22.08.2013 Views

Untitled

Untitled

Untitled

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

here, αm and β m represents the multiplicative distortion at the<br />

desired subcarrier and the ICI, respectively. If the channel keeps<br />

invariant during a block period, β m will be zero, which means no<br />

ICI.<br />

In the general case where the multipath cannot be regarded as<br />

time-invariant during a block period, (7) can be expressed in<br />

vector form as:<br />

Y = H X + W<br />

(9)<br />

where Y=[Y 0,…,Y N-1] T , X=[X 0,…,X N-1] T , W=[W 0,…,W N-1] T , and<br />

⎡a0,0 a0,1La0, N −1<br />

⎤<br />

⎢<br />

⎥<br />

a1,0 a1,1 a1,<br />

N −1<br />

H<br />

⎢ L<br />

=<br />

⎥<br />

⎢M O M ⎥<br />

⎢ ⎥<br />

⎢a a a ⎥<br />

⎣ N−1,0 N−1,1 N−1, N−1⎦<br />

here, a m,k in (10) is defined as<br />

( m−k) − j2 π k( L−1)/ N<br />

+ HL1 e ,0 ≤(<br />

m, k) ≤ N −1.<br />

* * * *<br />

= + (14)<br />

⎤<br />

⎦<br />

(10)<br />

( m−k) ( m−k) − j2 π k/ N<br />

amk , = H0 + H1 e + K (11)<br />

−<br />

In this paper, we assume the multipath fading channel is<br />

slowly-varying, so most energy concentrate in the dominant<br />

adjacent subcarriers, the ICI terms which do not significantly<br />

affect Y m in (10) can be ignored. And then by transforming the<br />

matrix H of order N× N to a block-diagonal matrix H * of order<br />

×<br />

(N-q)(q+1) (N-q)(q+1). We obtain:<br />

where,<br />

⎡Aq/2<br />

0 ⎤<br />

⎢<br />

⎥<br />

A<br />

*<br />

q /2+ 1<br />

H =<br />

⎢ ⎥<br />

⎢ O ⎥<br />

⎢ ⎥<br />

⎢⎣0AN−− 1 q/2⎥⎦<br />

⎛an−q/2, n−q/2 L an−q/2,<br />

n L<br />

0 ⎞<br />

⎜ ⎟<br />

⎜an− q/2+ 1, n−q/2 ⎟<br />

⎜ M M ⎟<br />

⎜ ⎟<br />

⎜ann , −q/2<br />

O<br />

⎟<br />

An<br />

= ⎜ ⎟<br />

⎜ O<br />

an+<br />

1, n+ q/<br />

2 ⎟<br />

⎜ M M ⎟<br />

⎜ ⎟<br />

⎜ an+<br />

q/2− 1, n+ q/2⎟<br />

⎜<br />

⎟<br />

0<br />

an+ q/2, n a ⎟<br />

⎝<br />

L L n+ q/2, n+ q/2<br />

⎠<br />

(12)<br />

(13)<br />

We can locate some of the unused subcarriers at the first q/2 and<br />

the last q/2 IFFT grids to get the estimation of all symbols. Then,<br />

the input-output relationship of the multipath channel is expressed<br />

in a vector form as:<br />

where,<br />

Y H X W<br />

*<br />

T<br />

X ⎡xq/2<br />

xq/2+ 1 L xN−1<br />

−q/2<br />

= ⎣<br />

xn = ⎡<br />

⎣Xn−q/2 Xn− q/2+ 1 L X ⎤ n+ q/2⎦<br />

*<br />

T<br />

Y ⎡yq/2 yq/2+ 1 L y ⎤ N−1 −q/2<br />

= ⎣ ⎦<br />

yn = ⎡<br />

⎣Xn−q/2 Xn− q/2+ 1 L X ⎤ n+ q/2⎦<br />

*<br />

T<br />

W ⎡wq/2 wq/2+ 1 L w ⎤ N−1 −q/2<br />

= ⎣ ⎦<br />

wn = ⎡<br />

⎣Wn−q/2 Wn− q/2+ 1 L W ⎤ n+ q/2⎦<br />

T<br />

T<br />

T<br />

(15)<br />

V. FREQUENCY DOMAIN EQUALIZER<br />

A. Zero-Forcing Equalizer<br />

Zero-Forcing equalization compensates both multiplicative<br />

distortion and ICI by multiplying the inverse of , which is the<br />

*<br />

estimated value of H , to (14). The resulting signal can be<br />

expressed as:<br />

* 1 * − % % (16)<br />

X = H Y<br />

where X % is the estimated signal, and<br />

H%<br />

* −1<br />

−1<br />

⎡ A%<br />

0 0 ⎤<br />

⎢ ⎥<br />

−1<br />

⎢ A%<br />

1 ⎥<br />

=<br />

⎢ O ⎥<br />

⎢ ⎥<br />

−1<br />

⎢0 A% ⎣<br />

⎥ N−− 1 q⎦<br />

Also, (16) can be rewritten as<br />

1<br />

X% −<br />

= A% y , q/2≤ n≤ N −1<br />

−q/<br />

2.<br />

(18)<br />

n n n<br />

Finally, the transmitted symbols are estimated by selecting<br />

the elements in the middle of X .<br />

n<br />

%<br />

B. MMSE Equalizer<br />

H %<br />

*<br />

(17)<br />

Using the fact that the ICI power is localized to the dominant<br />

adjacent subcarriers, only a few neighborhood subcarriers can be<br />

used for equalization without much performance penalty. We find<br />

the solution for each desired subcarrier individually[4]. The<br />

problem of MMSE is to find the equalizer coefficient vector<br />

g m=[g m,0,…,g m,q-1] which minimizes the mean-squared error<br />

2<br />

ˆ { m − m }<br />

E X X<br />

where ˆ T<br />

X m = gmy and , y<br />

m y ⎡Y , Y ⎤<br />

where Hm=A m.<br />

The MMSE solution is<br />

m ( m− q/2) L m is then<br />

( m+ q/2)<br />

= ⎣ ⎦<br />

ym = Hmx+ w (19)<br />

m<br />

g R R −<br />

= (20)<br />

1<br />

m X ,<br />

mym ym<br />

assuming H is known, x is a zero-mean i.i.d. random vector with<br />

variance<br />

2<br />

σ x , and w is an AWGN vector with variance<br />

{ } 2<br />

H H<br />

X mym m m<br />

2<br />

σ w ,then<br />

R = E X y = σ xh (21)<br />

m<br />

where hm is the q/2 th column of the matrix H m, i.e.,<br />

Also, we can get<br />

hm = ⎡<br />

⎣am− q/2, q/2 , L , a ⎤<br />

(22)<br />

m+ q/2, q/2⎦<br />

H 2 H 2<br />

{ } σ σ<br />

R = E y y = H H + wIq (23)<br />

ymm m x m m<br />

After inserting (21) and (23) into (20), the q-tap equalizer vector<br />

g m becomes<br />

T

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!