Multi-partition Resource Allocation Mehtod Using Partial Channel ...

Multi-partition resource allocation method
using partial channel matrix
in OFDMA system
Myeong Geol Lee, Yuqin Chen, Sung Jun Lee, Sung Hwan Sohn, Jae Moung Kim
The Graduate School of Information technology & Telecommunications INHA University, Incheon,
Korea
E-Mail : ddongri80@paran.com, river4416@sina.com, rhlek99@nate.com, kittisn@naver.com,
jeakim@inha.ac.kr
Abstract—In OFDMA system, resource allocation methods is an
approach to efficiently improve the spectrum usage. According
to the channel magnitude response each user selects the
subcarriers with the highest channel gain. The existing
algorithms allocate all required partition to users at one time,
which cause increase in the complexity. And when the partition
per user increases, the successful allocation probability will be
decreased rapidly. To support large amount of users, several
procedures are proposed, among which the main idea is to
allocate partition one by one until all users get enough partitions
based on users priority and consider how to avoid to select
channel partition with very serious channel quality gain.
Index terms— Adaptive resource allocation, Orthogonal
Frequency Division Modulation Access, frequency selective
fading, Diversity
I. INTRODUCTION
Before 1990s, Spectrum usage had low economical value.
Due to rapidly increasing demand on broadband wireless
communication. Demand on the spectrum is increased always.
It is difficult to use the resources at the same place and time
if some users are using. This leads to the unsuitable
frequency band to some users and sometimes give the low
coefficient of utilization. To overcome this problem an
efficient dynamic frequency allocation method is required for
the proper usage of the spectrum.
In order to provide high data rate services. Orthogonal
frequency division multiplexing (OFDM) is being considered
as a promising technique due to its ability to overcome
multipath fading. The subcarriers experience deep fading to
one user but may not be to other users. Therefore, to improve
system capacity we allocate the spectrum dynamically to each
users according to the channel quality.
In general case, among the users using the same channel
has difference communication environment owing to distance,
geographical features, and multipath etc. So we have to
allocate resources adaptively considering user’s
communication environment. To efficiently allocating
channel efficiency, In [1] the whole subcarriers are divided
into a number of partition and each user acts in parallel and
attempts to select the partition with highest average channel
gain.
In [2], method is proposed to divide the subcarriers into a
large number of partitions and allocate several smaller
partitions instead of one larger partition to each user. The
users channel quality gain rows are copied to support
different data rate of user’s matrix. In this method, K by K
matrix is computed to solve conflict among users and the
computation overload is too heavy.
In this paper, to reduce complexity, we will suggest some
method about the resource allocation base on the user’s
channel impulse response. The rest of this paper is organized
as follows, Section II describes Multi-Partition resource
allocation method, and in Section III, we describe our
simulation parameters and simulation result. In Section IV,
we discuss performance analysis and finally in Section V we
end up with conclusion.
II. MULTI-PARTITON RESOURCE ALLOCATION
METHOD
A. Conventional algorithm
In present systems, subcarriers which is assigned to each
user are nonadaptively fixed. One way of allocating the
available subcarrier is to divide them into as many partitions
as there are users and then allocate one partition to each user

in a pre-determined manner. However, the above method
does not take in account the fact that the channel magnitude
response of each user varies across subcarriers. Thus, certain
users are allocated partition that suffer from poor channel
gains resulting large path loss and random fading. It can be
seen that at each subcarrier, some user suffer from deep fade
while others are able to achieve a high channel gain.
It has two phase – initialization and iteration. In
initialization phase, the channel magnitude response for each
user is divided into a number of partition. After initial usage
value of each partition is multiplied by a ranking factor in
order to increase the probability of selection of the partition
with the highest average channel gain. Finally make the
matrix of channel information K user by K partition.
In iteration phase is carried out until each partition is only
allocated to one user. During each iteration the set of usage
value for each user is modified according to the selected
partitions of the other users at the preceding iteration. In
processing, many users select same partition, it can cause
conflict. In order to solve the conflict between users for
partition allocation, in [1] a novel method is proposed for
modification of usage value of each partition. The modified
usage value of each partition is based on the current usage
value and the inhibition perceived by the partition under
consideration. Consider the variables Un(t-1) as the usage
value of Partition N during the previous iteration, CN as the
cost of using Partition N, K as the total number of users in
the system and w as the weight factor, then the usage of
partition N during the current iteration is defined as follows
U
N ( t)
= U
N ( t)
U N ( t)
× ( 1−
w)
× w +
CN
+ 1
+ 1
( K −1)
In [2], making more partition is diving each users into a
number of partition. This leads to one user to select more
than 1 partition. In more selective channel environment, this
algorithm is more efficient. But dividing into more partition
requires larger matrix. If only 1 user has one partition we
need K by K matrix. When divided into N partition makes K
by N matrix and to apply this method, copying the user’s
information in row makes N by N matrix. So this method
needs complex calculation, more computation and when we
copy the row of the same user we can get the same
information and participation is competitive. This gives same
result for every iteration which are able to allocate many
partition to each user, base on [1] and [2].
When each user has more than 2 partition, N by N matrix is
computed to solve the conflicts among users. While doing so
the computation overload is too heavy. Using the
conventional algorithm, it leads to more competition and mis-
(1)
allocation of the band occurs. It leads to performance
degradation. To reduce this complexity, in this paper we
propose 1 simple method which are able to allocate many
partition to each users base on [1].
figure 1. Flow chart for proposed multi-partition resource allocation
B. Proposed algorithm
1) Partial Allocation Method
Our main goal is to allocate users successfully. First we make
K by N matrix based on the channel magnitude response.
Next each user selects the best K channel and participates in
the competition and makes another matrix. This matrix is
only K by K. This matrix only has information, another K by
(N – K) part is empty. Using the existing algorithm, it chose
only one of the best channels for each user. By less iteration,
we can allocate successfully. Next time the users choose the
partition and make indexing (already allocated users). Again
each user selects K (all user selects at least one partition)
except indexing partition and chooses the best partition for
each user.
This process’s advantage is one time choice to participate in
competition, and unselected partition don’t participate in the
competition, we can avoid unnecessary competition.

Computation is lower using this method. And the probability
of successful allocation to user is higher.
figure 2. Flow chart for proposed worst_band first_allocation
2) Worst Band-First allocation Method
Another method is the selection of the partition to avoid
the worst channel magnitude. Likely to partial allocation
using the channel magnitude, make K by N matrix and
summarize the whole band magnitude, the low magnitude
user is near worst channel environment. We assume that this
user’s channel is worst. So we first consider this band. In
same band, the users that have the highest channel response
are allocated this band. Next users having the lowest
magnitude are considered and find the worst channel band
and find the best user to use this band.
This method is not sure that each user has the best channel.
But it is certain that each user doesn’t choose the worst
channel band, over the whole users has fair allocation. it is
simple due to not using the matrix and no iteration. This only
use the sorting method of partition. The computation load is
very low
III. SIMULATION PARAMETER AND RESULT
In this paper we use follow system parameter and the ITU-R
M.1225 channel profile parameters.
ITU-R M.1225
Vehicular A
Table 1.
System parameter
Parameter Value
System BW 6 MHz
Modulation QPSK
Channel
ITU-R M.1225
Vehicular A,
FFT Size 2K
Number
1728
subcarrier
Number of
guard carrier
320
Table 2.
Channel profile
Path1 Path2 Path3 Path 4 Path 5 Path 6
Excess delay 0 310 ns 710 ns
Relative
amplitude
0
-1.0
dB
-9.0
dB
1090
ns
-10.0
dB
figure 3. Channel frequency response
1730
ns
-15.0
dB
2510
ns
-20.0
dB
Fig 4. shows the channel magnitude response for four users
in M.1225 Vehicular A channel. We assume that each user
has different position so they have different time delay and
different power.
In this system simulation, an FFT with 2048 subcarrier are
allocated to 4, 8, 24, 48 uses, respectively. And the proposed
methods are simulated in frequency selective fading with
AWGN..
Finally, it should be noted that no coding scheme is
employed in order to test the effectiveness of the proposed
algorithm.
Simulation is varies with user number and partition number

BER
1.E+00
1.E-01
1.E-02
1.E-03
1.E-04
1.E-05
1.E-06
figure 4. An example of channel magnitude response and partition for four users , each method.
BER vs SNR for Varying Number of Partitions
A
B
C
D
24 users, 24 partition convetional Method
24 users, 96 partition convetional Method
24 users, 24 partition Worst_First Method
24 users, 96 partition Worst_First Method
1 3 5 7 9 11 13 15
SNR(dB)
17 19 21 23 25 27 29
figure 5. Worst_first method performance
Fig.5 is the Worst_First Method. A case, each user has 1
partition. The performance is up to the expectation in the case
of existing algorithm. But B is same number of user with
larger number of partition. In this case, performance is
degraded due to the complexity and there is allocation of the
band. C is same condition performance is not as expected.
But D case, when the partition is larger (96partition)
proposed algorithm has good performance due to simplicity
and diversity effect as compare to conventional algorithm
(second line)
Fig.6 is the partial Method’s result. A is the existing
algorithm case. It has 48 partition it allocate whole spectrum.
So this matrix is very complex. In this case, many bands are
mis-allocated and performance degradation takes place. In
B’s performance is little better due to frequency diversity.
C,D are our proposed algorithm. It reduces the complexity
and performance is stable.
Fig.7 is performance of fixed number of users with varying
number of partition. As the number of partition increased, the
result shows that there is high increase in the performance.
But more than certain number of partition is saturated. It
shows the frequency diversity effect.

Figure 8 case is fixed 4 users, the number of partition is
increased. The number of partition is larger, we can get user
and frequency diversity effect by successful allocation using
the partial method.
BER
BER
1.E+00
1.E-01
1.E-02
1.E-03
1.E-04
1.E-05
1.E+00
1.E-01
1.E-02
1.E-03
1.E-04
1.E-05
1.E-06
BER vs SNR for Varying Num ber of Partitions
A
B
C
D
24 users, 48 partition convetional Method
24 users, 96 partition convetional Method
24 users, 48 partition Partial Method
24 users, 96 partition Partial Method
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29
SNR(dB)
figure 6. Performance of partial allocation method
BER vs SNR for Varying Num ber of Partitions
4 users, 2 partition per user
4 users, 6 partition per user
4 users, 12 partition per user
4 users, 24 partition per user
4 users, 48 partition per user
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
SNR(dB)
figure 7. Performance according to varying number of partition using partial
method
IV. CONCLUSION
If we use the frequency and the user diversity, we can get
better performance. But this consideration is complex for
existing algorithm. Our proposed algorithm is suitable under
this situation reducing the complexity. We verify this method
using computer simulation. And high performance is
achieved.
BER
1.E+00
1.E-01
1.E-02
1.E-03
1.E-04
1.E-05
1.E-06
BER vs SNR for Varying Number of Partitions
1 3 5 7 9 11 13 15 17 19 21 23 25 27
SNR(dB)
4 users, 6 partition per user
12 users, 6 partition per user
24 users, 6 partition per user
figure 8. Performance according to varying number of users with fixed
partition’s per user
ACKNOWLEDGEMENT
This work was supported by Korea Science & Engineering
Foundation through the NRL Program
(M1060000019406J000019410)
V. REFERENCE
[1] Teo Choon Alen, A.S. Madhukumar , Francols Chin,
“Capacity enhancement of a multi-user OFDM system
using dynamic frequency allocation”, IEEE Trans.
Braodcast. , vol.49, 2003, pp. 344-353.
[2] Yuqin Chen, SungHwan Sohn, Sang-jo Yoo, Jae Moung
Kim, “Dynamic Frequency Selection in OFDMA”,
ICACT. 2006
[3] Richard van Nee, Ramjee Prasad, “OFDM fore wireless
multimedia communications”, Artech House Publishers,
Boston London, 2000
[4] Mustafa Ergen, Sinem Coleri, and Pravin Varalya, “QoS
aware adaptive resource allocation techniques for fair
scheduling in OFDMA based broadband wireless access
systems”, IEEE Trans. Broadcast., vol. 49, 2003 pp.362-
370
[5] C.Y. Wong, C.Y.Tsui, R.S.Cheng. and K.B. Letaief, “A
real-time subcarrier allocation scheme for multiple
access downlink OFDM transmission, “In Proc. VTC
1999, vol.2, 1999, pp1124-1128.