 −2 −1 0<br />

δ [rad]<br />

v,u<br />

1 2 3<br />

QP SK<br />

Figure 5.18: Uu,DD (0) as a function of δv,u (- simulation, ¢ computation<br />

direct space, Æ computation reciprocal space)<br />

<br />

arg x0 <br />

v,u + δv,u<br />

x 0 u,u<br />

<br />

Ev<br />

Eu<br />

<br />

0 x <br />

v,u<br />

Figure 5.19: Phasor contributions of user u and interferer v to matched<br />

filter output y m u for QSPK-modulated data symbols


5.1 Feedback 137<br />

In the remain<strong>de</strong>r of this chapter, only the case of BPSK modulation will be<br />

consi<strong>de</strong>red.<br />

Numerical integration To fully exploit the expression (F.1), numerical<br />

integration (Romberg m<strong>et</strong>hod [117]) has been applied. In such approach,<br />

the fact that the integrand is divi<strong>de</strong>d by the integration variable in all<br />

terms of (F.1) is rather inconvenient. It can be solved with a classic change<br />

of variable (Ω =lnω). This leads to (F.2) which is more appropriate for<br />

numerical integration.<br />

The concordance b<strong>et</strong>ween (5.11) and (F.2) has been tested in different 2user<br />

snapshot scenarii consi<strong>de</strong>ring either the strongest or the weakest user<br />

with Eb =10or 30 dB. The results are shown in Figures 5.20 and 5.21 for<br />

N0<br />

an SU estimator in a single-user system (Reference curve), and for singleuser<br />

(SU curve) and multiuser (MU curve) estimators in the 2-user system.<br />

In each case, the first subfigure represents the consi<strong>de</strong>red snapshot<br />

scenario by drawing the phasors of the two users. The second subfigure<br />

illustrates U BPSK<br />

u,DD (0) computed according to (5.11) at equilibrium with respect<br />

to the phase offs<strong>et</strong> δu,v b<strong>et</strong>ween the two users. This phase offs<strong>et</strong> is<br />

the one generated by the phase oscillators. That is why this offs<strong>et</strong> does not<br />

match with the offs<strong>et</strong> shown in the first subfigure which also inclu<strong>de</strong>s the<br />

influence of the channel. The cross in the second subfigure indicates the<br />

value of U BPSK<br />

u,DD (0) given by numerical integration of relation (F.2) at the<br />

offs<strong>et</strong> value δu,v of the snapshot scenario. Finally, the third subfigure illustrates<br />

U BPSK<br />

u,DD (∆u, 0) numerically integrated from (F.2) at the offs<strong>et</strong> δu,v of<br />

the snapshot scenario. The value of U BPSK<br />

u,DD (0) <strong>de</strong>rived analytically with<br />

(5.11) is shown by a circle. The match b<strong>et</strong>ween the analytical <strong>de</strong>rivation<br />

(5.11) and the numerical integration (F.2) is thus measured by the concordance<br />

b<strong>et</strong>ween crosses and circles.<br />

In Figure 5.20, the strongest user of the system is consi<strong>de</strong>red. The second<br />

subfigure shows that no estimation process exhibits a bias at equilibrium,<br />

whatever the true phase difference b<strong>et</strong>ween users is, thanks to the fact that<br />

the MAI introduced by the interfering user is small with regard to the useful<br />

contribution of the user. It has thus no influence on the <strong>de</strong>cision stage.<br />

As a result, U BPSK<br />

1,DD (∆1) shown in the third subfigure has the form of a<br />

classic S-curve.<br />

The situation is quite different in the scenario where user 1 is the weak-


138 Decision Directed<br />

est (Figure 5.21). The MAI contribution of the interfering user provokes<br />

<strong>de</strong>cision errors, which introduce a bias affecting the SU estimator (second<br />

subfigure). This bias also appears on the drawing of U BPSK<br />

1,DD (∆1).


150<br />

210<br />

120<br />

240<br />

90<br />

270<br />

3.7124<br />

2.4749<br />

1.2375<br />

4.9499<br />

180 0<br />

v<br />

(a) Polar representation of the<br />

snapshot scenario<br />

60<br />

300<br />

u<br />

30<br />

330<br />

Bias<br />

Figure 5.20: U BPSK<br />

u,DD<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

−0.2<br />

−0.4<br />

−0.6<br />

−0.8<br />

Δ u = Δ v = 0<br />

SU<br />

MU<br />

−1<br />

−2 −1 0<br />

δ [rad]<br />

u,v<br />

1 2<br />

15<br />

10<br />

5<br />

0<br />

−5<br />

−10<br />

Δ v = 0, δ u,v = 0.88714 rad<br />

Reference<br />

SU<br />

MU<br />

−15<br />

−2 −1 0<br />

Δ [rad]<br />

u<br />

1 2<br />

(b) Comparison of biases computed by (5.11) and <strong>de</strong>rived from the<br />

numerical integration of (F.2)<br />

where user u is the strongest and Eb<br />

N0 =30dB<br />

5.1 Feedback 139


140 Decision Directed<br />

150<br />

210<br />

120<br />

240<br />

90<br />

270<br />

2.3839<br />

1.1919<br />

3.5758<br />

180<br />

u<br />

0<br />

(a) Polar representation of the<br />

snapshot scenario<br />

v<br />

60<br />

300<br />

30<br />

330<br />

Bias<br />

Figure 5.21: U BPSK<br />

u,DD<br />

0.03<br />

0.02<br />

0.01<br />

0<br />

−0.01<br />

−0.02<br />

Δ u = Δ v = 0<br />

SU<br />

MU<br />

−0.03<br />

−2 −1 0<br />

δ [rad]<br />

u,v<br />

1 2<br />

0.015<br />

0.01<br />

0.005<br />

0<br />

−0.005<br />

−0.01<br />

Δ v = 0, δ u,v = 0.55436 rad<br />

Reference<br />

SU<br />

MU<br />

−0.015<br />

−2 −1 0<br />

Δ [rad]<br />

u<br />

1 2<br />

(b) Comparison of biases computed by (5.11) and <strong>de</strong>rived from the<br />

numerical integration of (F.2)<br />

where user u is the weakest and Eb<br />

N0 =10dB


5.1 Feedback 141<br />

5.1.3 Actual <strong>de</strong>cisions - Closed-loop study<br />

Expressions (F.1) and (F.2) are too complex to be used for a closed-loop<br />

study as the one lead in the DA case. Nevertheless, by simplifying them<br />

to a 2-user environment without ISI nor AWGN, one can write<br />

U BPSK<br />

u,DD (∆u, ∆v) = U BPSK<br />

u,DD<br />

where U BPSK<br />

u,DD<br />

writes<br />

∂U BPSK<br />

u,DD<br />

∂∆u<br />

= 1<br />

while<br />

<br />

<br />

<br />

<br />

∆=0<br />

2 x0u,u <br />

Ev<br />

+ 1<br />

2<br />

⎧<br />

⎪⎨<br />

⎪⎩<br />

∂UBP SK<br />

u,DD<br />

∂∆v<br />

∂U BPSK<br />

u,DD<br />

∂UBPSK<br />

u,DD<br />

(0, 0) +<br />

∂∆u<br />

<br />

<br />

<br />

<br />

∆=0<br />

∆u + ∂UBPSK<br />

u,DD<br />

∂∆v<br />

<br />

<br />

is given by (5.11). After some calculations,<br />

∆=0<br />

0<br />

sign Ru,u + R 0 0<br />

u,v +signRu,u R 0 <br />

u,v<br />

Eu<br />

R 0 u,v<br />

<br />

<br />

<br />

<br />

∆=0<br />

∂UBP SK<br />

u,DD<br />

∂∆u<br />

∆v<br />

(5.13)<br />

<br />

<br />

<br />

∆=0<br />

<br />

sign R0 u,u + R0 <br />

u,v 1 sign R0 v,v + R0 <br />

v,u<br />

+sign R0 u,u R0 <br />

u,v 1 sign R0 v,v R0 <br />

<br />

v,u<br />

2 I0 <br />

2 δ R0 u,u + R<br />

u,v<br />

0 <br />

u,v 1 sign R0 v,v + R0 <br />

v,u<br />

+δ R0 u,u R0 <br />

u,v 1 sign R0 v,v R0 ⎫<br />

⎪⎬<br />

<br />

⎪⎭<br />

v,u<br />

(5.14)<br />

<br />

<br />

becomes<br />

∆=0<br />

<br />

<br />

<br />

<br />

∂∆v <br />

∆=0<br />

= 1<br />

⎧<br />

⎪⎨<br />

Ev<br />

2 Eu ⎪⎩<br />

<br />

sign R0 u,u + R0 <br />

u,v sign R0 v,u + R0 <br />

u,v<br />

sign R0 u,u R0 <br />

u,v sign R0 v,u R0 <br />

<br />

u,v<br />

2 I0 <br />

2 sign R0 u,u + R<br />

u,v<br />

0 <br />

u,v δ R0 v,v + R0 <br />

u,v<br />

+sign R0 u,u R0 <br />

u,v δ R0 v,v R0 ⎫<br />

⎪⎬<br />

.<br />

⎪⎭<br />

u,v<br />

(5.15)<br />

R 0 u,v<br />

The rea<strong>de</strong>r can notice that the explicit sensitivity of U BPSK (∆) with re-<br />

u,DD<br />

spect to the interfering user only appears in the case of the MU phase<br />

estimator which contains an MAI mitigation term in its error signal um u,DD .


142 Decision Directed<br />

This is shown by relation (5.15), which is not null if and only if MU estimation<br />

is applied.<br />

Using the previous equations, the phase recovery loop <strong>de</strong>scribed by these<br />

expressions can be drawn. It is shown in a 2-user system in Figure 5.22.<br />

φ m u<br />

φ m v<br />

+<br />

ˆφ m u<br />

+<br />

ˆφ m v<br />

-<br />

∆ m u<br />

- ∆ m v<br />

∂UBP SK<br />

u,DD<br />

∂∆m u<br />

∂UBP SK<br />

v,DD<br />

∂∆m u<br />

K0,u (z 1) 1<br />

∂UBP SK<br />

v,DD<br />

∂∆m v<br />

∂UBP SK<br />

u,DD<br />

∂∆m v<br />

K0,v (z 1) 1<br />

Figure 5.22: 2-user DD phase recovery loop<br />

ξ m u<br />

ξ m u<br />

Fu (z)<br />

Fv (z)<br />

Unfortunately, due to the complexity of the <strong>de</strong>velopments, the study of<br />

the recovery loop has been stopped here.<br />

5.2 Feedforward<br />

Before studying the performance of DD ML FF estimators, some implementation<br />

issues are worth a discussion. These issues are raised by the<br />

fact that the estimators no longer rely on training sequences but on <strong>de</strong>cisions.


5.2 Feedforward 143<br />

5.2.1 SU DD ML FF estimator<br />

The causal restriction introduced in the previous section <strong>de</strong>aling with DD<br />

ML FB estimators applies more forcibly in the FF case, for both SU and<br />

MU estimators. In<strong>de</strong>ed, since the DD ML FF estimator is based on fedback<br />

<strong>de</strong>cisions produced downstream in the communication chain after phase<br />

correction (see Figure 2.14 b), its estimate is, at first sight, constrained to<br />

use past <strong>de</strong>cisions only. Hence, the SU DD ML FF phase estimator writes<br />

ˆφ m u<br />

= arg<br />

m 1<br />

<br />

n=m N<br />

⋆ n<br />

Îu y n <br />

u . (5.16)<br />

Obviously, an initial phase estimate obtained in DA mo<strong>de</strong> is requested in<br />

or<strong>de</strong>r to start the whole process. With the help of this initial estimate, the<br />

phase of matched filter outputs is corrected and <strong>de</strong>cisions based on these<br />

corrected outputs are obtained. These <strong>de</strong>cisions are then exploited by the<br />

DD phase estimator to compute the first DD estimate.<br />

Several options are possible as far as the update rhythm is concerned. In<br />

the fastest mo<strong>de</strong> (Figure 5.23), a new phase estimate is computed as soon<br />

as a new <strong>de</strong>cision is produced. This new <strong>de</strong>cision is combined with N 1<br />

past ones in or<strong>de</strong>r to compute the updated estimate according to (5.16).<br />

On the other hand, slower mo<strong>de</strong>s can be consi<strong>de</strong>red, in which the phase<br />

e j ˆ φ 0 u<br />

m N yu m N Îu m N+1 yu m N+1 Îu m 1 yu m 1 Îu ˆφ<br />

e j ˆ φ m u<br />

y m u<br />

Î m u<br />

ˆφ<br />

y m+1<br />

u<br />

e j ˆ φ m+1<br />

u<br />

Î m+1<br />

u<br />

m+N 1 yu m+N 1 Îu Figure 5.23: SU DD ML FF estimator - Fastest update implementation<br />

estimate is applied to several (at most N) successive matched filter outputs<br />

(Figure 5.24). Once the <strong>de</strong>cisions based on these phase-corrected matched<br />

filter outputs are produced, an updated estimate of the phase is computed.<br />

With respect to the previous phase estimate, the slowly updated one is<br />

mostly based on brand new <strong>de</strong>cisions, while the fastest estimator <strong>de</strong>scribed<br />

here above recycles N 1 past <strong>de</strong>cisions. Notice also that the slow-


144 Decision Directed<br />

e j ˆ φ 0 u<br />

m N yu m N Îu m N+1 yu m N+1 Îu m 1 yu m 1 Îu ˆφ<br />

e j ˆ φ m u<br />

y m u<br />

Î m u<br />

y m+1<br />

u<br />

Î m+1<br />

u<br />

m+N 1 yu m+N 1 Îu ˆφ e j ˆ φ m+N<br />

u<br />

Figure 5.24: SU DD ML FF estimator - Slow update implementation<br />

est estimation mo<strong>de</strong> nicely applies to a burst transmission scenario. In<br />

fact, the choice of an update mo<strong>de</strong> is a tra<strong>de</strong>-off b<strong>et</strong>ween the computational<br />

load and the adaptability of the phase estimate. The slowest solution<br />

is <strong>de</strong>finitely the best one from the point of view of the computational<br />

load, since it requires up to N times less operations than the fastest one.<br />

However, it assumes that the phase to be estimated does not significantly<br />

change over the time span of the treated symbol block. If it does, wrong<br />

<strong>de</strong>cisions will be produced due to an erroneous phase estimate, and the<br />

whole reception process will collapse. The fastest estimator makes no such<br />

assumptions but implies much more computation.<br />

5.2.2 MU DD ML FF estimator<br />

From a multiuser perspective, the constraint to rely on available <strong>de</strong>cisions<br />

applies not only to the single-user part of the estimate, as studied in the<br />

previous paragraph, but also to the MAI mitigation term introduced in<br />

(4.26). Moreover, since the mitigation part uses phase estimates related to<br />

the interfering users, the estimator is also constrained to rely on the latest<br />

phase estimates. Hence, these restrictions can lead to two different implementations.<br />

A first option consists in using past <strong>de</strong>cisions and past estimates for all<br />

users. The MU estimator writes<br />

ˆφ m u = tan 1 (Cm u )<br />

(Cm u ) = arg (Cm u ) (5.17)


5.2 Feedforward 145<br />

where<br />

C m u<br />

=<br />

m 1<br />

n=m N<br />

Î n u<br />

⋆<br />

y n u<br />

Nu <br />

Ek<br />

k=1<br />

k=u<br />

Eu<br />

e j ˆm 1<br />

φk m 1<br />

n=m N p=<br />

m 1<br />

Î n u<br />

⋆ p p<br />

Îk xn<br />

u,k<br />

(5.18)<br />

This is the parallel implementation, since all users are simultaneously <strong>de</strong>alt<br />

with. A 2-user version is shown in Figure 5.25. With respect to the DA<br />

case, the MAI mitigation is incompl<strong>et</strong>e, because it is strictly limited to its<br />

causal part (p


146 Decision Directed<br />

related to weaker users. (4.26) writes then<br />

C m u<br />

=<br />

m 1<br />

Î n u<br />

⋆<br />

y n u<br />

n=m N<br />

Nu <br />

Ek<br />

e<br />

Eu<br />

k=1<br />

ku<br />

j ˆ m<br />

m 1 1<br />

m 1<br />

φl n=m N p=<br />

e j ˆ φ 0 u ˆ φ ˆ φ<br />

e j ˆ φ 0 v<br />

m N yu m N Îu m N Îv m N yv m N+1 yu m N+1 Îu m N+1 Îv m N+1 yv m 1 yu m 1 Îu m 1 Îv m 1 yv ˆφ<br />

y m u<br />

Î m u<br />

e j ˆ φ m+1<br />

u<br />

Î m v<br />

y m v<br />

ˆφ<br />

e j ˆ φ m+1<br />

u<br />

Î n u<br />

Î n u<br />

y m+1<br />

u<br />

Î m+1<br />

u<br />

Î m+1<br />

v<br />

e j ˆ φ m+1<br />

v<br />

y m+1<br />

v<br />

⋆ p p<br />

Îk xn<br />

u,k<br />

⋆ Î p<br />

l<br />

xn p<br />

u,l<br />

Figure 5.26: 2-user successive MU DD ML FF estimator<br />

m+N 1 yu m+N 1 Îu m+N 1 Îv m+N 1 yv (5.19)<br />

The successive implementation uses the most up-to-date information in its<br />

reception process. As a result, the MAI mitigation is slightly b<strong>et</strong>ter than in<br />

the parallel implementation in as much as the MAI related to the current<br />

symbols (p = m) of the more powerful users is also mitigated. The price<br />

to pay is a <strong>de</strong>lay on the reception of weak users. This <strong>de</strong>lay is as great as<br />

the user is weak, due to the fact that the estimator waits for new <strong>de</strong>cisions<br />

and updated estimates from more powerful users.


5.2 Feedforward 147<br />

Finally, notice that the <strong>de</strong>sign option discussed in the SU case (fast vs. slow<br />

update) is also applicable in the MU case.<br />

5.2.3 Decisions assumed correct<br />

Computational results are presented in the following pages. They have<br />

been obtained un<strong>de</strong>r the assumption that <strong>de</strong>cisions and phase estimates<br />

were correct. In that situation, the only difference b<strong>et</strong>ween DA and DD<br />

implementations lies in the fact that the mitigation performed in MU DD<br />

structures is limited to the causal part of the message, as mentioned in the<br />

previous paragraph. With this restriction, the following results are just a<br />

reinterpr<strong>et</strong>ation of DA results of Chapter 4.<br />

Introducing the causal restriction in (4.31) leads to the following MU DD<br />

ML FF estimator<br />

∆u =<br />

<br />

ISIu + MAIa <br />

v,u + Noiseu (Directv + ISIv)<br />

MAIc <br />

<br />

v,u (ISIv + Noisev)<br />

<br />

(Directu + ISIu) (Directv + ISIv)<br />

MAIc <br />

<br />

v,u MAIc u,v<br />

(5.20)<br />

Comparing to its DA counterpart (4.32), the rea<strong>de</strong>r can notice that the MAI<br />

contribution in (4.32) is now split into its causal MAI c v,u<br />

MAI c u,v =<br />

Ev<br />

Eu<br />

and anti-causal MAI a v,u parts<br />

MAI a v,u =<br />

Ev<br />

Eu<br />

j(φv φu)<br />

e<br />

j(φv φu)<br />

e<br />

N<br />

m=1 n=<br />

N<br />

m<br />

+<br />

m=1 n=m+1<br />

(I m u ) ⋆ I n m n<br />

v xu,v (I m u ) ⋆ I n n m<br />

v xv,u (5.21)<br />

(5.22)<br />

with respect to time in<strong>de</strong>x m. The anti-causal part of the MAI contributes<br />

to the biasing term (second numerator term of (5.20)) already mentioned<br />

in Section 4.2.2 in the case of ISI.<br />

From the point of view of the variance, the split of the MAI into its causal<br />

and anti-causal part adds new terms to the expressions presented in Appendix<br />

C. A simplified version is obtained by limiting the study to the


148 Decision Directed<br />

most important one, MAIa <br />

v,u (Directv). Its inci<strong>de</strong>nce in the variance<br />

expressions of Appendix C is a supplementary term whose numerator<br />

writes, <strong>de</strong>pending on the modulation,<br />

¯ BPSK modulation<br />

<br />

<br />

N<br />

0 2<br />

Nxu,u 4 Ev<br />

Eu<br />

¯ QPSK modulation<br />

+<br />

m=1 n=m+1<br />

4 Ev<br />

Eu<br />

x <br />

n m<br />

v,u<br />

0 2 N<br />

Nxu,u 2 <br />

+<br />

m=1 n=m+1<br />

<br />

e 2jδu,v x<br />

n m<br />

v,u<br />

2 <br />

(5.23)<br />

<br />

n m<br />

x 2 . (5.24)<br />

Its <strong>de</strong>nominator is given either by (C.7) for BPSK- or by (C.11) for QPSKmodulated<br />

symbols.<br />

Both expressions (5.23) and (5.24) induce a variance increase in<strong>de</strong>pen<strong>de</strong>nt<br />

of the Es ratio whose importance <strong>de</strong>pends on the relative weight of the<br />

N0<br />

unmitigated anti-causal part of the MAI. This variance increase graphically<br />

translates into a raise of the variance floor (See Figure 5.27, curve ”DD<br />

with”). The variance is even greater if the zero-shift contribution x0 u,v is not<br />

taken into account (curve ”DD without”), as already explained in Section<br />

5.1.1. However, in dispersive environments such as the one consi<strong>de</strong>red in<br />

Figure 5.27, a (slight) improvement with respect to the SU estimator is still<br />

noticeable thanks to the mitigation of the MAI due to past symbols.<br />

5.2.4 Actual <strong>de</strong>cisions<br />

No study similar to the one performed in DA phase estimation (Section<br />

4.2) has been ma<strong>de</strong> for its DD counterpart. Such analysis would have<br />

been pr<strong>et</strong>ty difficult, since it would have required to <strong>de</strong>al with the two<br />

different kinds of coupling, namely the coupling b<strong>et</strong>ween users and the<br />

coupling b<strong>et</strong>ween estimation and <strong>de</strong>tection stages. Since the resolution of<br />

the coupling b<strong>et</strong>ween users in the DA context already leads to simplifications,<br />

it could be expected that the study of the DD context would require<br />

stronger approximations.<br />

v,u


5.3 Conclusions 149<br />

Variance [rad 2 ]<br />

10 1<br />

2−user system − 31−chip Gold co<strong>de</strong>s − BPSK modulation − R = 1e5 Bauds − HT channel − 2 B T = 0.1<br />

N<br />

10 0<br />

10 −1<br />

10 −2<br />

10 −3<br />

10 −4<br />

10 −5<br />

DA<br />

DD with<br />

DD without<br />

Single−user<br />

Multiuser<br />

Uniform distribution<br />

CRLB<br />

10<br />

0 5 10 15 20 25 30 35 40<br />

−6<br />

E /N [dB]<br />

s 0<br />

Figure 5.27: Variance of ML FF estimators in presence of ISI (BSPK)<br />

On the other hand, as mentioned in Section 2.3.2, the study of DD ML FF<br />

phase estimators is usually not performed as is in the estimation literature.<br />

One relies on a correspon<strong>de</strong>nce b<strong>et</strong>ween FF and FB estimators [84, p. II-<br />

3] to extend the performance <strong>de</strong>rived for FB estimators to FF estimators.<br />

Such extension is only valid as long as some conditions regarding the error<br />

<strong>de</strong>tector and the closed-loop impulse response are respected [85, p. 344].<br />

5.3 Conclusions<br />

The study of MU DD estimators has been performed in this chapter by<br />

following two different tracks.<br />

The first one was based on the assumptions of correct <strong>de</strong>cisions. Introducing<br />

the causal restriction of DD estimators into DA relations <strong>de</strong>rived in<br />

Chapter 4, it has led to a reinterpr<strong>et</strong>ation of these relations in a DD perspective.<br />

It has been shown that the variance of the MU DD estimator is<br />

greater than, or at best equal to the one of the MU DA estimator, but lower<br />

than, or at worst equal to the one of the SU estimator. The quality of the<br />

estimator mainly <strong>de</strong>pends on the way it <strong>de</strong>als with the interference related<br />

to the present symbol.


150 Decision Directed<br />

The second track ma<strong>de</strong> no assumption with respect to the correctness of<br />

the <strong>de</strong>cisions. It has focused on the open-loop study of a MU DD phase<br />

recovery loop. Several m<strong>et</strong>hods for <strong>de</strong>riving the mean of its error signal<br />

have been presented, namely a bidimensional brute-force m<strong>et</strong>hod in the<br />

direct space and a multidimensional one in the reciprocal space. Using<br />

the first one, a graphical representation of the mean has been illustrated.<br />

In the consi<strong>de</strong>red 2-user system, it stands as the bidimensional extension<br />

of well-known S-curves, therefore called S-surfaces. For systems encompassing<br />

more than 2 users, one speaks of S-hypersurfaces.<br />

On the other hand, the closed-loop study of the MU DD recovery loop has<br />

been initiated. It has been <strong>de</strong>monstrated that its structure could be seen as<br />

a s<strong>et</strong> of loops whose error signals are function of all estimation errors.


Chapter 6<br />

Conclusions<br />

6.1 Achievements<br />

This thesis has tackled the param<strong>et</strong>er estimation issue in a multiuser spread-spectrum<br />

communication system. The following table summarises<br />

the options consi<strong>de</strong>red as far as the interaction b<strong>et</strong>ween estimator and <strong>de</strong>tector<br />

(DA/DD/NDA) and the structure of the estimator (FB/FF) are concerned.<br />

The <strong>de</strong>gree of achievement within each option is also mentioned.<br />

Estimator Performance study<br />

DA:<br />

<br />

minθ Î,θ<br />

DD:<br />

Λ(I,θ)<br />

<br />

Λ Î,θ<br />

FB<br />

FF<br />

FB<br />

Open-loop Closed-loop<br />

Closed-form solution<br />

Decisions assumed correct<br />

minθ (I,θ) Open-loop<br />

FF Decisions assumed correct<br />

NDA: ¯ Λ(θ) FB<br />

minθ EI [ (I,θ)] FF<br />

Table 6.1: Synth<strong>et</strong>ic view of the achievements of the thesis<br />

Regarding the param<strong>et</strong>er to be estimated as an uniformly distributed random<br />

variable, the log-likelihood function has been <strong>de</strong>rived in Chapter 3 as<br />

the cost function to be minimised by the optimal estimator, wh<strong>et</strong>her DA,<br />

DD, or NDA. In the case of DA and DD estimations, it has been shown<br />

that taking into account the conditional signal energy in the writing of the<br />

log-likelihood function leads to the addition of a supplementary term in


152 Conclusions<br />

the expression of the estimators, besi<strong>de</strong> the well-known contribution <strong>de</strong>pending<br />

on the correlation b<strong>et</strong>ween symbols and phase-corrected matched<br />

filter outputs. Usually, the conditional signal energy is disregar<strong>de</strong>d in the<br />

literature as being in<strong>de</strong>pen<strong>de</strong>nt of the param<strong>et</strong>er to estimate. This is valid<br />

in SU environments but not in MU ones. This observation has been the<br />

starting point of this thesis. As a result, the supplementary contribution<br />

appearing in the expressions of the estimators has been shown afterwards<br />

to mitigate the inci<strong>de</strong>nce of the MAI entering the system through the matched<br />

filter outputs.<br />

Moving to implementations, two MU DA ML phase estimators, one FB<br />

and one FF, have been first <strong>de</strong>scribed in Chapter 4. Their performance<br />

have been <strong>de</strong>rived analytically and compared to those of their SU counterparts<br />

working in the same multiuser environment. The mitigation effect<br />

mentioned here above leads to performance improvement with respect to<br />

SU estimators: MU DA ML phase estimators exhibit less jitter variance<br />

than their SU counterparts. Furthermore, they reach the CRLB and appear<br />

to be Near-Far resistant. However, they suffer from ISI.<br />

Following the analysis of DA estimators, the study of DD structures presented<br />

in Chapter 5 has been twofold. On the one hand, assuming correct<br />

<strong>de</strong>cisions, the <strong>de</strong>velopments led for DA estimators have been reinterpr<strong>et</strong>ed<br />

in a DD perspective. On the other hand, relaxing assumptions<br />

on the quality of the <strong>de</strong>cisions, this work focused on FB implementations<br />

(recovery loops). The investigations have been limited to the open-loop<br />

study. S-hypersurfaces have been introduced as multi-dimensional extensions<br />

of well-known S-curves. Moreover, the results presented in [87] for<br />

a single-user case and AWGN channels have been generalised for a multiuser<br />

case and possibly frequency-selective channels.<br />

For all the investigated estimators, the performance study has been performed<br />

analytically. To the knowledge of the author, this is the first analytical<br />

<strong>de</strong>monstration of the efficiency of multiuser param<strong>et</strong>er estimation<br />

in spread-spectrum environments.<br />

6.2 Perspectives<br />

This thesis does not claim, in any way, to have covered the subject in its<br />

entir<strong>et</strong>y. There remains several issues that can be used as starting points


6.2 Perspectives 153<br />

and/or central themes for future studies. They shall be the subject of this<br />

concluding section.<br />

Phase mo<strong>de</strong>l<br />

The first action point is related to the param<strong>et</strong>er at the centre of this thesis.<br />

In the present work, the phase has been mo<strong>de</strong>lled as constant during the<br />

estimation. However, it might be mo<strong>de</strong>lled as a slowly changing param<strong>et</strong>er<br />

whose fluctuations are characterised by the phase noise. The filtering<br />

of this noise through the estimation <strong>de</strong>vice affects the performance of<br />

the estimator [84, part IV]. The inci<strong>de</strong>nce of the phase noise should thus<br />

be taken into account in future <strong>de</strong>velopments.<br />

Close to the concern of phase noise, a specific fluctuation of the phase<br />

would be of great interest, namely its continuous raise due to the integration<br />

over time of a frequency offs<strong>et</strong>. This issue is of crucial importance for<br />

coherent transmissions.<br />

Validity of working hypotheses<br />

The estimators presented in this work have been <strong>de</strong>rived un<strong>de</strong>r the assumption<br />

that all other param<strong>et</strong>ers of the communication system (timing,<br />

power, channel impulse response, <strong>et</strong>c.) either were known or had been<br />

perfectly recovered. There is very little knowledge in as much stringent<br />

these hypotheses are. As reviewed in Section 2.3.3, there are numerous<br />

references regarding the inci<strong>de</strong>nce of estimation errors on the <strong>de</strong>tection<br />

performance. However, the issue of imperfect recovery of one param<strong>et</strong>er<br />

on the estimation of another one has not received much attention. The<br />

sensibility of the proposed estimators to incorrect values of those param<strong>et</strong>ers<br />

ought thus to be studied in or<strong>de</strong>r to g<strong>et</strong> a b<strong>et</strong>ter view on the working<br />

of the proposed structures.<br />

Generalisation to other param<strong>et</strong>ers - Joint 2 estimation<br />

The previous action point stressed the importance of the other param<strong>et</strong>ers<br />

of the link. In<strong>de</strong>ed, beyond the sensibility study consi<strong>de</strong>red, the scope of<br />

the present work might be broa<strong>de</strong>ned so as to encompass all those param<strong>et</strong>ers.<br />

The phase is <strong>de</strong>finitely not the only param<strong>et</strong>er to estimate. From<br />

an analytical point of view, and interestingly enough, it is characterised by<br />

the fact that it stands explicitly through a phasor in relations <strong>de</strong>fining for


154 Conclusions<br />

instance the matched filter output. On the other hand, other param<strong>et</strong>ers,<br />

such as the timing, are implicit. They appear in these relations through<br />

non-linear functions, which makes their estimation more complex. However,<br />

recovering them is of crucial interest, as for instance timing in the<br />

case of spread-spectrum receivers. Hence, investigating the multiuser estimation<br />

of other param<strong>et</strong>ers is most probably the next step ahead.<br />

Furthermore, it is obvious that the estimation of any param<strong>et</strong>er is connected<br />

to the recovery of the other ones. It might thus be more straightforward<br />

to tackle the problem in its globality from the very beginning, that<br />

is to say <strong>de</strong>sign a param<strong>et</strong>er estimator which simultaneously addresses all<br />

param<strong>et</strong>ers of the link for all active users. This would lead to a ”joint 2 ” estimation<br />

process, the exponent indicating that the joint characteristic has a<br />

double meaning: for each param<strong>et</strong>er, this global estimator would take all<br />

active users into account and, for each user, it would attempt to recover all<br />

the param<strong>et</strong>ers of this link.<br />

Coupled structures for DD estimation<br />

Figure 5.22 can be seen both as an achievement and as a starting point.<br />

On the one hand, it is an achievement in that the means of coupling in an<br />

MU DD ML FB phase estimator have been ma<strong>de</strong> obvious. Y<strong>et</strong>, it is also a<br />

starting point, as all elements for further study are available because the<br />

expression of U BPSK<br />

u,DD has been established. Further investigations of such<br />

MU DD phase recovery loops will <strong>de</strong>finitely lead to the study of strongly<br />

coupled feedback systems.<br />

NDA estimation<br />

The closed-loop study of DD estimators mentioned here above would conclu<strong>de</strong><br />

the treatment of DD structures. However, even with this ad<strong>de</strong>ndum,<br />

the review of possible estimation structures would still miss NDA estimators.<br />

This lack ought to be filled through a <strong>de</strong>dicated analysis of NDA structures,<br />

whose main property is to be in<strong>de</strong>pen<strong>de</strong>nt of the <strong>de</strong>tection stage.<br />

Monte-Carlo simulations for validating calculations and investigating<br />

dynamic phenomena<br />

The analytical <strong>de</strong>velopments presented in this thesis are interesting in that<br />

they give a precise insight in the ins and outs of the param<strong>et</strong>er estima-


6.2 Perspectives 155<br />

tion process in a multiuser spread-spectrum environment. However, the<br />

confirmation of the calculations through Monte-Carlo simulations would<br />

ground more strongly the conclusions drawn in the previous section. Moreover,<br />

Monte-Carlo simulations are not just a tool for validating calculations.<br />

They also enable to investigate dynamic behaviours, such as the<br />

estimator’s one, and to <strong>de</strong>al with time-varying channels.<br />

As far as the estimator is concerned, the investigation performed during<br />

this thesis has focused on its steady-state performance. Y<strong>et</strong>, as was<br />

stressed in Section 2.3.2, the dynamic behaviour (acquisition, cycle slips,<br />

<strong>et</strong>c.) of the MU estimator also <strong>de</strong>serves much attention. It ought thus to<br />

be investigated in or<strong>de</strong>r to g<strong>et</strong> a really compl<strong>et</strong>e view of its performance.<br />

On the other hand, the channels used throughout this work have been<br />

regar<strong>de</strong>d as static, although the thesis took place in a mobile radio environment.<br />

The time-varying characteristic of the channels have thus not<br />

been taken into account. Hence, it would be more realistic to investigate<br />

the behaviour of the presented estimators in such channels. Monte-Carlo<br />

simulations are a powerful tool in that perspective.<br />

Gaussian issues<br />

Section 2.3.3 stressed that the performance study of <strong>de</strong>vices operating in<br />

MAI-plagued environments often relied on a Gaussian mo<strong>de</strong>l of this interference.<br />

This thesis has taken the exact opposite view in that it <strong>de</strong>alt with<br />

the actual MAI throughout all <strong>de</strong>velopments. However, it might be worth<br />

raising the question of the efficiency of this approach. In<strong>de</strong>ed, the use of<br />

the Gaussian approximation which mo<strong>de</strong>ls the effect of a large number of<br />

in<strong>de</strong>pen<strong>de</strong>nt random sources, implies that the load of the system, i.e. the<br />

number of active users, is large. As a result, the exhaustive approach followed<br />

in this thesis should apply for small to mo<strong>de</strong>rate system loads but<br />

might be outperformed by the Gaussian approximation from the point of<br />

view of calculation complexity when the load becomes heavy. An interesting<br />

issue is the limit of the system load at which it becomes more efficient<br />

to change of MAI mo<strong>de</strong>l. Some preliminary results were already presented<br />

in [89]. However, it would <strong>de</strong>serve <strong>de</strong>eper investigation.<br />

The other issue related to Gaussian matters refers to the <strong>de</strong>velopments<br />

presented in [107]. It is clear from Appendix E that analytical <strong>de</strong>velopments<br />

related to DD structures often involve the calculation of a probab-


156 Conclusions<br />

ility of the type Pr (A >0), where A is a random variable. In<strong>de</strong>ed, A is a<br />

linear combination of several random variables and of Gaussian noise. Assuming<br />

the other variables, the random nature of A only comes from the<br />

noise. As a result, the probability mentioned here above can be obtained<br />

assuming the other random variables, as a Q function whose argument<br />

<strong>de</strong>pends on the assumed variables. Compl<strong>et</strong>ing the calculation requires to<br />

average this expression over the joint pdf of the assumed variables. However,<br />

the Q function that has appeared due to the Gaussian random variable<br />

is not really well suited for such a <strong>de</strong>rivation. Y<strong>et</strong>, another writing of<br />

the Q function has been proposed [107] (See Section 3.5.1), which might<br />

help to solve this calculation issue.<br />

MC-CDMA<br />

The question of coherent reception is the last key-issue for future works to<br />

mention. Now topical for third-generation <strong>de</strong>velopments, it is also of crucial<br />

importance for another kind of air interface, viz. OFDM systems. In<br />

the last few years, an hybrid air interface, MC-CDMA, combining OFDM<br />

and CDMA, has attracted much interest. It would be worth extending the<br />

study led in this thesis for single-carrier CDMA to MC-CDMA systems.


Appendix A<br />

Correlation function of the<br />

loop noise in a DA recovery<br />

loop<br />

General expressions of the cross-correlation function C m n<br />

u,v (∆) of two loop<br />

noise samples ξ m u and ξ n v are given in this appendix as a function of the<br />

vector estimation error ∆=Φ ˆΦ.


158 Correlation function of the loop noise in a DA recovery loop<br />

A.1 BPSK modulation<br />

After long but easy calculations, Cm n<br />

u,v (∆) writes, in the case of BPSKmodulated<br />

data symbols<br />

m n<br />

Cu,v (∆)<br />

= E [ξ m u ξn v ]<br />

= δ (u<br />

⎧<br />

v)<br />

⎪⎨<br />

⎧<br />

⎪⎨<br />

δ (m n)<br />

+<br />

p=<br />

+ N0x0 u,u<br />

2EuT<br />

+ Nu Ek<br />

Eu<br />

k=1 p=<br />

k=u<br />

+ Nu Ek<br />

Eu<br />

k=1 p=<br />

k=u<br />

<br />

j∆u p 2 e xu,u +<br />

+<br />

<br />

e j(δk,u+∆u)<br />

2 p<br />

xu,k <br />

e j(δk,u+∆u<br />

2 ∆k) p<br />

xu,k ⎧<br />

⎨<br />

<br />

e j(δk,u+∆u)<br />

<br />

p<br />

x<br />

<br />

u,k<br />

e j(δk,u+∆u<br />

<br />

∆k) p<br />

xu,k ⎪⎩<br />

2<br />

⎪⎩<br />

Nu + Ek<br />

Eu<br />

k=1 p= ⎩<br />

k=u<br />

+ ej∆u <br />

xm n j∆u<br />

u,u e xn m<br />

u,u<br />

+[1 δ (u v)]<br />

⎧<br />

<br />

⎪⎨<br />

⎪⎩<br />

ej(δv,u+∆u) <br />

xm n<br />

u,v ej(δu,v+∆v) xn m<br />

v,u<br />

ej(δv,u+∆u) <br />

xm n j(δu,v+∆v<br />

u,v e ∆u) xn m<br />

v,u<br />

ej(δu,v+∆v) <br />

xn m j(δv,u+∆u<br />

v,u e ∆v) xm n<br />

u,v<br />

ej(δv,u+∆u ∆v) ⎫<br />

⎪⎬<br />

.<br />

<br />

xm n 2<br />

⎪⎭<br />

u,v<br />

⎫<br />

⎬<br />

⎭<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

(A.1)


A.2 QPSK modulation 159<br />

A.2 QPSK modulation<br />

Likewise, with QPSK-modulated data symbols,<br />

m n<br />

Cu,v (∆)<br />

= E [ξ m u (ξn v )⋆ ]<br />

⎧<br />

⎡<br />

⎢<br />

⎪⎨<br />

⎢<br />

δ (m n) ⎢<br />

= δ (u v)<br />

⎢<br />

⎣<br />

1<br />

2<br />

p=<br />

+<br />

x p u,u 2<br />

+ N0x0 u,u<br />

2EuT<br />

+ 1<br />

Nu + Ek<br />

2 Eu<br />

k=1 p=<br />

k=u<br />

<br />

<br />

x p<br />

<br />

<br />

u,k<br />

2<br />

+ 1<br />

Nu + Ek<br />

2 Eu<br />

k=1 p=<br />

k=u<br />

<br />

<br />

x p<br />

<br />

<br />

u,k<br />

2<br />

Nu Ek<br />

Eu<br />

k=1<br />

k=u<br />

+<br />

cos ∆k<br />

p=<br />

⎪⎩ 1<br />

<br />

<br />

2 xm n<br />

u,u<br />

2 cos 2∆u<br />

+ 1<br />

⎡ <br />

xm n<br />

u,v<br />

⎢<br />

[1 δ (u v)] ⎢<br />

2 ⎣<br />

2 cos (∆u +∆v)<br />

+ <br />

xm n<br />

u,v<br />

2 cos ∆u<br />

+ <br />

xm n<br />

u,v<br />

2 cos ∆v<br />

<br />

xm n2<br />

u,v<br />

<br />

<br />

x p<br />

<br />

<br />

u,k<br />

2<br />

⎤ ⎫<br />

⎥ ⎪⎬<br />

⎥<br />

⎦<br />

⎪⎭<br />

⎤<br />

⎥ . (A.2)<br />


Appendix B<br />

Pdf of Single-User DA ML FF<br />

phase estimator<br />

In this appendix, the pdf of the Single-User DA ML FF phase estimator<br />

given by [7, p. 326] is <strong>de</strong>rived analytically in a multiuser context. Calculations<br />

are simplified by restricting the study to BPSK modulated data<br />

symbols and by using a single-tap averaging window (N =1). For <strong>de</strong>tection<br />

[38] as well as for param<strong>et</strong>er estimation, such one-shot approach ends<br />

in a sub-optimal implementation in the case of asynchronous signals.<br />

B.1 First step: characteristic function ψˆxu,ˆyu (ωr,ωi)<br />

Using (3.77) and (3.78), the characteristic function writes<br />

ψˆxu,ˆyu (ωr,ωi)<br />

⎧ ⎧ ⎧ ⎡<br />

Nu<br />

N<br />

+ Ek<br />

ωr ⎣ Eu<br />

⎪⎨ ⎪⎨ ⎪⎨<br />

k=1 n=<br />

m=1<br />

+I<br />

= E exp j<br />

⎪⎩ ⎪⎩ ⎪⎩<br />

m u νm u e jφu<br />

⎡<br />

Nu<br />

N<br />

+ Ek<br />

+ωi ⎣ Eu<br />

k=1 n=<br />

m=1<br />

+Im u νm u e jφu<br />

Im u In n<br />

k<br />

Rm<br />

u,k<br />

Im u In n<br />

k<br />

Im<br />

u,k<br />

⎤<br />

⎦<br />

⎤<br />

⎦<br />

⎫⎫⎫<br />

⎪⎬ ⎪⎬ ⎪⎬<br />

⎪⎭ ⎪⎭ ⎪⎭<br />

(B.1)<br />

Developing the expectation in (B.1) is pr<strong>et</strong>ty intricate due to the ISI terms<br />

E (Im u In u ). As a result of their presence, calculating the expectation over<br />

all data symbols progressively, assuming one data symbol after the other,


162 Pdf of Single-User DA ML FF phase estimator<br />

generates nested functions which are not easy to handle. However, a simplifying<br />

hypothesis, s<strong>et</strong>ting the span N equal to 1, enables to <strong>de</strong>rive these<br />

expectations without too much effort. Such narrow averaging window<br />

is unfortunately not realistic for practical implementations. The present<br />

choice should then be seen as a tra<strong>de</strong>-off b<strong>et</strong>ween the will to reach an analytical<br />

result and the tolerable complexity of the <strong>de</strong>rivations.<br />

With this hypothesis, (B.1) becomes<br />

ψˆxu,ˆyu (ωr,ωi)<br />

<br />

= exp<br />

+ <br />

n=<br />

n=0<br />

N0x 0 u,u<br />

4EuT<br />

2<br />

ωr + ω 2 0<br />

i + jxu,uωr cos R n u,uωr + I n u,uωi Nu <br />

k=1<br />

k=u<br />

n=<br />

+ <br />

<br />

cos R n u,kωr + I n u,kωi <br />

.<br />

(B.2)<br />

On the other hand, neglecting the ISI terms enables to avoid the nested<br />

functions issue mentioned here above. It is thus possible to write the characteristic<br />

function for widths N of the averaging window greater than 1.<br />

It can easily be shown that the characteristic function obtained in a case<br />

where N > 1 is the N-th power of the characteristic function obtained<br />

assuming N =1.<br />

B.2 Second step: pdf Tˆxu,ˆyu (ˆxu, ˆyu)<br />

The product of cosines functions in (B.2) does not facilitate the search for<br />

an analytical solution of the inverse Fourier transform. However, such<br />

closed form solution is within reach if this product can be written as a<br />

sum, like in the following<br />

where<br />

N<br />

cos (dkω) =<br />

k=1<br />

Dl =<br />

q=1<br />

1<br />

2 (N 1)<br />

2 (N 1)<br />

<br />

l=1<br />

N<br />

<br />

(l 1) mod 2<br />

1 2<br />

[cos (Dlω) sin (Dlω)] (B.3)<br />

2 (N q)<br />

(N+1 q)<br />

<br />

dq<br />

(B.4)


B.2 Second step: pdf Tˆxu,ˆyu (ˆxu, ˆyu) 163<br />

with x being the greatest integer value lower than or equal to x.<br />

This trick is very helpful to turn the integration of a product into a sum of<br />

integrations. However, it has a major drawback. It was stressed in Section<br />

3.5.2 that the <strong>de</strong>rivation in the reciprocal space, using the characteristic<br />

function, helped to reduce the number of computations from exponential<br />

down to linear complexity. Expanding the product as in (B.3) cancels this<br />

advantage, since the final result now exhibits a sum whose span enlarges<br />

exponentially with the number of symbols contributing to the interference.<br />

In<strong>de</strong>ed, this sum performs the averaging operation over all possible outcomes.<br />

Despite this loss of performance, the <strong>de</strong>rivation can and will go on.<br />

Defining Sx as the time span of the normalised channel correlation coefficient,<br />

that is to say the number of non-zero coefficients x q<br />

k,l for a pair of<br />

users (k, l), dq in (B.4), will be in the present case the elements of Nu ¢ Sx<br />

vectors ¯ R p u and Īp u so that<br />

¯R p u =<br />

p %Sx<br />

Ru, p<br />

Sx<br />

Ī p u =<br />

p %Sx<br />

Iu, p<br />

Sx<br />

(B.5)<br />

(B.6)<br />

where % stands for the modulo operator and p [1,NuSx] . Using (B.3)<br />

and [118, p. 15, relation (11)], Tˆxu,ˆyu (ˆxu, ˆyu) writes<br />

Tˆxu,ˆyu (ˆxu, ˆyu)<br />

+ 2<br />

1<br />

=<br />

2π<br />

=<br />

+<br />

1<br />

2 (NuSx+1) cuπ<br />

⎧<br />

2 (NuSx 2)<br />

<br />

k=1<br />

⎪⎨<br />

⎪⎩<br />

exp<br />

ψˆxu,ˆyu (ωr,ωi) e j(ωrxu+ωiyu) dωrdωi<br />

⎧<br />

⎪⎨<br />

+exp<br />

⎪⎩ 1<br />

4cu<br />

⎧<br />

⎪⎨<br />

⎪⎩ 1<br />

4cu<br />

(B.7)<br />

⎡ <br />

x2 u + y<br />

⎢<br />

⎣<br />

2 <br />

u +2xu F k<br />

u x0 <br />

u,u<br />

+2yuGk u + x0 2 u,u 2F k<br />

u x0 u,u<br />

+ F k ⎤⎫<br />

⎪⎬<br />

⎥<br />

⎦<br />

2 <br />

u + Gk 2<br />

⎪⎭<br />

⎡ u<br />

x2 u + y<br />

⎢<br />

⎣<br />

2 <br />

u 2xu F k<br />

u + x0 <br />

u,u<br />

2yuGk u + x0 2 u,u +2Fk u x0 u,u<br />

+ F k ⎫<br />

⎪⎬<br />

⎤⎫<br />

⎪⎬<br />

⎥<br />

⎦<br />

2 <br />

u + Gk 2<br />

⎪⎭ ⎪⎭<br />

u<br />

(B.8)


164 Pdf of Single-User DA ML FF phase estimator<br />

where<br />

F k u =<br />

G k u =<br />

cu = N0x 0 u,u<br />

4EuT =<br />

<br />

x0 2<br />

u,u<br />

N<br />

<br />

(l 1) mod 2<br />

1 2<br />

q=1<br />

4<br />

2 (N q)<br />

N<br />

<br />

(l 1) mod 2<br />

1 2<br />

q=1<br />

2 (N q)<br />

1<br />

Es<br />

N0<br />

(N+1 q)<br />

(N+1 q)<br />

B.3 Third step: change of variables<br />

<br />

<br />

¯R q u<br />

(B.9)<br />

(B.10)<br />

Ī q u . (B.11)<br />

Switching from the cartesian (x, y) to the polar (r, ∆) coordinate system<br />

according to ˆxu =ˆru cos ∆u and ˆyu =ˆru sin ∆u enables us, with the help of<br />

[118, p. 146, relation (31)], to write the joint pdf Tˆru,∆u (ˆru, ∆u) as follows<br />

where<br />

Tˆru,∆u (ˆru, ∆u)<br />

= Tˆxu,ˆyu (ˆxu,<br />

<br />

<br />

ˆyu) <br />

∂ (ˆxu, ˆyu) <br />

<br />

∂<br />

(ˆru, ∆u) <br />

1<br />

=<br />

2 (NuSx+1) cuπ<br />

⎧ <br />

2 (NuSx 2)<br />

⎨<br />

g u exp ˆru exp<br />

4cu <br />

⎩ +exp ˆru exp<br />

k=1<br />

g + u<br />

4cu<br />

2f u<br />

4au ˆru<br />

<br />

1 2<br />

(ˆru) 4cu<br />

1<br />

4cu (ˆru) 2 + 2f + u<br />

4cu ˆru<br />

<br />

⎫<br />

⎬<br />

⎭<br />

(B.12)<br />

(B.13)<br />

f + <br />

u = F k u + x 0 <br />

u,u cos ∆u + G k u sin ∆u<br />

(B.14)<br />

f u =<br />

<br />

F k u x0 <br />

u,u cos ∆u + G k u sin ∆u<br />

(B.15)<br />

g + u = x 0 2 k<br />

u,u +2Fux 0 <br />

u,u + F k 2 <br />

u + G k 2 u<br />

(B.16)<br />

g u = x 0 2 k<br />

u,u 2Fu x 0 u,u +<br />

<br />

F k 2 <br />

u + G k 2 u . (B.17)


B.4 Fourth step: pdf T∆u(∆u) 165<br />

B.4 Fourth step: pdf T∆u (∆u)<br />

Integrating ˆru out produces the marginal <strong>de</strong>nsity of ∆u which is the <strong>de</strong>sired<br />

pdf<br />

=<br />

1<br />

2 (NuSx) π⎧<br />

2 (NuSx 2)<br />

<br />

k=1<br />

⎪⎨<br />

⎪⎩<br />

<br />

exp<br />

<br />

1<br />

<br />

+exp<br />

<br />

1+<br />

T∆u(∆u)<br />

+<br />

= Tˆru,∆u (ˆru, ∆u)dˆru<br />

0<br />

<br />

g u<br />

4cu<br />

<br />

π<br />

cu<br />

f u<br />

2 exp<br />

2 <br />

(f u ) f u 1 erf<br />

4cu<br />

2 Ô <br />

cu<br />

<br />

g + u<br />

4cu<br />

<br />

π<br />

cu<br />

B.5 Analytical validation<br />

<br />

f + u<br />

2 exp<br />

(f + u ) 2<br />

4cu<br />

<br />

1 erf<br />

f + u<br />

2 Ô <br />

cu<br />

<br />

⎫<br />

⎪⎬<br />

.<br />

⎪⎭<br />

(B.18)<br />

In or<strong>de</strong>r to validate (B.18), an AWGN scenario was consi<strong>de</strong>red. In such<br />

environment, F k u = Gku =0, leading to f u = f + u and g u = g+ then becomes<br />

u . T∆u(∆u)<br />

T∆u(∆u)<br />

= 1<br />

2π exp<br />

<br />

1+<br />

<br />

<br />

Eb<br />

N0<br />

<br />

π Eb<br />

cos ∆u exp<br />

N0<br />

<br />

Eb<br />

cos<br />

N0<br />

2 ∆u<br />

<br />

1+erf<br />

<br />

Eb<br />

cos ∆u<br />

N0<br />

<br />

.<br />

(B.19)<br />

which is the expression presented in [7, p. 262, relation (4.2.103)]. This<br />

expression has been <strong>de</strong>rived in a non frequency-selective environment.<br />

However, it was shown in [119] that it could be exten<strong>de</strong>d to dispersive<br />

environments by including ISI in the SNR.


Appendix C<br />

Variance of DA ML FF phase<br />

estimators<br />

This appendix presents the variance expressions of the closed-form DA<br />

ML FF phase estimators <strong>de</strong>rived in Section 4.2.2. These expressions differ<br />

from BPSK- to QPSK-modulated data symbols, since<br />

<br />

E (I m k )2<br />

<br />

= E I m k 2<br />

(C.1)<br />

= σ 2 Ik (C.2)<br />

in the case of BPSK, while, in the case of QPSK<br />

<br />

E (I m k )2 =0. (C.3)<br />

C.1 Multiuser estimator<br />

C.1.1 BPSK modulation<br />

Consi<strong>de</strong>ring first BPSK-modulated data symbols, the variance of the multiuser<br />

phase estimator given by (4.32) can be <strong>de</strong>rived according to the<br />

m<strong>et</strong>hod <strong>de</strong>scribed in Section 3.4.2. It finally writes<br />

BPSK 2<br />

σ =<br />

∆u<br />

1<br />

Es,u<br />

x<br />

N0<br />

0 u,u<br />

Num BPSK<br />

u<br />

Den BPSK<br />

u<br />

+<br />

<br />

σBPSK 2<br />

ISIu<br />

Den BPSK<br />

u<br />

(C.4)


168 Variance of DA ML FF phase estimators<br />

where the numerator Num BPSK<br />

u<br />

Num BPSK<br />

u<br />

= N 3 x0 <br />

u,u x0 2<br />

v,v<br />

+N x0 ⎡<br />

u,u<br />

N<br />

⎣ N <br />

x<br />

m=1<br />

N<br />

n=1<br />

n=m<br />

Nx0 v,v<br />

m=1 n=1<br />

N x expands into<br />

m n<br />

v,v<br />

m n<br />

u,v<br />

<br />

<br />

2 + +<br />

2 + <br />

n=<br />

n=m<br />

x m n<br />

v,v<br />

2<br />

⎤<br />

⎦<br />

<br />

e2j(φu φv) x<br />

m n<br />

u,v<br />

2 <br />

Nx0 <br />

N N <br />

v,v xn m<br />

v,u x0 u,v + x<br />

m=1 n=1<br />

0 2j(φu<br />

v,u e φv)<br />

N N N <br />

<br />

xn m p m n p<br />

v,v xu,v xv,u +2 xn m m p<br />

v,v xv,u x<br />

m=1<br />

N<br />

m=1<br />

<br />

<br />

⎡<br />

n=1<br />

n=m<br />

+<br />

n=<br />

n=m<br />

p=1<br />

N<br />

N<br />

p=1<br />

N<br />

xn m<br />

x0 u,v<br />

m=1 n=1 p=1<br />

⎣e2j(φu φv) x0 v,u<br />

p n<br />

v,v xu,v x<br />

N<br />

x<br />

N<br />

n m<br />

v,u<br />

N<br />

m p<br />

v,u<br />

<br />

p n<br />

xv,v + x<br />

N<br />

x<br />

m=1 n=1 p=1<br />

p=n<br />

n m<br />

v,u<br />

n p<br />

v,v<br />

<br />

<br />

p n<br />

xv,v + x<br />

n p<br />

v,v<br />

<br />

N + N <br />

p m x <br />

+<br />

x m n<br />

v,v v,u<br />

m=1 n= p=1<br />

2 + e2j(φu φv) x<br />

<br />

N + N <br />

m p x <br />

+<br />

x n m<br />

v,v v,u<br />

m=1 n= p=1<br />

2 + e2j(φu φv) x<br />

<br />

N N N<br />

<br />

<br />

+<br />

xn m p<br />

v,u xm u,v xn m m<br />

v,v + xp v,v<br />

m=1 n=1 p=1<br />

<br />

<br />

N N N<br />

<br />

2j(φu + e φv) xn m p m p m<br />

v,u xv,u xv,v + x<br />

m=1 n=1 p=1<br />

<br />

N N N<br />

<br />

<br />

+<br />

xn m m p<br />

v,u xu,v xm n m p<br />

v,v + xv,v m=1 n=1 p=1<br />

<br />

<br />

N N N<br />

<br />

2j(φu + e φv) xn m m m p<br />

xp v,u xv,v + x<br />

m=1 n=1 p=1<br />

while the variance of the ISI contribution is given by<br />

v,u<br />

n p<br />

v,u<br />

⎤<br />

⎦<br />

m n<br />

v,u<br />

n m<br />

v,u<br />

n m<br />

v,v<br />

m n<br />

v,v<br />

<br />

<br />

<br />

2 <br />

2 <br />

(C.5)


C.1 Multiuser estimator 169<br />

BPSK 2<br />

σISIu = 4 Nx 0 u,u<br />

+2<br />

2<br />

2 Ev<br />

Eu<br />

⎧<br />

⎪⎨<br />

⎪⎩<br />

⎧<br />

⎪⎨<br />

⎪⎩<br />

⎧<br />

⎪⎨<br />

⎪⎩<br />

N<br />

m=1<br />

N<br />

m=1<br />

N<br />

m=1<br />

+<br />

n=<br />

n=m<br />

N<br />

N<br />

n=1<br />

n=m<br />

+<br />

n=<br />

n=m<br />

m=1 n=1<br />

n=m<br />

N<br />

m=1<br />

+ N<br />

m=1<br />

+<br />

x <br />

n m<br />

N<br />

n=<br />

n=m<br />

N<br />

n=1<br />

n=m<br />

u,u<br />

2 <br />

x <br />

n m2<br />

x<br />

u,u<br />

x <br />

n m<br />

u,u<br />

2 <br />

x <br />

n m2<br />

x<br />

u,u<br />

x <br />

n m<br />

v,v<br />

2 + <br />

x <br />

n m2<br />

+ x<br />

On the other hand, the <strong>de</strong>nominator Den BPSK<br />

u<br />

Den BPSK<br />

u<br />

⎧<br />

⎨<br />

=2<br />

⎩<br />

N 2 x 0 u,u x0 v,v<br />

1<br />

2<br />

N<br />

m=1 n=1<br />

C.1.2 QPSK modulation<br />

N x <br />

n m<br />

v,u<br />

v,v<br />

2 + <br />

x <br />

n m 2<br />

u,u<br />

n m<br />

u,u<br />

writes<br />

2 <br />

x <br />

n m 2<br />

u,u<br />

n m<br />

u,u<br />

2 <br />

x <br />

n m 2<br />

v,v<br />

n m<br />

v,v<br />

2 <br />

<br />

e2j(φu φv) x<br />

n m<br />

v,u<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

⎫<br />

⎪⎬<br />

.<br />

⎪⎭<br />

(C.6)<br />

⎫2<br />

⎬<br />

<br />

2 .<br />

⎭<br />

(C.7)<br />

Moving to QPSK-modulated data symbols, the variance expression does<br />

not apparently differ from the corresponding expression (C.4) obtained<br />

consi<strong>de</strong>ring BPSK-modulated symbols<br />

<br />

σ<br />

2 QP SK<br />

∆u<br />

=<br />

1<br />

Es,u<br />

x<br />

N0<br />

0 u,u<br />

QP SK<br />

Numu QP SK<br />

Denu 2 QP SK<br />

ISIu<br />

QP SK<br />

Denu <br />

σ<br />

+<br />

. (C.8)<br />

In<strong>de</strong>ed, the differences lie in the <strong>de</strong>finitions of the numerator Num QP SK<br />

u


170 Variance of DA ML FF phase estimators<br />

QP SK<br />

Numu = N 3 x0 <br />

u,u x0 2<br />

v,v<br />

+N x0 N +<br />

u,u<br />

m=1 n=<br />

n=m<br />

N<br />

Nx0 v,v<br />

m=1 n=1<br />

N<br />

m=1<br />

<br />

<br />

<br />

N<br />

+<br />

<br />

N<br />

+<br />

+<br />

n=<br />

n=m<br />

N<br />

N x N<br />

x<br />

p=1<br />

N<br />

<br />

xm n 2<br />

v,v<br />

m n<br />

u,v<br />

n m n<br />

v,v xp<br />

N<br />

x<br />

x0 u,v<br />

m=1 n=1 p=1<br />

+<br />

m=1 n=<br />

N<br />

N<br />

<br />

2 + x0 <br />

u,vxn m<br />

v,u<br />

<br />

u,v x<br />

N <br />

xm n<br />

p=1<br />

<br />

the variance of the ISI contribution σ<br />

<br />

σ<br />

2 QP SK<br />

ISIu<br />

= 4 Nx 0 u,u<br />

+2<br />

2<br />

2 Ev<br />

Eu<br />

v,u<br />

m p<br />

v,u<br />

n m p<br />

v,u xn v,v<br />

2 <br />

xn m m p<br />

v,u xu,v m=1 n=1 p=1<br />

2 QP SK<br />

ISIu<br />

<br />

p m<br />

xv,v + x<br />

<br />

p m<br />

xv,v + x<br />

m p<br />

v,v<br />

m n<br />

v,v<br />

<br />

<br />

,<br />

⎧<br />

⎪⎨ N + <br />

n m<br />

x <br />

u,u<br />

⎪⎩ m=1 n=<br />

n=m<br />

2<br />

N N <br />

n m<br />

x <br />

u,u<br />

m=1 n=1<br />

n=m<br />

2<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

⎧<br />

⎨ N + <br />

xn m<br />

u,u<br />

⎩m=1<br />

n=<br />

n=m<br />

2 N N <br />

xn m<br />

u,u<br />

m=1 n=1<br />

n=m<br />

2<br />

⎫<br />

⎬<br />

⎭<br />

⎧<br />

⎨ N + <br />

xn m<br />

v,v<br />

⎩<br />

2 + N N <br />

xn m<br />

v,v<br />

2<br />

⎫<br />

⎬<br />

⎭ ,<br />

m=1<br />

QP SK<br />

and the <strong>de</strong>nominator Denu QP SK<br />

Denu =2<br />

<br />

n=<br />

n=m<br />

N 2 x 0 u,u x 0 v,v<br />

1<br />

2<br />

N<br />

m=1 n=1<br />

m=1<br />

n=1<br />

n=m<br />

N <br />

n m<br />

x<br />

v,u<br />

2<br />

2<br />

(C.9)<br />

(C.10)<br />

. (C.11)<br />

Comparing (C.11) to(C.7), the rea<strong>de</strong>r can notice that the terms weighted<br />

by I2 <br />

k have disappeared due to (C.3).


C.2 Single-user estimator 171<br />

C.2 Single-user estimator<br />

After <strong>de</strong>riving the variance of the Multiuser DA ML FF phase estimator<br />

(4.32), the variance of the Single-User estimator (4.37) is to be <strong>de</strong>rived.<br />

C.2.1 BPSK modulation<br />

Applying the same procedure as the one applied for obtaining (C.4), the<br />

variance of a Single-User DA ML FF estimator writes in the case of BPSKmodulated<br />

data symbols<br />

BPSK 2<br />

σ =<br />

∆u<br />

1<br />

2 N<br />

⎧<br />

+<br />

⎪⎨<br />

⎪⎩<br />

+ Ev<br />

Eu<br />

1<br />

Es,u<br />

N<br />

N0<br />

m=1<br />

N<br />

m=1<br />

N<br />

+<br />

n=<br />

n=m<br />

N<br />

n=1<br />

n=m<br />

m=1 n=1<br />

C.2.2 QPSK modulation<br />

x <br />

n m<br />

x<br />

N x u,u<br />

n m<br />

u,u<br />

2 <br />

<br />

2 x<br />

2 N 2 x0 2 u,u<br />

n m<br />

v,u<br />

<br />

2 <br />

x <br />

n m 2<br />

u,u<br />

n m<br />

u,u<br />

2 <br />

<br />

e2j(φu φv) x<br />

2 N 2 x0 2 u,u<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

n m<br />

v,u<br />

2 <br />

.<br />

(C.12)<br />

Finally, the variance of a Single-User DA ML FF phase estimator writes for<br />

QPSK-modulated data symbols


172 Variance of DA ML FF phase estimators<br />

<br />

σ<br />

2 QP SK<br />

∆u<br />

=<br />

1<br />

2 N<br />

N<br />

+<br />

m=1<br />

+ Ev<br />

Eu<br />

1<br />

Es,u<br />

N0<br />

+<br />

n=<br />

n=m<br />

N<br />

m=1 n=1<br />

<br />

xn m<br />

u,u<br />

2<br />

N<br />

2 N 2 x0 u,u<br />

N <br />

xn m2<br />

v,u<br />

N<br />

m=1 n=1<br />

n=m<br />

2 <br />

xn m<br />

u,u<br />

2 N 2 x0 2 . (C.13)<br />

u,u<br />

Again, the terms weighted by I 2 k<br />

disappear when switching from BPSKto<br />

QPSK-modulated data symbols.<br />

2


Appendix D<br />

First or<strong>de</strong>r statistics in a linear<br />

channel<br />

The purpose of this appendix is to <strong>de</strong>velop the expressions of the first or<strong>de</strong>r<br />

statistics of products involving data symbols and <strong>de</strong>cisions present<br />

in (5.8) and (5.8), consi<strong>de</strong>ring a 2-user synchronous communication systems<br />

transmitting over a linear channel. The following <strong>de</strong>velopments will<br />

be limited to Signal ¢ Signal first or<strong>de</strong>r statistics since these are the only<br />

statistics required to compute (5.8) and (5.8). In<strong>de</strong>ed, products involving<br />

Noise disappear from these relations thanks to the constellation symm<strong>et</strong>ry<br />

[87].<br />

Using the Rice component of the additive noise samples ν m k and νm k<br />

given by (3.80) and with the following <strong>de</strong>finitions<br />

¯ BPSK<br />

¯ QPSK<br />

X ¦ = x 0 u,u cos ∆u ¦<br />

Y ¦ =<br />

Xk = x 0 u,u cos<br />

Eu<br />

Ev<br />

π<br />

4<br />

k 1 <br />

+( 1) 2<br />

Ev<br />

Eu<br />

x 0 u,v cos (δv,u +∆u) (D.1)<br />

x 0 u,v cos (δu,v +∆v) ¦ x 0 v,v cos ∆v (D.2)<br />

<br />

k 1<br />

+( 1) 4 ∆u<br />

Ev<br />

Eu<br />

x 0 u,v cos<br />

<br />

π<br />

4 +( 1)k 1 <br />

(δv,u +∆u)<br />

(D.3)


174 First or<strong>de</strong>r statistics in a linear channel<br />

Yk =<br />

where k =1...8.<br />

Eu<br />

Ev<br />

x 0 v,u cos<br />

<br />

π<br />

4<br />

k 1 <br />

+( 1) 2 x 0 v,v cos<br />

<br />

k 1<br />

+( 1) 4 (δu,v +∆v)<br />

π<br />

4 +( 1)k 1 ∆v<br />

<br />

(D.4)<br />

the analytical expressions of the expectations appearing in (5.8) and (5.10)<br />

can be <strong>de</strong>rived. But before writing any expression, notice that the linearity<br />

of the channel and the synchronous transmission modify the expression<br />

of the matched filter output (3.7) in such a way that it only <strong>de</strong>pends now<br />

on the current symbols and is thus subject to MAI only<br />

y m u = ejφuI m u x0u,u +e<br />

Useful term, no ISI<br />

jφv<br />

<br />

Ev<br />

Eu Im v x0u,v MAI<br />

Additive noise<br />

+ν m u<br />

(D.5)<br />

This writing, combined with the statistical in<strong>de</strong>pen<strong>de</strong>nce b<strong>et</strong>ween the data<br />

flows of the users, drives to zero most of the expectations in (5.8) and<br />

(5.10), namely those for which temporal in<strong>de</strong>xes n and m differ (n m = 0).<br />

As a result, only terms for which n = m are non zero. Their computation<br />

follows.<br />

D.1 Expectations of data ¢ <strong>de</strong>cision products<br />

Data ¢ <strong>de</strong>cision products come from the expansion of the matched filter<br />

output. Two different kinds of products can be consi<strong>de</strong>red. The first one<br />

involves only signals from the user of interest and is thus the useful contribution.<br />

On the other hand, the second one <strong>de</strong>pends on User ¢ Interferer<br />

products as a result of the MAI introduced through the matched filter output.<br />

D.1.1 User ¢ User<br />

Same I/Q branch<br />

Consi<strong>de</strong>r first the expectation of products involving data symbols and <strong>de</strong>cisions<br />

related to the same user over the same branch.


D.1 Expectations of data ¢ <strong>de</strong>cision products 175<br />

BPSK<br />

QPSK<br />

Cross-talk<br />

<br />

E â m u an u <br />

ˆΦ<br />

n=m<br />

=0, Φ=∆<br />

= 1 P ν m u X+ P ν m u<br />

<br />

E â m u a n <br />

u ˆΦ<br />

n=m<br />

=0, Φ=∆<br />

= 1<br />

16<br />

X <br />

(D.6)<br />

8<br />

[1 2P (ν m u Xk)] (D.7)<br />

k=1<br />

<br />

E ˆb m<br />

u b n <br />

<br />

u<br />

ˆΦ=0, Φ=∆<br />

= 1<br />

16<br />

n=m<br />

8<br />

[1 2P (ν m u Xk)] . (D.8)<br />

k=1<br />

Moving to products involving contributions from different I/Q branches,<br />

the <strong>de</strong>rived expectations quantify the inci<strong>de</strong>nce of cross-talk. Obviously,<br />

there are no contribution to take into account in the case of BPSK-modulated<br />

data symbols. On the contrary, in the QPSK case<br />

<br />

E â m u bn u <br />

ˆΦ<br />

n=m<br />

=0, Φ=∆<br />

= 1<br />

8<br />

4<br />

[P (ν m u Xk+4) P (ν m u Xk)] (D.9)<br />

k=1<br />

<br />

E ˆb m<br />

u a n <br />

<br />

u<br />

ˆΦ=0, Φ=∆<br />

= 1<br />

8<br />

n=m<br />

4<br />

[P (ν m u Xk) P (ν m u Xk+4)] . (D.10)<br />

k=1<br />

Having <strong>de</strong>alt so far with the useful User ¢ User products only, it is time to<br />

consi<strong>de</strong>r User ¢ Interferer contributions due to MAI.


176 First or<strong>de</strong>r statistics in a linear channel<br />

D.1.2 User ¢ Interferer<br />

Similarly to the previous section, several combinations will be consi<strong>de</strong>red,<br />

either on the same branch or with cross-talk.<br />

Same I/Q branch<br />

On the same branch, the expectations write<br />

BPSK<br />

QPSK<br />

E<br />

<br />

E â m u anv <br />

ˆΦ<br />

n=m<br />

=0, Φ=∆<br />

= P ν m u X P ν m u<br />

<br />

â m u a n <br />

v ˆΦ<br />

n=m<br />

=0, Φ=∆<br />

= 1<br />

8<br />

4<br />

k=1<br />

<br />

P ν m u X k 1<br />

k+2 2 <br />

<br />

<br />

E ˆb m<br />

u b n <br />

<br />

v ˆΦ=0, Φ=∆<br />

= 1<br />

8<br />

4<br />

k=1<br />

n=m<br />

<br />

P ν m u X k 1<br />

k+2 2 <br />

<br />

P<br />

P<br />

X+<br />

<br />

ν m u X k+2 k 1<br />

2<br />

<br />

ν m u X k 1<br />

k+2 2<br />

where k represents the least integral value strictly greater than k<br />

Cross-talk<br />

<br />

<br />

(D.11)<br />

(D.12)<br />

(D.13)<br />

Consi<strong>de</strong>ring cross-talk, the QPSK case needs again to be studied alone.<br />

The expectations of User ¢ Interferer products are given by<br />

<br />

E â m u bnv <br />

ˆΦ<br />

n=m<br />

=0, Φ=∆<br />

= 1<br />

8<br />

4<br />

k=1<br />

<br />

P ν m u Xk+2 k 1 1 2<br />

<br />

P<br />

<br />

ν m u X 3<br />

k 1<br />

+2( 1)k<br />

2<br />

<br />

(D.14)


D.2 Expectations of <strong>de</strong>cision ¢ <strong>de</strong>cision products 177<br />

<br />

E ˆb m<br />

u a n <br />

<br />

v ˆΦ=0, Φ=∆<br />

= 1<br />

8<br />

4<br />

k=1<br />

n=m<br />

<br />

P ν m <br />

u X k 1 3 +2( 1)k<br />

2<br />

<br />

P ν m <br />

u X k 1<br />

k+2 1 .<br />

2<br />

(D.15)<br />

These were the expectations quantifying the inci<strong>de</strong>nce of the MAI introduced<br />

in the system through the matched filter output. On the other hand,<br />

the inci<strong>de</strong>nce of the MAI mitigation has now to be computed. It <strong>de</strong>pends<br />

on <strong>de</strong>cision ¢ <strong>de</strong>cision products.<br />

D.2 Expectations of <strong>de</strong>cision ¢ <strong>de</strong>cision products<br />

Dealing with MAI mitigation terms, the current section has only to consi<strong>de</strong>r<br />

User ¢ Interferer products, first on the same branch, then with crosstalk,<br />

and always in both BPSK and QPSK cases if applicable.<br />

D.2.1 Same I/Q branch<br />

The expectations of <strong>de</strong>cision ¢ <strong>de</strong>cision products related to a unique branch<br />

write<br />

BPSK<br />

<br />

E â m u ânv ˆΦ =0, Φ=∆<br />

n=m<br />

= 2P (νm u X + ) P (νm u X + ) P (νm u X + ) P (νm u X + )<br />

2P (νm u X ) P (νm u X ) P (νm u X ) P (νm u X )<br />

+1<br />

(D.16)<br />

QPSK<br />

<br />

E â m u â n <br />

v ˆΦ<br />

n=m<br />

=0, Φ=∆<br />

= 1<br />

16<br />

8<br />

k=1<br />

4P (ν m u Xk) P (ν m v Yk)<br />

2P (ν m u Xk) 2P (ν m v Yk)+1<br />

<br />

(D.17)


178 First or<strong>de</strong>r statistics in a linear channel<br />

<br />

E ˆb mˆn u bv ˆΦ=0,<br />

<br />

n=m<br />

Φ=∆<br />

= 1<br />

8<br />

<br />

4P (νm u Xk) P (ν<br />

16<br />

m v Yk)<br />

2P (νm u Xk) 2P (νm v Yk)+1<br />

k=1<br />

D.2.2 Cross-talk<br />

<br />

. (D.18)<br />

Consi<strong>de</strong>ring cross-talk, the corresponding expectations become<br />

<br />

E â m u ˆb n <br />

<br />

v ˆΦ=0,<br />

<br />

n=m<br />

Φ=∆<br />

= 1<br />

⎧ <br />

4<br />

⎪⎨<br />

4P (νm u Xk) P<br />

16<br />

k=1 ⎪⎩<br />

νm <br />

v Y5+(k%4) 2P (νm u Xk) 2P νm <br />

<br />

v Y5+(k%4) +1<br />

4P νm u X5+(k%4) P (νm v Yk)<br />

2P νm ⎫<br />

⎪⎬<br />

<br />

⎪⎭<br />

u X5+(k%4) 2P (νm v Yk)+1<br />

(D.19)<br />

<br />

E ˆb m<br />

u â n <br />

<br />

v ˆΦ=0,<br />

<br />

n=m<br />

Φ=∆<br />

= 1<br />

⎧ <br />

4<br />

⎪⎨<br />

4P (νm v Yk) P<br />

16<br />

k=1 ⎪⎩<br />

νm v X <br />

5+(k%4)<br />

2P (νm v Yk) 2P νm u X <br />

<br />

5+(k%4) +1<br />

4P νm v Y5+(k%4) P (νm u Xk)<br />

2P νm v Y ⎫<br />

⎪⎬<br />

<br />

⎪⎭<br />

5+(k%4) 2P (νm u Xk)+1<br />

(D.20)<br />

where k%4 is the rest of the division of k by 4.<br />

D.3 Conclusion<br />

Comparing the expectations of data ¢ <strong>de</strong>cision products (D.6-D.15) to<br />

the ones of <strong>de</strong>cision ¢ <strong>de</strong>cision products (D.16-D.20), it appears that the<br />

former only <strong>de</strong>pends on X while the latter also <strong>de</strong>pends on Y . As<br />

far as open-loop performance is concerned, contributions from the matched<br />

filter output are thus only a function of the phase estimation error<br />

related to the user of interest, while MAI mitigation terms are sensitive to<br />

estimation errors related to both user and interferer(s).


Appendix E<br />

Expectations for DD FB<br />

open-loop performance<br />

evaluation<br />

The present appendix <strong>de</strong>scribes the calculation of the expectations of products<br />

b<strong>et</strong>ween true data symbols and <strong>de</strong>cisions. Such expectations are<br />

encountered in the <strong>de</strong>rivation of the open-loop performance of DD FB estimators.<br />

They <strong>de</strong>pend on the data symbols in such a way that, at first<br />

sight, their computation would require to consi<strong>de</strong>r all hypotheses regarding<br />

the messages sent. Fortunately, the use of the characteristic functions<br />

helps to avoid this time-consuming approach.<br />

E.1 BPSK Modulation<br />

⋆ E Îm u In <br />

<br />

v ˆ ⋆Î <br />

Φ=0, Φ=∆ and E Îm n u v ˆ <br />

Φ=0, Φ=∆ will be <strong>de</strong>rived<br />

in this section for BPSK modulated data symbols. Since the modulation<br />

is binary, there is neither conjugate in the argument of the expectation<br />

nor quadrature component in data symbols and hard <strong>de</strong>cisions. Thus,<br />

⋆ m<br />

E Îu I n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆ = E â m u anv ˆ <br />

Φ=0, Φ=∆ (E.1)<br />

⋆Î <br />

m n E Îu v ˆ <br />

Φ=0, Φ=∆ = E â m u â n v ˆ <br />

Φ=0, Φ=∆ . (E.2)


180 Expectations for DD FB open-loop performance evaluation<br />

<br />

E.1.1 Derivation of E âm u an v ˆ <br />

Φ=0, Φ=∆<br />

Using A p<br />

k introduced in (3.83) which, in the current BPSK context (no information<br />

on the Q-branch), writes<br />

A p<br />

k =<br />

+<br />

q=<br />

a q q<br />

kRp k,k +<br />

Nu <br />

l=1<br />

l=k<br />

+<br />

<br />

the expectation E âm u an v ˆ <br />

Φ=0, Φ=∆ becomes<br />

<br />

E<br />

â m u anv ˆ <br />

Φ=0, Φ=∆<br />

q=<br />

a q<br />

<br />

q<br />

l<br />

Rp<br />

k,l + e j ˆ φk p<br />

νk <br />

= E sign (A m u ) a n v ˆ <br />

Φ=0, Φ=∆<br />

= 1<br />

⎧ <br />

Pr A<br />

⎪⎨<br />

2<br />

⎪⎩<br />

m u > 0 an v =1, ˆ <br />

Φ=0, Φ=∆<br />

<br />

Pr Am u < 0 anv =1, ˆ <br />

Φ=0, Φ=∆<br />

<br />

Pr Am u > 0 anv = 1, ˆ <br />

Φ=0, Φ=∆<br />

<br />

+Pr Am u < 0 an v = 1, ˆ ⎫<br />

⎪⎬<br />

<br />

⎪⎭<br />

Φ=0, Φ=∆<br />

<br />

(E.3)<br />

(E.4)<br />

(E.5)<br />

where an 1<br />

v = ¦1 with equal probability 2 . In or<strong>de</strong>r to <strong>de</strong>rive the probabilities<br />

in (E.5), the pdf of Am u is requested. It will then be integrated over<br />

[ , 0] and [0, + ]. This pdf can be <strong>de</strong>rived as the inverse Fourier transform<br />

of the characteristic function<br />

ψA m u (ωu) =E e jAm u ωu . (E.6)<br />

Switching the inverse Fourier transform and the integration over [ , 0]<br />

and [0, + ] and using the fact that<br />

+<br />

0<br />

0<br />

<br />

e jEω dE = 1<br />

+ πδ (ω) (E.7)<br />

jω<br />

e jEω dE = πδ (ω)<br />

1<br />

jω<br />

(E.8)


E.1 BPSK Modulation 181<br />

leads to a new writing of (E.4)<br />

<br />

E â m u a n v ˆ <br />

Φ=0, Φ=∆<br />

=<br />

1<br />

⎡<br />

+<br />

⎢ ψA<br />

⎢<br />

2jπ ⎣<br />

m <br />

ωu a u<br />

n v =1, ˆ <br />

dωu<br />

Φ=0, Φ=∆ ωu<br />

+ <br />

ωu an v = 1, ˆ Φ=0, Φ=∆<br />

ψA m u<br />

dωu<br />

ωu<br />

⎤<br />

⎥ . (E.9)<br />

⎦<br />

Introducing (E.3) into (E.6) enables to write the characteristic function in<br />

open-loop conditions as follows<br />

ψAm u (ωu a n v = ¦1, ˆ Φ=0, Φ=∆)<br />

⎧ ⎧ ⎡ +<br />

⎪⎨ ⎪⎨ ⎢<br />

a<br />

⎢ p=<br />

= E exp jωu ⎢<br />

⎣<br />

⎪⎩ ⎪⎩<br />

p m p<br />

uRu,u + Nu +<br />

<br />

= exp<br />

⎡<br />

⎢<br />

⎣<br />

+ <br />

p=<br />

p=n<br />

σ2 (νu) ω2 u<br />

2<br />

cos ωuR<br />

k=1<br />

k=u<br />

p=<br />

m n<br />

¦ jωuRu,v m p<br />

u,v<br />

⎤ ⎡<br />

Nu ⎥<br />

⎢<br />

<br />

⎦ ⎣<br />

k=1<br />

k=u<br />

where, in open-loop, (3.77) turns into<br />

p q<br />

Rk,l =<br />

El<br />

Ek<br />

a p p<br />

kRm u,k + (νm u )<br />

<br />

q=<br />

+ <br />

<br />

<br />

<br />

cos ωuR<br />

m q<br />

u,k<br />

e j(δk,l+∆k) x p q<br />

k,l<br />

⎤⎫<br />

<br />

⎥⎪⎬<br />

<br />

<br />

⎥ <br />

⎥ <br />

⎦ <br />

⎪⎭ <br />

<br />

Finally, inserting (E.11) into (E.9) produces the expectation<br />

<br />

E â m u a n v ˆ <br />

Φ=0, Φ=∆<br />

= 1<br />

π<br />

+<br />

<br />

exp<br />

⎡<br />

⎣ +<br />

p=<br />

p=n<br />

σ2<br />

(νu) ω2 <br />

u<br />

2<br />

<br />

cos ωuR<br />

m n<br />

u,v<br />

sin ωuR<br />

⎤ ⎡<br />

⎦ ⎣ Nu <br />

m p<br />

u,v<br />

k=1<br />

k=v<br />

<br />

q=<br />

+<br />

a n v<br />

⎫<br />

⎪⎬<br />

= ¦1<br />

⎪⎭<br />

(E.10)<br />

⎤<br />

<br />

⎥<br />

⎦ (E.11)<br />

<br />

. (E.12)<br />

<br />

cos ωuR<br />

m q<br />

u,k<br />

⎤<br />

⎦ dωu<br />

ωu .<br />

(E.13)


182 Expectations for DD FB open-loop performance evaluation<br />

(E.13) applies to the multiuser context un<strong>de</strong>r investigation. To validate it,<br />

one can move to the situation studied in [87], i.e. a single user transmission<br />

(u = v, El =0 l = u) over an AWGN channel (xm n<br />

u,u = x0u,uδ(m n)).<br />

This produces<br />

<br />

E â m u amu ˆ <br />

Φ=0, Φ=∆<br />

= 1<br />

π<br />

+<br />

exp<br />

<br />

σ2 (νu) ω2 u<br />

2<br />

<br />

sin ωu cos ∆ux 0 dωu<br />

u,u . (E.14)<br />

Using [120, p. 123], (E.14) finally turns into the same result as in [87]<br />

<br />

E â m u a m u ˆ <br />

Φ=0, Φ=∆<br />

=1 2Q<br />

⎛<br />

⎝ cos ∆ux 0 u,u<br />

<br />

σ 2 (νu)<br />

with Q (x) <strong>de</strong>fined in (3.89). The rea<strong>de</strong>r can also notice that<br />

<br />

E â m u amu ˆ <br />

Φ=0, Φ=∆<br />

ωu<br />

⎞<br />

⎠ (E.15)<br />

= (1 PE)+( 1) PE (E.16)<br />

= 1 2 PE (E.17)<br />

where PE represents the BPSK error probability in AWGN channels as a<br />

function of the phase misalignment ∆u. If∆u =0, PE becomes [121, p. 94]<br />

<br />

PE = Q σ 1<br />

<br />

(ν) = Q<br />

<br />

2Eb<br />

. (E.18)<br />

N0<br />

<br />

E.1.2 Derivation of E âm u ân v ˆ <br />

Φ=0, Φ=∆<br />

<br />

The <strong>de</strong>rivation of E âm u ân v ˆ <br />

Φ=0, Φ=∆ is a little more intricate but fol-<br />

lows the same procedure as in the previous section. Am u and Anv being the<br />

arguments of the discriminating functions producing hard <strong>de</strong>cisions âm u


E.1 BPSK Modulation 183<br />

and ân <br />

v , the expectation E âm u ân v ˆ <br />

Φ=0, Φ=∆ becomes<br />

<br />

E â m u â n v ˆ <br />

Φ=0, Φ=∆<br />

<br />

= E sign (A m u )sign(A n v ) ˆ <br />

Φ=0, Φ=∆<br />

(E.19)<br />

<br />

= Pr A m u > 0,Anv > 0 ˆ <br />

Φ=0, Φ=∆<br />

<br />

Pr A m u > 0,Anv < 0 ˆ <br />

Φ=0, Φ=∆<br />

<br />

Pr A m u < 0,Anv > 0 ˆ <br />

Φ=0, Φ=∆<br />

<br />

+Pr A m u < 0,A n v < 0 ˆ <br />

Φ=0, Φ=∆ . (E.20)<br />

Once again, the characteristic function ψA m u ,A n v (ωu,ωv) is of great help to<br />

<strong>de</strong>rive the probabilities in (E.20). Following the steps which lead from<br />

(E.5) to(E.9), one g<strong>et</strong>s<br />

E<br />

<br />

â m u â n v ˆ <br />

Φ=0, Φ=∆<br />

= 1<br />

π 2<br />

+<br />

+<br />

ψA m u ,An v<br />

+ (ν n v )<br />

<br />

ωu,ωv ˆ <br />

dωu dωv<br />

Φ=0, Φ=∆<br />

ωu ωv<br />

(E.21)<br />

where the characteristic function writes in open-loop conditions<br />

ψAm u ,An <br />

ωu,ωv v<br />

ˆ <br />

Φ=0, Φ=∆<br />

<br />

= E e j(Amu ωu+An <br />

v ωv) <br />

ˆ <br />

Φ=0, Φ=∆<br />

(E.22)<br />

=<br />

⎧ ⎧ ⎡<br />

+<br />

⎢ a<br />

jωu ⎣ p=<br />

⎪⎨ ⎪⎨<br />

E exp<br />

⎪⎩ ⎪⎩<br />

p m p<br />

uRu,u + Nu +<br />

k=1 p=<br />

k=u<br />

a p<br />

+ (ν<br />

p<br />

kRm u,k<br />

m u )<br />

⎡<br />

+<br />

⎢ a<br />

+jωv ⎣ q=<br />

⎤<br />

⎥<br />

⎦<br />

q n q<br />

vRv,v + Nu +<br />

l=1 q=<br />

l=v<br />

a q<br />

⎫⎫<br />

⎪⎬ ⎪⎬<br />

⎤<br />

q<br />

l<br />

Rn<br />

v,l ⎥<br />

⎦<br />

⎪⎭ ⎪⎭<br />

<br />

= exp<br />

<br />

Nu <br />

k=1 p=<br />

1<br />

<br />

2<br />

<br />

cos ωuR<br />

+ <br />

σ 2 (νu) ω2 m n<br />

u +2ρu,v ωuωv + σ 2 (νv) ω2 v<br />

m p<br />

u,k<br />

<br />

(E.23)<br />

<br />

n p<br />

+ ωvRv,k <br />

. (E.24)


184 Expectations for DD FB open-loop performance evaluation<br />

Using (E.24) in(E.21) finally gives the expectation<br />

<br />

E â m u â n v ˆ <br />

Φ=0, Φ=∆<br />

= 1<br />

π 2<br />

+<br />

+<br />

<br />

exp<br />

<br />

Nu <br />

k=1 q=<br />

1<br />

<br />

2 σ2 (νu) ω2 u<br />

+<br />

E.2 QPSK Modulation<br />

<br />

cos ωuR<br />

+2ρm n<br />

u,v ωuωv + σ2 (νv) ω2 <br />

v<br />

m q<br />

u,k<br />

+ ωvR n q<br />

v,k<br />

<br />

dωu<br />

ωu<br />

dωv<br />

ωv .<br />

(E.25)<br />

Dealing now with QPSK modulated data symbols, both in-phase and quadrature<br />

components have to be taken into account so that<br />

⋆ m<br />

E Îu I n v ˆ <br />

Φ=0, Φ=∆<br />

<br />

= E â m u a n v ˆ <br />

Φ=0, Φ=∆ + E ˆb m<br />

u b n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆<br />

<br />

+j E â m u b n v ˆ <br />

Φ=0, Φ=∆ E ˆb m<br />

u a n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆<br />

(E.26)<br />

⋆Î m n<br />

E Îu v ˆ <br />

Φ=0, Φ=∆<br />

<br />

= E â m u â n v ˆ <br />

Φ=0, Φ=∆ + E ˆb mˆn u bv ˆ <br />

Φ=0, Φ=∆<br />

<br />

+j E â m u ˆb n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆ E ˆb m<br />

u â n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆ .<br />

(E.27)<br />

It can be shown that the computation of the eight expectations listed here<br />

above is unnecessary. Thanks to the symm<strong>et</strong>ry of the QPSK constellation,<br />

the following relationships apply<br />

<br />

E<br />

E<br />

â m u a n v ˆ <br />

Φ=0, Φ=∆<br />

<br />

â m u b n v ˆ Φ=0, Φ=∆<br />

<br />

<br />

= E ˆb m<br />

u b n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆ (E.28)<br />

<br />

= E ˆb m<br />

u a n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆ . (E.29)<br />

Analogous relationships may be written regarding the expectation of the<br />

product of <strong>de</strong>cisions. Thus, only four expectations will be <strong>de</strong>rived in the<br />

next sections.


E.2 QPSK Modulation 185<br />

<br />

E.2.1 Derivation of E âm u an v ˆ <br />

Φ=0, Φ=∆<br />

<br />

The <strong>de</strong>rivation of E âm u an v ˆ <br />

Φ=0, Φ=∆ in a QPSK environment is very<br />

similar to the one lead in Section E.1.1 for BPSK. The main difference lies in<br />

the argument A p<br />

k (3.83) of the <strong>de</strong>cision function âp<br />

k =<br />

Ô<br />

2<br />

2 sign A p<br />

k . Unsurprisingly,<br />

cross-talk from quadrature components into in-phase <strong>de</strong>cisions<br />

of the same user appear due to the phase misalignment of oscillators.<br />

Moreover, MAI comes out from both in-phase and quadrature components.<br />

Thus, the characteristic function in open-loop conditions now writes<br />

<br />

ωu a n Ô<br />

2<br />

v = ¦<br />

2 , ˆ <br />

Φ=0, Φ=∆<br />

⎧ ⎧ ⎡ + <br />

a<br />

⎪⎨ ⎪⎨<br />

⎢ p=<br />

⎢<br />

= E exp jωu ⎢<br />

⎣<br />

⎪⎩ ⎪⎩<br />

p m p<br />

uRu,u f p uI<br />

+ Nu + <br />

a<br />

k=1 p=<br />

k=u<br />

p p<br />

kRm ψA m u<br />

<br />

= exp<br />

⎡<br />

Nu<br />

⎢<br />

<br />

⎣<br />

k=1<br />

k=v<br />

q=<br />

σ2 (νu) ω2 u<br />

+ <br />

2<br />

<br />

cos ωuR<br />

+ (ν m u )<br />

m n<br />

¦ jωuRu,v m q<br />

u,k<br />

⎡<br />

⎢<br />

⎣<br />

<br />

cos ωuI<br />

<br />

m p<br />

u,u<br />

u,k b p<br />

<br />

p<br />

kIm u,k<br />

+ <br />

p=<br />

p=n<br />

m q<br />

u,k<br />

cos ωuR<br />

m p<br />

u,v<br />

⎤⎫<br />

<br />

<br />

⎥<br />

⎥⎪⎬<br />

<br />

<br />

⎥ <br />

⎥ a<br />

⎥ <br />

⎦ <br />

⎪⎭ <br />

<br />

n ⎫<br />

Ô ⎪⎬<br />

2<br />

v = ¦<br />

2<br />

⎪⎭<br />

cos ωuI<br />

(E.30)<br />

⎤<br />

⎥<br />

⎦<br />

m p<br />

u,v<br />

⎤<br />

<br />

⎥<br />

⎦ . (E.31)<br />

Comparing (E.11) and (E.31), the rea<strong>de</strong>r can notice the <strong>de</strong>pen<strong>de</strong>ncy of the<br />

p q<br />

characteristic function on cosines functions of Ik,l introduced in (3.78). In<br />

p q<br />

open-loop conditions, consi<strong>de</strong>ring QPSK, R<br />

p q<br />

Rk,l p q<br />

Ik,l =<br />

=<br />

Ô<br />

2<br />

2<br />

Ô<br />

2<br />

2<br />

El<br />

Ek<br />

<br />

El<br />

Ek<br />

<br />

<br />

<br />

<br />

k,l<br />

and Ip q<br />

k,l write<br />

e j(δk,l+∆k) x p q<br />

k,l<br />

e j(δk,l+∆k) x p q<br />

k,l<br />

<br />

(E.32)<br />

<br />

. (E.33)


186 Expectations for DD FB open-loop performance evaluation<br />

Finally, the expectation writes<br />

<br />

E â m u a n v ˆ <br />

Φ=0, Φ=∆<br />

= 1<br />

2π<br />

+<br />

<br />

exp<br />

⎡<br />

⎣ +<br />

p=<br />

p=n<br />

⎡<br />

<br />

⎣ Nu<br />

k=1 q=<br />

k=v<br />

σ2<br />

(νu) ω2 <br />

u<br />

2<br />

<br />

cos ωuR<br />

+<br />

sin <br />

ωuRm n<br />

u,v cos ωuI<br />

<br />

m p<br />

cos ωuI ⎤<br />

⎦<br />

m p<br />

u,v<br />

<br />

cos ωuR<br />

m q<br />

u,k<br />

u,v<br />

<br />

cos ωuI<br />

m q<br />

u,k<br />

m n<br />

u,v<br />

<br />

⎤<br />

⎦ dωu<br />

ωu .<br />

(E.34)<br />

(E.34) was validated in the same way as done previously with (E.13). In<br />

the context <strong>de</strong>scribed in Section E.1.1, the expectation reduces to<br />

<br />

E<br />

<br />

â m u am u ˆ Φ=0, Φ=∆<br />

= 1<br />

2π<br />

+<br />

<br />

exp<br />

cos<br />

σ2 (νu) ω2 Ô<br />

u<br />

2<br />

2 sin<br />

Ô 2<br />

2 sin ∆ux 0 u,uωu<br />

2 cos ∆ux 0 u,uωu<br />

dωu<br />

which, after integration [120, p. 123], leads to<br />

<br />

E â m u amu ˆ <br />

Φ=0, Φ=∆<br />

= 1<br />

⎧<br />

⎨<br />

2 ⎩ Q<br />

⎡<br />

⎣ x0u,u cos ∆u + 3π<br />

<br />

⎤ ⎡<br />

4<br />

⎦ Q ⎣ x0u,u cos ∆u + π<br />

<br />

⎤⎫<br />

⎬<br />

4<br />

⎦<br />

⎭ .<br />

σ 2 (νu)<br />

ωu<br />

σ 2 (νu)<br />

<br />

(E.35)<br />

(E.36)<br />

This is exactly the relation (36) presented in [87]. Moreover, noticing that<br />

<br />

E â m u a m u ˆ <br />

Φ=0, Φ=∆<br />

=<br />

Ô 2<br />

2<br />

2<br />

(1 PE)<br />

Ô 2<br />

2<br />

2<br />

PE<br />

=<br />

(E.37)<br />

1<br />

2 (1 2PE) (E.38)


E.2 QPSK Modulation 187<br />

s<strong>et</strong>ting ∆u =0turns (E.36) into (E.18), the BPSK error probability which is<br />

to be un<strong>de</strong>rstood here as the bit error probability affecting each branch of<br />

a QPSK constellation in an AWGN channel.<br />

<br />

E â m u a m u ˆ <br />

Φ=0, Φ=∆<br />

1 <br />

2<br />

= 1<br />

1 2Q 2σ<br />

2<br />

2 (ν) (E.39)<br />

= 1<br />

<br />

2Eb<br />

1 2Q<br />

(E.40)<br />

2<br />

N0<br />

<br />

2Eb<br />

PE = Q . (E.41)<br />

N0<br />

<br />

E.2.2 Derivation of E âm u bnv ˆ <br />

Φ=0, Φ=∆<br />

Since the <strong>de</strong>cision with which it has been <strong>de</strong>alt in this section is the same<br />

as the one used in the previous section, the characteristic function used<br />

to <strong>de</strong>rive the expectation is still ψA m u (ωu). However, its inverse Fourier<br />

transform is now conditioned on b n v , leading to<br />

<br />

E<br />

=<br />

â m u bnv ˆ <br />

Φ=0, Φ=∆<br />

1<br />

4jπ<br />

⎡<br />

+<br />

⎢<br />

⎣ +<br />

ψA m u<br />

ψA m u<br />

<br />

ωu bn Ô<br />

2<br />

v = 2 , ˆ <br />

dωu<br />

Φ=0, Φ=∆ ωu<br />

<br />

ωu b n v =<br />

Ô 2<br />

2 , ˆ Φ=0, Φ=∆<br />

dωu<br />

ωu<br />

⎤<br />

⎥ . (E.42)<br />

⎦<br />

Inserting (E.31) in(E.42) produces a slightly modified version of (E.34)<br />

<br />

E â m u b n v ˆ <br />

Φ=0, Φ=∆<br />

= 1<br />

2π<br />

+<br />

<br />

exp<br />

⎡<br />

⎣ +<br />

p=<br />

p=n<br />

⎡<br />

<br />

⎣ Nu<br />

k=1 q=<br />

k=v<br />

σ2 (νu) ω2 <br />

u<br />

2<br />

<br />

cos ωuR<br />

+<br />

sin <br />

ωuI m n<br />

u,v cos ωuR<br />

<br />

m p<br />

cos ωuI ⎤<br />

⎦<br />

m p<br />

u,v<br />

<br />

cos ωuR<br />

m q<br />

u,k<br />

u,v<br />

<br />

cos ωuI<br />

m q<br />

u,k<br />

m n<br />

u,v<br />

<br />

⎤<br />

⎦ dωu<br />

ωu .<br />

(E.43)


188 Expectations for DD FB open-loop performance evaluation<br />

which, adapted to the conditions of the validation test, brings out<br />

<br />

E â m u b m u ˆ <br />

Φ=0, Φ=∆<br />

= 1<br />

⎧ ⎡<br />

⎨<br />

1 Q ⎣<br />

2 ⎩ x0u,u cos ∆u + 3π<br />

<br />

⎤ ⎡<br />

4<br />

⎦ Q ⎣ x0u,u cos ∆u + π<br />

<br />

⎤⎫<br />

⎬<br />

4<br />

⎦<br />

⎭ .<br />

σ 2 (νu)<br />

σ 2 (νu)<br />

(E.44)<br />

Again, it is exactly the same relation as the one presented in [87] un<strong>de</strong>r<br />

reference (37). Moreover, the rea<strong>de</strong>r can notice that (E.44) is equal to zero<br />

if there is no phase error (∆u =0). In this case there is no cross-talk b<strong>et</strong>ween<br />

in-phase and quadrature components. Decisions ma<strong>de</strong> on one Rice<br />

component no longer <strong>de</strong>pend on the other component.<br />

<br />

E.2.3 Derivation of E âm u ânv ˆ <br />

Φ=0, Φ=∆<br />

Like in Section E.1.2, the arguments of two <strong>de</strong>cisions functions of the type<br />

(3.83) need to be taken into account. Building up the expectation according<br />

to (E.20), the characteristic function ψA m u ,A n v (ωu,ωv) is <strong>de</strong>rived by applying<br />

the same procedure as before<br />

ψAm u ,An v (ωu,ωv ˆ Φ=0, Φ=∆)<br />

<br />

= E e j(Amu ωu+An <br />

v ωv) <br />

ˆ <br />

Φ=0, Φ=∆<br />

⎧ ⎧ ⎡ + <br />

a<br />

⎢ p=<br />

⎢<br />

jωu ⎢<br />

⎣<br />

⎪⎨ ⎪⎨<br />

= E exp<br />

⎪⎩ ⎪⎩<br />

p m p<br />

uRu,u f p uI<br />

+ Nu + <br />

a<br />

k=1 p=<br />

k=u<br />

p p<br />

kRm + (νm u )<br />

⎡ + <br />

a<br />

⎢ q=<br />

⎢<br />

+jωv ⎢<br />

⎣<br />

q n q<br />

vRv,v b q vI<br />

+ Nu + <br />

a<br />

l=1 q=<br />

l=v<br />

q q<br />

l<br />

Rn<br />

+ (ν n v )<br />

<br />

m p<br />

u,u<br />

u,k b p<br />

<br />

p<br />

kIm u,k<br />

<br />

n q<br />

v,v<br />

v,l b q<br />

<br />

q<br />

l<br />

In<br />

v,l<br />

(E.45)<br />

⎤ ⎫⎫<br />

⎥<br />

⎦<br />

⎪⎬ ⎪⎬<br />

⎤ (E.46)<br />

⎥<br />

⎦<br />

⎪⎭ ⎪⎭


E.2 QPSK Modulation 189<br />

<br />

= exp<br />

<br />

Nu <br />

k=1 p=<br />

1<br />

<br />

σ<br />

2<br />

2 (νu) ω2 n<br />

u +2ρm u,v ωuωv + σ 2 (νv) ω2 <br />

v<br />

<br />

+ <br />

<br />

cos ωuR<br />

m p<br />

u,k<br />

Using (E.47) in(E.21) finally gives<br />

<br />

E<br />

<br />

â m u ân v ˆ Φ=0, Φ=∆<br />

=<br />

1<br />

2π 2<br />

+<br />

+<br />

<br />

exp<br />

⎡<br />

<br />

⎣ Nu<br />

k=1 q=<br />

<br />

n p<br />

+ ωvRv,k cos ωuI<br />

1<br />

<br />

2 σ2 (νu) ω2 u +2ρm n<br />

<br />

m q<br />

cos ωuRu,k +<br />

<br />

cos ωuI<br />

m p<br />

u,k<br />

<br />

E.2.4 Derivation of E âm u ˆb n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆<br />

m p<br />

u,k<br />

<br />

n p<br />

+ ωvIv,k <br />

.<br />

u,v ωuωv + σ 2 (νv) ω2 v<br />

<br />

n q<br />

+ ωvRv,k n p<br />

+ ωvIv,k Not only are the arguments of two <strong>de</strong>cision functions, A p<br />

k<br />

⎤<br />

⎦ dωu<br />

ωu<br />

<br />

(E.47)<br />

dωv<br />

ωv .<br />

(E.48)<br />

and Bp<br />

k ,tobe<br />

consi<strong>de</strong>red in this section, but the second one is related to the quadrature<br />

component. This has been <strong>de</strong>fined in (3.84). The characteristic function of<br />

interest is thus ψA m u ,Bn v (ωu,ωv) which writes in open-loop conditions<br />

ψAm u ,Bn v (ωu,ωv ˆ Φ=0, Φ=∆)<br />

<br />

= E e j(Amu ωu+Bn <br />

v ωv) <br />

ˆ <br />

Φ=0, Φ=∆<br />

⎧ ⎧ ⎡ + <br />

a<br />

⎢ p=<br />

⎢<br />

jωu ⎢<br />

⎣<br />

⎪⎨ ⎪⎨<br />

= E exp<br />

⎪⎩ ⎪⎩<br />

p m p<br />

uRu,u f p <br />

uI<br />

+ Nu + <br />

a<br />

k=1 p=<br />

k=u<br />

p p<br />

kRm + (νm u )<br />

⎡ + <br />

a<br />

⎢ q=<br />

⎢<br />

+jωv ⎢<br />

⎣<br />

q n q<br />

vIv,v + b p <br />

n q<br />

vRv,v + Nu + <br />

a<br />

l=1 q=<br />

l=v<br />

q<br />

<br />

q<br />

q<br />

l<br />

In<br />

v,l + bq<br />

l<br />

Rn<br />

v,l<br />

+ (νn v )<br />

m p<br />

u,u<br />

u,k b p<br />

<br />

p<br />

kIm u,k<br />

(E.49)<br />

⎤ ⎫⎫<br />

⎥<br />

⎦<br />

⎪⎬ ⎪⎬<br />

⎤ (E.50)<br />

⎥<br />

⎦<br />

⎪⎭ ⎪⎭


190 Expectations for DD FB open-loop performance evaluation<br />

<br />

= exp<br />

<br />

Nu <br />

k=1 p=<br />

1<br />

<br />

σ<br />

2<br />

2 (νu) ω2 n<br />

u +2ρm u,v ωuωv + σ 2 (νv) ω2 <br />

v<br />

<br />

+ <br />

<br />

cos ωuI<br />

m p<br />

u,k<br />

Using (E.51) in(E.21) finally gives<br />

<br />

E â m u ˆb n <br />

<br />

v ˆ <br />

Φ=0, Φ=∆<br />

=<br />

1<br />

2π2 + + <br />

exp 1<br />

<br />

⎡ 2<br />

⎣ Nu + cos<br />

k=1 q= cos<br />

<br />

n p<br />

m p<br />

+ ωvRv,k cos ωuRu,k ωvI<br />

n p<br />

v,k<br />

σ2 (νu) ω2 u +2ρm n<br />

u,v ωuωv + σ2 (νv) ω2 v<br />

<br />

⎤<br />

m q n q<br />

ωuRu,k + ωvI<br />

<br />

v,k ⎦<br />

m p n p<br />

ωuIu,k ωvRv,k dωu<br />

ωu<br />

<br />

.<br />

(E.51)<br />

<br />

dωv<br />

ωv .<br />

(E.52)


Appendix F<br />

Expressions of U u,DD <strong>de</strong>rived<br />

in the reciprocal space<br />

The general expressions of the mean Uu,DD of the error signal u m u,DD driving<br />

a multiuser DD phase recovery loop, obtained in the reciprocal space,<br />

are presented in this appendix for both BPSK- and QPSK-modulated data<br />

symbols.<br />

F.1 BPSK modulation<br />

Inserting the results of Appendix E into (5.8) leads to<br />

U BPSK<br />

u,DD (∆)<br />

= 1<br />

π<br />

+<br />

p=<br />

ej∆ux p <br />

u,u<br />

+<br />

N0x0 u,u<br />

4EuT ω2 <br />

a sin (ωaR p u,u)<br />

<br />

exp<br />

+<br />

cos (ωaR q u,u)<br />

q=<br />

q=p<br />

Nu<br />

<br />

l=1<br />

l=u<br />

+<br />

q=<br />

cos<br />

<br />

ωaR q<br />

u,l<br />

dωa<br />

ωa


192 Expressions of Uu,DD <strong>de</strong>rived in the reciprocal space<br />

+ 1<br />

π<br />

+ 1<br />

π 2<br />

Nu <br />

l=1<br />

l=u<br />

Nu <br />

l=1<br />

l=u<br />

+<br />

p=<br />

+<br />

p=<br />

<br />

El<br />

Eu <br />

<br />

e j(δl,u+∆u)<br />

<br />

p<br />

xu,l + <br />

exp<br />

N0x0 u,u<br />

4EuT ω2 <br />

a sin ωaR p<br />

<br />

u,l<br />

+ <br />

cos ωaR q<br />

<br />

u,l<br />

q=<br />

q=p<br />

Nu <br />

k=1<br />

k=l<br />

+<br />

q=<br />

<br />

cos ωaR q<br />

<br />

dωa<br />

u,k ωa<br />

<br />

El<br />

Eu <br />

<br />

e j(δl,u+∆u ∆l) p<br />

x<br />

+<br />

+<br />

⎧<br />

⎪⎨<br />

exp<br />

⎪⎩<br />

Nu <br />

k=1 q=<br />

N0<br />

4T<br />

+<br />

⎡<br />

⎢<br />

⎣<br />

cos<br />

u,l<br />

<br />

x0 u,u<br />

Eu ω2 a<br />

+2 (xpu,v) Ô ωaωb<br />

EuEl<br />

+ x0 l,l<br />

El ω2 b<br />

<br />

ωR q<br />

u,k<br />

⎤⎫<br />

⎥⎪⎬<br />

⎥<br />

⎦<br />

⎪⎭<br />

<br />

p q<br />

+ ωbRl,k dωa<br />

ωa<br />

dωb<br />

ωb<br />

.<br />

(F.1)<br />

The writing of (F.1) is not appropriate for numerical integration due to the<br />

presence of the integration variable in the <strong>de</strong>nominator of the integrand.<br />

Applying a classic change of variable (Ω =lnω), another writing of (F.1)is<br />

obtained, avoiding this shortcoming<br />

U BPSK<br />

u,DD (∆)<br />

= 2<br />

π<br />

+<br />

p=<br />

<br />

ej∆u <br />

m p<br />

xu,u + <br />

exp<br />

N0x0 u,u<br />

4EuT e2Ωa<br />

<br />

sin eΩa <br />

ej∆ux + <br />

cos eΩa <br />

ej∆u <br />

m q<br />

x<br />

q=<br />

q=p<br />

Nu<br />

<br />

l=1<br />

l=u<br />

+<br />

q=<br />

u,u<br />

<br />

cos eΩa <br />

El<br />

Eu <br />

<br />

m p<br />

u,u<br />

<br />

e j(δl,u+∆u) x m q<br />

u,l<br />

<br />

dΩa


F.1 BPSK modulation 193<br />

+ 2<br />

π<br />

+ 2<br />

π 2<br />

2<br />

π 2<br />

Nu <br />

l=1<br />

l=u<br />

Nu <br />

l=1<br />

l=u<br />

Nu <br />

l=1<br />

l=u<br />

p=<br />

+<br />

+<br />

p=<br />

+<br />

p=<br />

<br />

El<br />

Eu <br />

<br />

ej(δl,u+∆u) x<br />

+ <br />

exp<br />

N0x0 u,u<br />

<br />

sin eΩa <br />

El<br />

+<br />

q=<br />

q=p<br />

Nu<br />

<br />

k=1<br />

k=l<br />

cos<br />

+<br />

q=<br />

<br />

El<br />

Eu <br />

<br />

+<br />

+<br />

<br />

El<br />

Eu <br />

<br />

+<br />

+<br />

m p<br />

u,l<br />

4EuT e2Ωa<br />

Eu <br />

<br />

e Ωa<br />

<br />

<br />

<br />

ej(δl,u+∆u) <br />

m p<br />

xu,l <br />

ej(δl,u+∆u) x<br />

El<br />

Eu <br />

<br />

cos eΩa <br />

Ek<br />

Eu <br />

<br />

ej(δl,u+∆u ∆l) m p<br />

x<br />

⎧ u,l<br />

x0 u,u<br />

Eu e2Ωa<br />

Nu <br />

k=1 q=<br />

⎪⎨<br />

exp<br />

⎪⎩<br />

+<br />

⎪⎨<br />

exp<br />

⎪⎩<br />

N0<br />

4T<br />

<br />

m q<br />

u,l<br />

<br />

e j(δk,u+∆u) x m q<br />

u,k<br />

⎡<br />

⎢<br />

p<br />

⎢ (xm<br />

u,l )<br />

⎢ +2 Ô e<br />

⎣ EuEl<br />

(Ωa+Ωb)<br />

⎧<br />

⎪⎨<br />

cos<br />

+ e<br />

⎪⎩<br />

Ωb<br />

ej(δl,u+∆u ∆l) m p<br />

x<br />

⎧ u,l<br />

x0 u,u<br />

Eu e2Ωa<br />

Nu <br />

+<br />

k=1 q=<br />

N0<br />

4T<br />

⎧<br />

⎡<br />

⎢<br />

⎣<br />

⎪⎨<br />

cos<br />

⎪⎩<br />

+ x0 l,l<br />

El e2Ωb<br />

eΩa <br />

Ek<br />

Eu <br />

e j(δk,u+∆u) m q<br />

xu,k <br />

Ek<br />

El <br />

e j(δk,l+∆l) p q<br />

xl,k 2<br />

<br />

<br />

<br />

p<br />

(xm<br />

u,l )<br />

Ô e<br />

EuEl<br />

(Ωa+Ωb)<br />

+ x0 l,l<br />

El e2Ωb<br />

eΩa <br />

Ek<br />

Eu <br />

e j(δk,u+∆u) m q<br />

xu,k eΩb <br />

Ek<br />

El <br />

e j(δk,l+∆l) p q<br />

xl,k <br />

dΩa<br />

⎤⎫<br />

⎥⎪⎬<br />

⎥<br />

⎦<br />

⎪⎭<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

dΩadΩb<br />

<br />

<br />

⎤⎫<br />

⎥⎪⎬<br />

⎥<br />

⎦<br />

⎪⎭<br />

⎫<br />

⎪⎬<br />

⎪⎭<br />

dΩadΩb.<br />

(F.2)


194 Expressions of Uu,DD <strong>de</strong>rived in the reciprocal space<br />

F.2 QPSK modulation<br />

Similarly, inserting the results of Appendix E into (5.10) leads to<br />

QP SK<br />

Uu,DD (∆)<br />

= 1<br />

π<br />

+<br />

p=<br />

+ 1<br />

π<br />

+ 1<br />

π 2<br />

Nu <br />

l=1<br />

l=u<br />

Nu <br />

l=1<br />

l=u<br />

ej∆ux p <br />

u,u<br />

+<br />

+<br />

p=<br />

+<br />

p=<br />

N0x0 u,u<br />

4EuT ω2 <br />

a sin (ωaR p u,u)cos(ωaI p u,u)<br />

<br />

exp<br />

+<br />

cos (ωaR q u,u)cos(ωaI q u,u)<br />

q=<br />

q=p<br />

Nu<br />

<br />

+<br />

l=1 q=<br />

l=u<br />

<br />

El<br />

Eu <br />

+<br />

<br />

cos ωaR q<br />

<br />

u,l cos ωaI q<br />

<br />

dωa<br />

u,l ωa<br />

<br />

e j(δl,u+∆u)<br />

<br />

p<br />

xu,l <br />

exp<br />

N0x0 u,u<br />

4EuT ω2 <br />

a sin ωaR p<br />

<br />

u,l cos<br />

+ <br />

cos ωaR q<br />

<br />

u,l cos ωaI q<br />

<br />

u,l<br />

q=<br />

q=p<br />

Nu <br />

k=1<br />

k=l<br />

+<br />

q=<br />

<br />

El<br />

Eu <br />

<br />

e j(δl,u+∆u ∆l) p<br />

x<br />

+<br />

+<br />

⎧<br />

⎪⎨<br />

exp<br />

⎪⎩<br />

Nu <br />

k=1 q=<br />

ωaI p<br />

u,l<br />

<br />

cos ωaR q<br />

<br />

u,k cos ωaI q<br />

<br />

dωa<br />

u,k ωa<br />

+<br />

N0<br />

4T<br />

<br />

u,l<br />

x0 u,u<br />

Eu ω2 a<br />

+2 (xp<br />

Ô u,l)<br />

ωaωb<br />

EuEl<br />

⎡<br />

⎢<br />

⎣<br />

+ x0 l,l<br />

El ω2 ⎤⎫<br />

⎥⎪⎬<br />

⎥<br />

⎦<br />

⎪⎭<br />

b<br />

<br />

cos ωR q<br />

<br />

p q<br />

u,k + ωbRl,k <br />

cos ωI q<br />

<br />

p q<br />

u,k + ωbIl,k dωa<br />

ωa<br />

dωb<br />

ωb


F.2 QPSK modulation 195<br />

1<br />

π 2<br />

Nu <br />

l=1<br />

l=u<br />

+<br />

p=<br />

<br />

El<br />

Eu <br />

<br />

e j(δl,u+∆u ∆l) p<br />

x<br />

+<br />

+<br />

⎧<br />

⎪⎨<br />

exp<br />

⎪⎩<br />

Nu <br />

+<br />

k=1 q=<br />

N0<br />

4T<br />

<br />

⎡ u,l<br />

x<br />

⎢<br />

⎣<br />

0 u,u<br />

Eu ω2 a<br />

2 (xp<br />

Ô u,l)<br />

ωaωb<br />

EuEl<br />

+ x0 l,l<br />

El ω2 ⎤⎫<br />

⎥⎪⎬<br />

⎥<br />

⎦<br />

⎪⎭<br />

b<br />

<br />

cos ωI q<br />

<br />

p q<br />

u,k + ωbR<br />

<br />

l,k<br />

cos ωI q<br />

<br />

p q<br />

u,k ωbIl,k dωb<br />

ωb .<br />

dωa<br />

ωa<br />

(F.3)


Appendix G<br />

COST 207 Channel Mo<strong>de</strong>ls<br />

The COST 207 channel mo<strong>de</strong>ls are tapped <strong>de</strong>lay lines mo<strong>de</strong>lling the behaviour<br />

of a mobile radio channel in several typical environments. Usually,<br />

one distinguishes the Rural Area (RA), the Typical Urban (TU), and<br />

the Hilly Terrain (HT) environments. Values and spacing of their taps are<br />

given in [97, Section 2.4.4]. Their impulse responses are illustrated hereafter.<br />

Power [W]<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

COST 207 Rural Area (RA) channel impulse response<br />

−1 0 1 2 3 4 5 6<br />

x 10 −7<br />

0<br />

Time [s]<br />

Figure G.1: The Rural Area (RA) channel mo<strong>de</strong>l is ma<strong>de</strong> of 6 taps and its<br />

power <strong>de</strong>lay profile spreads over 0.5 µs.


198 COST 207 Channel Mo<strong>de</strong>ls<br />

Power [W]<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

COST 207 Typical Urban (TU) channel impulse response<br />

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5<br />

x 10 −6<br />

0<br />

Time [s]<br />

Figure G.2: The Typical Urban (TU) channel mo<strong>de</strong>l is ma<strong>de</strong> of 12 taps and<br />

its power <strong>de</strong>lay profile spreads over 5 µs<br />

Power [W]<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

COST 207 Hilly Terrain (HT) channel impulse response<br />

0 1 2<br />

x 10 −5<br />

0<br />

Time [s]<br />

Figure G.3: The Hilly Terrain (HT) channel mo<strong>de</strong>l is ma<strong>de</strong> of 12 taps and<br />

its power <strong>de</strong>lay profile spreads over 20 µs.


Appendix H<br />

Curriculum vitae<br />

Laurent Schumacher<br />

Born in Mons, Belgium, on April 28th, 1971.<br />

Education<br />

1994 – 1999 Ph.D. in Applied Sciences – Université catholique <strong>de</strong><br />

Louvain<br />

Thesis: About Maximum-Likelihood Phase Estimation in<br />

DS-CDMA Communication Systems<br />

Supervisor: Prof. L. Van<strong>de</strong>ndorpe.<br />

1988 – 1993 Electrical Engineer – Faculté Polytechnique <strong>de</strong> Mons,<br />

Belgium<br />

Orientation: Telecommunications<br />

Masters thesis: Segmentation of cursive writing<br />

Supervisor: Prof. H. Leich.<br />

Professional experience<br />

October 1995 –<br />

September 1999<br />

October 1994 –<br />

September 1995<br />

Degree candidate, Fonds National <strong>de</strong> la Recherche<br />

Scientifique, Université catholique <strong>de</strong> Louvain,<br />

Communication and Remote Sensing Laboratory.<br />

Scholar, Fonds National <strong>de</strong> la Recherche Scientifique,<br />

Université catholique <strong>de</strong> Louvain, Communication<br />

and Remote Sensing Laboratory.


200 Curriculum vitae<br />

August 1993 –<br />

September 1994<br />

July 1992 –<br />

September 1992<br />

August 1991 –<br />

September 1991<br />

Publications<br />

Refereed conference papers<br />

Research assistant, Université catholique <strong>de</strong><br />

Louvain, Communication and Remote Sensing<br />

Laboratory.<br />

Trainee, IBM Danmark A/S.<br />

Trainee, Generale Bank, Human Resources and IT<br />

Departments.<br />

L. Schumacher and L. Van<strong>de</strong>ndorpe, ”Maximum likelihood joint phase estimators<br />

in CDMA communications systems”, Proc. IEEE Third Symposium<br />

on Communications and Vehicular Technology in the Benelux, Eindhoven, The<br />

N<strong>et</strong>herlands, October 1995, pp. 76-82.<br />

L. Schumacher and L. Van<strong>de</strong>ndorpe, ”Open loop analysis of maximumlikelihood<br />

<strong>de</strong>cision-directed phase estimation in CDMA communication<br />

systems with QPSK modulation”, Proc. IEEE Fourth Symposium on Communication<br />

and Vehicular Technology in the Benelux, Ghent, Belgium, October<br />

1996, pp. 114-121.<br />

L. Van<strong>de</strong>ndorpe and L. Schumacher, ”Maximum likelihood data-ai<strong>de</strong>d<br />

phase estimators in CDMA communication systems with QPSK modulation”,<br />

Proc. IEEE Globecom’96 Communication Theory - Mini-Conference,<br />

London, United Kingdom, November 17-22, 1996, pp. 219-223.<br />

L. Schumacher and L. Van<strong>de</strong>ndorpe, ”MAI Mitigation in DA ML Carrier<br />

Phase Recovery Loops for DS-CDMA Systems”, Proc. IEEE Vehicular Technology<br />

Conference VTC 1999-Fall, Amsterdam, The N<strong>et</strong>herlands, September<br />

19-22, 1999, pp. 1850-1854.<br />

Submitted papers<br />

L. Schumacher and L. Van<strong>de</strong>ndorpe, ”MAI Mitigation in DA ML Carrier<br />

Phase Recovery Loops for DS-CDMA Systems”, submitted to IEEE Transactions<br />

on Communications, April 1999.


201<br />

L. Schumacher and L. Van<strong>de</strong>ndorpe, ”Performance study of DD ML Phase<br />

Estimators for DS-CDMA Communications Systems”, submitted to EU-<br />

SIPCO 2000 - European Signal Processing Conference, Tampere, Finland, September<br />

4-9, 2000.


Bibliography<br />

[1] Gil<strong>de</strong>r Technology Group Inc. Gil<strong>de</strong>r: The Gil<strong>de</strong>r Technology Report.<br />

[online]. No Date [Cited 6 August 1999]. Available from:<br />

.<br />

[2] George Gil<strong>de</strong>r. Telecosm and Beyond: Over the Paradigm Cliff. Forbes<br />

ASAP, February 1997. [Cited 9 August 1999] Available from:<br />

.<br />

[3] Bernard Guidon. Next-generation telecommunication n<strong>et</strong>works<br />

take shape. Telecommunications News, September<br />

1998. [Cited 9 August 1999] Available from:<br />

.<br />

[4] Roger L. P<strong>et</strong>erson, Rodger E. Ziemer, and David E. Borth. Introduction<br />

to Spread Spectrum Communications. Prentice Hall, 1995.<br />

[5] Tero Ojanperä and Ramjee Prasad. An Overview of Air Interface<br />

Multiple Access for IMT-2000/UMTS. IEEE Communications<br />

Magazine, 36(9):82–95, September 1998.<br />

[6] Raymond L. Pickholtz, Laurence B. Milstein, and Donald L.<br />

Schilling. Spread Spectrum for Mobile Communications. IEEE<br />

Transactions on Vehicular Technology, 40(2):313–322, 1991.<br />

[7] John Proakis. Digital Communications. Mac Graw Hill, 1992.<br />

[8] Ricardo De Gau<strong>de</strong>nzi, Filippo Gian<strong>et</strong>ti, and Marco Luise. Signal<br />

Synchronization for Direct-Sequence Co<strong>de</strong>-Division Multiple Access<br />

Radio Mo<strong>de</strong>ms. European Transactions on Telecommunications,<br />

9(1):73–89, February 1998.<br />

[9] Andrew J. Viterbi. CDMA - Principles of Spread Spectrum Communication.<br />

Addison-Wesley, 1995.


204 BIBLIOGRAPHY<br />

[10] Dilip V. Sarwate and Michael B. Pursley. Crosscorrelation Properties<br />

of Pseudorandom and Related Sequences. Proceedings of the IEEE,<br />

68(5):593–619, May 1980.<br />

[11] William C. Y. Lee. Overview of Cellular CDMA. IEEE Transactions<br />

on Vehicular Technology, 40(2):291–302, May 1991.<br />

[12] Eric Lawrey. The suitability of OFDM as a modulation technique for<br />

wireless telecommunications, with a CDMA comparison. Master’s<br />

thesis, James Cook University of North Queensland, Australia, 1997.<br />

[online]. 18 October 1997 [Cited 13 August 1999]. Available from:<br />

.<br />

[13] Klein S. Gilhousen, Irwin M. Jacobs, Roberto Pavodani, Andrew J.<br />

Viterbi, Lindsay A. Weaver, Jr., and Charles E. Wheatley III. On the<br />

Capacity of a Cellular CDMA System. IEEE Transactions on Vehicular<br />

Technology, 40:303–312, May 1991.<br />

[14] Ruxandra Lupas and Sergio Verdú. Linear Multiuser D<strong>et</strong>ectors for<br />

Synchronous Co<strong>de</strong>-Division Multiple-Access Channels. IEEE Transactions<br />

on Information Theory, 35(1):123–136, January 1989.<br />

[15] ISO. ISO homepage. [online]. No date [Cited 3 August 1999]. Available<br />

from: .<br />

[16] ITU. International Mobile Telecommunications IMT-2000. [online].<br />

28 July 1999 [Cited 3 August 1999]. Available from:<br />

.<br />

[17] Ermanno Berruto, Mikael Gudmundson, Raffaele Menolascino,<br />

Werner Mohr, and Marta Pizarroso. Research Activities on UMTS<br />

Radio Interface, N<strong>et</strong>work Architectures, and Planning. IEEE Communications<br />

Magazine, 46(2):82–95, February 1998.<br />

[18] CDG. CDMA Development Group. [online]. No date [Cited 3 August<br />

1999]. Available from: .<br />

[19] Douglas N. Knisely, Sarath Kumar, Subhasis Laha, and Sanjiv<br />

Nanda. Evolution of Wireless Data Services: IS-95 to cdma2000.<br />

IEEE Communications Magazine, 36(10):140–149, October 1998.<br />

[20] ETSI. ETSI - Standardizing Telecommunications Products and Services.<br />

[online]. No date [Cited 3 August 1999]. Available from:<br />

.


BIBLIOGRAPHY 205<br />

[21] Barry Miller. Satellites Free the Mobile Phone. IEEE Spectrum,<br />

35(3):26–35, March 1998.<br />

[22] A. Kajiwara. Mobile Satellite CDMA System Robust to Doppler<br />

Shift. IEEE Transactions on Vehicular Technology, 44(3):480–486, August<br />

1995.<br />

[23] Arild Flystveit and Arvid Bertheau Johannessen. Global Mobile Personal<br />

Communications by Satellite. Telektronikk, 94(2):22–33, 1998.<br />

[24] Iridium LLC. :: Welcome to Iridium ::. [online].<br />

No date [Cited 3 August 1999]. Available from:<br />

.<br />

[25] Globalstar. Globalstar. [online]. No date [Cited 6 April 1999]. Available<br />

from: .<br />

[26] Ellipso. Ellipso. [online]. No date [Cited 3 August 1999]. Available<br />

from: .<br />

[27] Tele<strong>de</strong>sic LLC. Tele<strong>de</strong>sic. [online]. July 1999 [Cited 3 August 1999].<br />

Available from: .<br />

[28] Skybridge. SkyBridge. [online]. No date [Cited 3 August 1999].<br />

Available from: .<br />

[29] Kaveh Pahlavan, Thomas H. Probert, and Mitchell E. Chase.<br />

Trends in Local Wireless N<strong>et</strong>works. IEEE Communications Magazine,<br />

33(3):88–95, March 1995.<br />

[30] Kaveh Pahlavan and Prashant Krishnamurthy. Wi<strong>de</strong>band Local<br />

Access: Wireless LAN and Wireless ATM. IEEE Communications<br />

Magazine, 35(11):34–40, November 1997.<br />

[31] Richard O. LaMaire, Arvind Krishna, Pravin Bhagwat, and James<br />

Panian. Wireless LANs and Mobile N<strong>et</strong>working: Standards and Future<br />

Directions. IEEE Communications Magazine, 34(8):86–94, August<br />

1996.<br />

[32] Luis Correia and Ramjee Prasad. An Overview of Wireless Broadband<br />

Communications. IEEE Communications Magazine, 35(1):28–33,<br />

January 1997.


206 BIBLIOGRAPHY<br />

[33] Norihiko Morinaga, Masao Nakagawa, and Ryuji Kohno. New Concepts<br />

and Technologies for Achieving Highly Reliable and High-<br />

Capacity Multimedia Wireless Communications Systems. IEEE<br />

Communications Magazine, 35(1):34–40, January 1997.<br />

[34] ETSI. BRAN Homepage. [online]. 2 July 1999 [Cited 3 August 1999].<br />

Available from: .<br />

[35] Dan Spoelman and Gary Law. Mapping Out A Mo<strong>de</strong>m Strategy.<br />

CED Magazine, 24(4):72–92, April 1998.<br />

[36] Terayon. S-CDMA, The Upstream Advantage. [online].<br />

1998 [Cited 3 August 1999]. Available from:<br />

.<br />

[37] Matt Brandt. IEEE 802.14 Information. [online].<br />

1998 [Cited 3 August 1999]. Available from:<br />

.<br />

[38] Sergio Verdú. Minimum Probability of Error for Asynchronous<br />

Gaussian Multiple-Access Channels. IEEE Transactions on Information<br />

Theory, 32(1):85–96, January 1986.<br />

[39] Ruxandra Lupas and Sergio Verdú. Near-Far Resistance of Multiuser<br />

D<strong>et</strong>ectors in Asynchronous Channels. IEEE Transactions on<br />

Communications, 38(4):496–508, April 1990.<br />

[40] Stefan Parkvall, Erik G. Ström, and Björn Ottersten. The Impact of<br />

Timing Errors on the Performance of Linear DS-CDMA Receivers.<br />

IEEE Journal on Selected Areas in Communications, 14(8):1660–1668,<br />

October 1996.<br />

[41] Alexandra Duel-Hallen, Jack Holtzman, and Zoran Zvonar. Multiuser<br />

D<strong>et</strong>ection for CDMA Systems. IEEE Personal Communications,<br />

2(2):46–58, April 1995.<br />

[42] Shimon Moshavi. Multi-User D<strong>et</strong>ection for DS-CDMA Communications.<br />

IEEE Communications Magazine, pages 124–136, October 1996.<br />

[43] Geert Leus and Marc Moonen. Multi-User D<strong>et</strong>ection in Frequency-<br />

Selective Fading Channels. In Seminar on Digital Signal Processing and<br />

Wireless Communications, 28 May 1999, Leuven, Belgium, May 1999.


BIBLIOGRAPHY 207<br />

[44] Michael Honig, Upamanyu Madhow, and Sergio Verdú. Blind Adaptive<br />

Multiuser D<strong>et</strong>ection. IEEE Transactions on Information Theory,<br />

41(4):944–960, July 1995.<br />

[45] Franco Mazzenga and Giovanni Emanuele Corazza. Blind Least-<br />

Squares Estimation of Carrier Phase, Doppler Shift, and Doppler<br />

Rate for m-PSK Burst Transmission. IEEE Communication L<strong>et</strong>ters,<br />

2(3):73–75, March 1998.<br />

[46] Kaj Go<strong>et</strong>hals. DA Chip Synchronizers for Bandlimited DS/SS M-<br />

PSK Signals Using CDMA on Mobile Satellite Communications<br />

Channels. In IEEE International Conference on Communications ICC’94,<br />

1-5 May 1994, New-Orleans, United States of America, volume 2, pages<br />

1150–1154, May 1994.<br />

[47] Kaj Go<strong>et</strong>hals and Marc Moeneclaey. NDA Chip Synchronizers for<br />

Bandlimited DS/SS Signals Using CDMA on Nonselective Fading<br />

Channels. In 3rd European Conference on Satellite Communications, 2-4<br />

November 1993, Manchester, United Kingdom, pages 336–340, November<br />

1993.<br />

[48] Stephen E. Bensley and Behnaam Aazhang. Subspace-Based Estimation<br />

of Multipath Channel Param<strong>et</strong>ers for CDMA Communication<br />

Systems. In Proceedings IEEE Telecommunications Conference, Communication<br />

theory Mini-Conference Record, pages 154–158, 1994.<br />

[49] Eric G. Ström, Stefan Parkvall, Scott L. Miller, and Björn E. Ottersten.<br />

Propagation Delay Estimation in Asynchronous Direct-<br />

Sequence Co<strong>de</strong>-Division Multiple Access Systems. IEEE Transactions<br />

on Communications, 44(1):84–93, January 1996.<br />

[50] Meir Fe<strong>de</strong>r and Josko A. Catipovic. Algorithms For Joint Channel<br />

Estimation and Data Recovery - Application to Equalization in<br />

Un<strong>de</strong>rwater Communications. IEEE Journal of Oceanic Engineering,<br />

16(1):42–55, January 1991.<br />

[51] Todd K. Moon. The Expectation-Maximization Algorithm. IEEE<br />

Signal Processing Magazine, pages 47–60, November 1996.<br />

[52] Bernard H. Fleury, Martin Tschudin, Ralf Hed<strong>de</strong>rgott, Dirk Dalhaus,<br />

and Klaus Ingeman Per<strong>de</strong>sen. Channel Param<strong>et</strong>er Estimation in Mobile<br />

Radio Environments Using the SAGE Algorithm. IEEE Journal<br />

on Selected Areas in Communications, 17(3):434–450, March 1999.


208 BIBLIOGRAPHY<br />

[53] Simon Haykin. Adaptive Filter Theory. Prentice Hall International,<br />

1991.<br />

[54] Li-Chung Chu and Urbashi Mitra. Analysis of MUSIC-Based Delay<br />

Estimators for Direct-Sequence Co<strong>de</strong>-Division Multiple-Access Systems.<br />

IEEE Transactions on Communications, 147(1):133–138, January<br />

1999.<br />

[55] Stephen E. Bensley and Behnaam Aazhang. Subspace-Based Channel<br />

Estimation for Co<strong>de</strong> Division Multiple Access Communication<br />

Systems. IEEE Transactions on Communications, 44(8):1009–1020, August<br />

1996.<br />

[56] Eric G. Ström, Stefan Parkvall, Scott L. Miller, and Björn E. Ottersten.<br />

DS-CDMA Synchronization in Time-Varying Fading Channels.<br />

IEEE Journal on Selected Areas in Communications, 14(8):1636–<br />

1642, August 1996.<br />

[57] Hui Liu and Guanghan Xu. A Subspace-M<strong>et</strong>hod for Signature<br />

Waveform Estimation in Synchronous CDMA Systems. IEEE Transactions<br />

on Communications, 44(10):1346–1354, October 1996.<br />

[58] Stephen E. Bensley and Behnaam Aazhang. Maximum-Likelihood<br />

Synchronization of a Single User for Co<strong>de</strong>-Division Multiple-Acces<br />

Communication Systems. IEEE Transactions on Communications,<br />

46(3):392–399, March 1998.<br />

[59] Steven M. Kay. Fundamentals of Statistical Signal Processing. Prentice<br />

Hall International, 1993.<br />

[60] Jerry M. Men<strong>de</strong>l. Lessons in Estimation Theory for Signal Processing<br />

Communications and Control. Prentice Hall, 1995.<br />

[61] Teng Joon Lim and Lars K. Rasmussen. Adaptive Symbol and Param<strong>et</strong>er<br />

Estimation in Synchronous Multiuser CDMA D<strong>et</strong>ectors. IEEE<br />

Transactions on Communications, 45(2):213–220, February 1997.<br />

[62] Ronald A. Iltis. Joint Estimation of PN Co<strong>de</strong> Delay and Multipath<br />

Using the Exten<strong>de</strong>d Kalman Filter. IEEE Transactions on Communications,<br />

38(10):1677–1685, October 1990.<br />

[63] Ronald A. Iltis. An EKF-Based Joint Estimator for Interference, Multipath,<br />

and Co<strong>de</strong> Delay in a DS Spread-Spectrum Receiver. IEEE<br />

Transactions on Communications, 42(2/3/4):1288–1299, April 1994.


BIBLIOGRAPHY 209<br />

[64] Alfred W. Fuxjaeger and Roland A. Iltis. Adaptive Param<strong>et</strong>er Estimation<br />

using Parallel Kalman Filtering for Spread Spectrum Co<strong>de</strong> and<br />

Doppler Tracking. IEEE Transactions on Communications, 42(6):2227–<br />

2230, June 1994.<br />

[65] Todd K. Moon, Zhenhua Xie, Craig K. Rushforth, and Robert T.<br />

Short. Param<strong>et</strong>er Estimation in a Multi-User Communication System.<br />

IEEE Transactions on Communications, 42(8):2553–2560, August<br />

1994.<br />

[66] F. Mazzenga and F. Valataro. Param<strong>et</strong>er Estimation in CDMA Multiuser<br />

D<strong>et</strong>ection Using Cyclostationnary Statistics. Electronic L<strong>et</strong>ters,<br />

32(3):179–181, February 1996.<br />

[67] Alfred W. Fuxjaeger and Roland A. Iltis. Acquisition of Timing and<br />

Doppler-Shift in a Direct-Sequence Spread-Spectrum System. IEEE<br />

Transactions on Communications, 42(10):2870–2880, October 1994.<br />

[68] Bernd Steiner and P<strong>et</strong>er Jung. Optimum and Suboptimum Channel<br />

Estimation for the Uplink of CDMA Mobile Radio Systems with<br />

Joint D<strong>et</strong>ection. European Transactions on Telecommunications, 5(1):39–<br />

50, January 1994.<br />

[69] Hui Liu, Guanghan Xu, Lang Tong, and Thomas Kailath. Recent<br />

<strong>de</strong>velopments in blind channel equalization: From cyclostationarity<br />

to subspaces. Signal Processing, 50(1-2):83–99, April 1996.<br />

[70] Xiaodong Wang and H. Vincent Poor. Blind Equalization and Multiuser<br />

D<strong>et</strong>ection in Dispersive CDMA Channels. IEEE Transactions on<br />

Communications, 46(1):91–103, January 1998.<br />

[71] Emre Akta and Urbashi Mitra. Blind Channel Estimation for Multiuser<br />

CDMA Systems. In ICC ’98 Conference Record, volume 2, pages<br />

1064–1068, June 1998.<br />

[72] Urs Fawer and Behnaam Aazhang. A Multiuser Receiver for Co<strong>de</strong><br />

Division Multiple Access Communications over Multipath Channels.<br />

IEEE Transactions on Communications, 43(2/3/4):1556–1565,<br />

1995.<br />

[73] Andrej Radovic. An Iterative Near-Far Resistant Algorithm for<br />

Joint Param<strong>et</strong>er Estimation in Asynchronous CDMA Systems. In<br />

PIMRC’94, pages 199–203, 1994.


210 BIBLIOGRAPHY<br />

[74] Dirk Dalhaus, Bernard H. Fleury, and Andrej Radović. A Sequential<br />

Algorithm for Joint Param<strong>et</strong>er Estimation and Multiuser D<strong>et</strong>ection<br />

in DS/CDMA Systems with Multipath Propagation. Wireless Personal<br />

Communications, pages 161–178, 1998.<br />

[75] Stefan Parkvall and Erik G. Ström. Param<strong>et</strong>er Estimation and D<strong>et</strong>ection<br />

of DS-CDMA Signal subject to Multipath Propagation. In<br />

Proceedings of IEEE/IEE Workshop on Signal Processing in Multipath Environments,<br />

Glasgow, Scotland, 1995.<br />

[76] Erik G. Ström and Stefan Parkvall. Joint Param<strong>et</strong>er Estimation and<br />

D<strong>et</strong>ection of DS-CDMA Signals in Fading Channels. In Proceedings<br />

IEEE Global Telecommunications Conference, volume 2, pages 1109–<br />

1113, 1995.<br />

[77] Zhenhua Xie, Craig K. Rushforth, Robert T. Short, and Todd K.<br />

Moon. Joint Signal D<strong>et</strong>ection and Param<strong>et</strong>er Estimation in Multiuser<br />

Communications. IEEE Transactions on Communications, 41(7):1208–<br />

1216, August 1993.<br />

[78] John R. Barry and Anuj Batra. A Multidimensionnal Phase-Locked<br />

Loop for Blind Equalization of Multi-Input Multi-Output Channels.<br />

In ICC ’96 Conference Record, volume 3, pages 1307–1312, 1996.<br />

[79] Floyd M. Gardner. Phaselock Techniques. John Wiley & Sons, Inc.,<br />

1979.<br />

[80] L. E. Franks. Carrier and Bit Synchronization in Data Communication<br />

- A Tutorial Review. IEEE Transactions on Communications,<br />

28(8):1107–1120, August 1980.<br />

[81] Umberto Mengali and Aldo N. D’Andrea. Synchronization Techniques<br />

for Digital Receivers. Plenum press, New York, 1997.<br />

[82] Andrew J. Viterbi and Audrey M. Viterbi. Nonlinear Estimation<br />

of PSK-Modulated Carrier Phase with Application to Burst Digital<br />

Transmission. IEEE Transactions on Information Theory, 29(4):543–551,<br />

July 1983.<br />

[83] Floyd Gardner. Demodulator Reference Recovery Techniques<br />

Suited for Digital Implementation. ESTEC Contract No.<br />

6487/86/NL/DG, European Space Agency, August 1988.


BIBLIOGRAPHY 211<br />

[84] Thierry Jesupr<strong>et</strong>, Marc Moeneclaey, and Gerd Ascheid. Digital Demodulator<br />

Synchronization. ESTEC Contract No. 8437/89/NL/RE,<br />

European Space Agency, June 1991.<br />

[85] Heinrich Meyr, Marc Moeneclaey, and Stefan A. Fechtel. Digital<br />

Communication Receivers - Synchronization, Channel Estimation, and<br />

Signal Processing. Wiley Series in Telecommunications and Signal<br />

Processing. John Wiley & Sons, Inc., 1998.<br />

[86] Andrew J. Viterbi. Principles of Coherent Communications. McGraw-<br />

Hill Book Company, 1966.<br />

[87] Ricardo De Gau<strong>de</strong>nzi, Tobias Gar<strong>de</strong>, and Vieri Vanghi. Performance<br />

Analysis of Decision-Directed Maximum-Likelihood Phase Estimators<br />

for M-PSK Modulated Signals. IEEE Transactions on Communications,<br />

43(12):3090–3100, December 1995.<br />

[88] Dae Sun Oh, Won Gi Jeon, Yong Soo Cho, Hyung Woon Park, and<br />

Ki Ho Kim. Convergence Analysis of a PLL for a Digital Recording<br />

Channel with an Adaptive Partial Response Equalizer. In IEEE<br />

Globecom ’96, pages 979–983, November 1996.<br />

[89] M. O˘guz Sunay and P<strong>et</strong>er J. McLane. Calculating Error Probabilities<br />

for DS CDMA Systems: When Not to Use the Gaussian Approximation.<br />

In IEEE Globecom ’96, pages 1744–1748, November 1996.<br />

[90] Erik G. Ström and Fredrik Malmsten. Maximum Likelihood Synchronization<br />

of DS-CDMA Signals Transmitted over Multipath<br />

Channels. In ICC ’98 Conference Record, volume 3, pages 1546–1550,<br />

June 1998.<br />

[91] Michel B. Jeruchim, Philip Balaban, and K. Sam Shanmugan. Simulation<br />

of Communication Systems. Plenum Press, 1992.<br />

[92] Fu-Chun Zheng and Stephen K. Barton. On the Performance of<br />

Near-Far Resistant CDMA D<strong>et</strong>ectors in the Presence of Synchronization<br />

Errors. IEEE Transactions on Communications, 43(12):3037–3045,<br />

December 1995.<br />

[93] Sergio Verdú. Optimum Multiuser Asymptotic Efficiency. IEEE<br />

Transactions on Communications, 34(9):890–897, September 1986.


212 BIBLIOGRAPHY<br />

[94] Bas W’t Hart, Richard D. J. Van Nee, and Ramjee Prasad. Performance<br />

Degradation Due to Co<strong>de</strong> Tracking Errors in Spread-Spectrum<br />

Co<strong>de</strong>-Division Multiple-Access Systems. IEEE Journal on Selected<br />

Areas in Communications, 14(8):1669–1679, October 1996.<br />

[95] Rick Cameron and Brian Woerner. Performance Analysis of CDMA<br />

with Imperfect Power Control. IEEE Transactions on Communications,<br />

44(7):777–808, July 1996.<br />

[96] Wei Huang, Ivan Andonovic, and Masao Nakagawa. PLL Performance<br />

of DS-CDMA Systems in the Presence of Phase Noise, Multiuser<br />

Interference, and Additive Gaussian Noise. IEEE Transactions<br />

on Communications, 46(11):1507–1515, November 1998.<br />

[97] COST 207 Management Committee. COST 207 Digital Land Mobile<br />

Radio Communications - Final Report. Office for Official Publications<br />

of the European Communities, 1989.<br />

[98] Harry L. Van Trees. D<strong>et</strong>ection, Estimation and Modulation Theory.<br />

Wiley, 1968.<br />

[99] M. H. Meyers and L. E. Franks. Joint Carrier Phase and Symbol<br />

Timing Recovery for PAM Systems. IEEE Transactions on Communications,<br />

28(8):1121–1129, August 1980.<br />

[100] Marc Moeneclaey and Geert <strong>de</strong> Jonghe. Tracking Performance Comparison<br />

of Two Feedforward ML-Oriented Carrier-In<strong>de</strong>pen<strong>de</strong>nt<br />

NDA Symbol Synchronizers. IEEE Transactions on Communications,<br />

40:1423–1425, September 1992.<br />

[101] Thomas Alberty. Frequency Domain Interpr<strong>et</strong>ation of the Cramér-<br />

Rao Bound for Carrier and Clock Synchronization. IEEE Transactions<br />

on Communications, 43(2/3/4):1185–1191, April 1995.<br />

[102] Aldo N. D’Andrea, Umberto Mengali, and Ruggero Reggiannini.<br />

The Modified Cramér-Rao Bound and Its Application to<br />

Synchronization Problems. IEEE Transactions on Communications,<br />

42(2/3/4):1391–1399, March 1994.<br />

[103] Marc Moeneclaey. On the True and the Modified Cramér-Rao<br />

Bounds for the Estimation of a Scalar Param<strong>et</strong>er in the Presence<br />

of Nuisance Param<strong>et</strong>ers. IEEE Transactions on Communications,<br />

46(11):1536–1544, November 1998.


BIBLIOGRAPHY 213<br />

[104] Martin Oer<strong>de</strong>r and Heinrich Meyr. Digital Filter and Square Timing<br />

Recovery. IEEE Transactions on Communications, 36(5):605–612, May<br />

1988.<br />

[105] Kaj Go<strong>et</strong>hals. ML-Oriented Symbol Synchronizers for Mobile Satellite<br />

Communications Using Narrowband M-PSK. In 1st IEEE Symposium<br />

on Communications and Vehicular Technology in the Benelux<br />

SCVT’93, October 1993, Louvain-la-Neuve, Belgium, pages 1.3.1–1.3.8,<br />

October 1993.<br />

[106] Kaj Go<strong>et</strong>hals and Marc Moeneclaey. ML-Oriented DA Symbol Synchronization<br />

for Nonselective Fading Channels Using Imperfect<br />

Channel Gain Estimates. In 4th International Workshop on Digital Signal<br />

Processing Techniques, 26-28 September 1994, London, United Kingdom,<br />

pages 2–8, September 1994.<br />

[107] Marvin K. Simon and Dariush Divsalar. Some New Twists to Problems<br />

Involving the Gaussian Probability Integral. IEEE Transactions<br />

on Communications, 46(2):200–210, February 1998.<br />

[108] Marvin K. Simon and Mohamed-Slim Alouini. A Unified Approach<br />

to the Performance Analysis of Digital Communications over Generalized<br />

Fading Channels. Proceedings of the IEEE, 86(9):1860–1877,<br />

September 1998.<br />

[109] Norman C. Beaulieu. The Evalutation of Error Probabilities for Intersymbol<br />

and Cochannel Interference. IEEE Transactions on Communications,<br />

39(12):1740–1749, December 1991.<br />

[110] O. Shimbo and M. I. Celebiler. The Probability Of Error Due To Intersymbol<br />

Interference And Gaussian Noise In Digital Communication<br />

Systems. IEEE Transactions on Communications, 19(4):115–119, April<br />

1971.<br />

[111] Luc Van<strong>de</strong>ndorpe and Olivier van <strong>de</strong> Wiel. Performance Analysis of<br />

Linear Joint Equalization and Multiple Access Interference Cancellation<br />

for Multitone CDMA. Wireless Personal Communications, 3(1-<br />

2):17–36, February 1996.<br />

[112] Carl W. Helstrom. Calculating Error Probabilities for Intersymbol<br />

and Cochannel Interference. IEEE Transactions on Communications,<br />

34(5):430–435, May 1986.


214 BIBLIOGRAPHY<br />

[113] Saïd Moridi and Hikm<strong>et</strong> Sari. Analysis of Four Decision-Feedback<br />

Carrier Recovery Loops in the Presence of Intersymbol Interference.<br />

IEEE Transactions on Communications, 33(6):543–550, June 1985.<br />

[114] Sami Hinedi and William C. Lindsey. Intersymbol Interference Effects<br />

on BPSK and QPSK Carrier Tracking Loops. IEEE Transactions<br />

on Communications, 38(10):1670–1676, October 1990.<br />

[115] Sami M. Hinedi and Marvin K. Simon. Suppressed Carrier<br />

Synchronizers for ISI Channels. In Communication Theory Mini-<br />

Conference on behalf of IEEE Globecom ’96, pages 62–66, 1996.<br />

[116] I.S. Gradshteyn. Table of Integrals and Products. Aca<strong>de</strong>mic Press, 1965.<br />

[117] W.H. Press, S.A. Teukolsky, W.T. V<strong>et</strong>terling, and B.P. Flannery. Numerical<br />

Recipes in C - The Art of Scientific Computing. Cambridge Univesity<br />

Press, 1992.<br />

[118] A. Erdélyi, W. Magnus, F. Oberh<strong>et</strong>tinger, and F.G. Tricomi. Tables<br />

of Integral Transforms, volume 1. McGraw-Hill Book Company Inc.,<br />

1954.<br />

[119] Sirikiat Ariyavisitakul. Equalization of a Hard-Limited Slowly-<br />

Fading Multipath Signal Using a Phase Equalizer with Time-<br />

Reversal Structure. IEEE Journal on Selected Areas in Communications,<br />

10(3):589–596, April 1992.<br />

[120] F. Oberh<strong>et</strong>tinger. Tabellen zur Fourier Transformation. Springer, Berlin,<br />

1957.<br />

[121] Richard D. Gitlin, Jeremiah F. Hayes, and Stephen B. Weinstein. Data<br />

Communications Principles. Plenum Press, New York, 1992.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!