Fast Bit and Power Allocation Algorithm

for OFDM systems

Sang-Min Lee, Yeun-Soo Park and Dong-Jo Park
Department of Electrical Engineering and Computer Science
Korea Advance Institute of Science and Technology
373-1 Guseong-dong, Yuseong-gu, Daejon 305-701, Republic of Korea
Tel: +82-42-869-5438, Fax: +82-42-869-8038
E-mail: lesam@kaist.ac.kr, freeysp@kaist.ac.kr, djpark@ee.kaist.ac.kr

Abstract We present a new bit and power allocation algo-
rithm for OFDM systems with low computational complexity.
In order to reduce the computational complexity, we divide
all subcarriers into groups, and the same number of bits is
allocated to the subcarriers in the same group. To minimize
the performance loss due to grouping, subcarriers are grouped
according to the channel gains. Based on this scheme, the bit and
power allocation problem is redefined and the algorithm for the
optimal solution is proposed. Simulation results verify that the
proposed algorithm shows comparable throughput performance
in comparison to an ideal allocation algorithm while computa-
tional load is significantly reduced.
I. I NTRODUCTION
Orthogonal frequency division multiplexing (OFDM) is a
kind of a multicarrier transmission technique, which has been
regarded as a solution to provide broadband, high-speed data
transmission over wireless channels. In OFDM system, the
channel is divided into K narrowband subchannels that can
be considered as parallel independent additive white Gaussian
noise (AWGN) channels provided K is sufficiently large. So
a high-rate data stream can be obtained by means of multiple
parallel low-rate data streams, which are then simultaneously
transmitted on a number of orthogonal subcarriers. Owing to
this merit, OFDM systems have aroused great interest in recent
years as potential solution to the problem of transmitting data
over wireless channels with large delay spread.
Assuming that the transmitter knows the instantaneous chan-
nel state information, adaptive modulation can improve the
performance significantly. In particular, subcarriers with large
channel gains employ higher order modulation to carry more
bits/symbol, while subcarriers in deep fade carry one or even
zero bits/symbol. As different subcarriers experience different
fades and transmit different numbers of bits, the transmit
power levels must be changed accordingly. Researchers have
developed many bit and power allocation algorithms [1]-[8].
Hughes-Hartogs algorithm [1] can be applied to minimize
the total transmission power with the constraint of the fixed
channel capacity or to maximize the channel capacity with the
total transmission power constraint. Chows algorithm [2] is a
computationally efficient bit and power allocation algorithm
for maximizing the channel capacity. Krongold [3] used a
Lagrange multiplier bisection search to find the optimal bit
and power allocation effectively constrained by the constant
transmission power and a required bit error rate (BER).
However, because the conventional algorithms search the
number of bits and the power level allocated to every single
subcarrier, very large computational load is required. In this
paper, in order to reduce the computational load, all subcarriers
are divided into subcarrier groups, and the same number of bits
are allocated to the subcarriers in the same group. Under this
assumption, a bit and power allocation problem is reformulated
and the algorithms for the optimal solution of the reformulated
problem is proposed. Also we propose the scheme to divide
subcarriers in order to minimize the performance loss due to
subcarrier grouping.
The organization of this paper is as follows. In Section II,
we first give the system model and formulate a bit and power
allocation problem. The optimal bit and power allocation
algorithm is described in Section III. In Section IV, we propose
a new bit and power allocation algorithm based on subcarrier
grouping and subchannel sorting. In Section V, our results are
verified through Monte Carlo simulations. Finally, conclusion
is followed in Section VI.
II. S YSTEM M ODEL
We consider an adaptive multi-carrier transmission sys-
tem in which the transmitter uses combined bit and power
allocation algorithms based on the channel information. A
modulation mode for each subcarrier is selected corresponding
to the number of bits allocated to the subcarrier and the symbol
modulated by the selected mode is then scaled to the allocated
power. We define c(k) and p(k) as the number of allocated
bits and the transmission power level of the kth subcarrier,
respectively.
We denote f (c) as the required received power per a symbol
in the kth subcarrier to satisfy a given BER requirement in a
c bits/symbol modulation scheme. Assuming that the channel
state information is known, the transmission power of the kth
subcarrier can be given by

The work in this paper was supported by Korea Research Foundation(KRF). p(k) = f (c(k))/|H(k)|2 , (1)

0-7803-8521-7/04/$20.00 (C) 2004 IEEE 503

0-7803-8521-7/04/$20.00 (C) 2004 IEEE

 Fig. 1. OFDM system with bit and power allocation

where H(k) is the channel gain of the kth subcarrier. Thus Repeat the following RT times:
the combined bit and power allocation algorithm should find k = arg mink p(k)
the optimal assignment of c(k) so that the total transmission c(k) = c(k) + 1
power is minimized satisfying the transmission rate and BER p(k) = (f (c(k) + 1) f (c(k)))/|H(k)|2
requirements. The optimization problem can be mathemati-
cally formulated as following: Finish:
Bit allocation: c(k), k


K Power allocation: p(k) = f (c(k))/|H(k)|2 , k
min PT = f (c(k))/|H(k)|2
c(k)D
k=1
(2)


K
subject to c(k) = RT , IV. P ROPOSED A LLOCATION A LGORITHM
k=1

where D is the set of all possible values for c(k), and c(k) =
0 means that no information is transmitted through the kth
subcarrier.
III. O PTIMAL A LLOCATION A LGORITHM
As the power needed to transmit a certain number of bits in
a subcarrier is independent of the numbers of bits allocated to
other subcarriers, it turns out that a greedy approach is optimal.
A greedy algorithm assigns bits to the subcarriers one bit at a
time, and in each assignment, the subcarrier that requires the
least additional power is selected. The bit allocation process
will be completed when all bits are assigned. The basic
structure of the greedy algorithm can be described as follows:
Initialization:
c(k) = 0, k
p(k) = (f (1) f (0))/|H(k)|2 , k
The conventional greedy algorithm converges to the optimal
solution, but requires very high computational complexity.
To reduce this complexity, we divide all subcarriers into N
groups and allocate bits to each group, not to each subcarrier.
The same number of bits is allocated to all the subcarriers
in the same group although they have different channel gain.
However, the difference of the channel gains in a group may
cause the performance loss.
To minimize this performance loss, all subcarriers are
sorted by their channel gains in ascending order. The sorted
subcarriers index set is denoted as {a1 , a2 , , aK }
with elements satisfying |H(a1 )| |H(a2 )|
|H(aK )|. Then all subcarriers are grouped into N groups
according to their indices. The index set of the nth group
is {a(n1)M +1 , a(n1)M +2 , , anM }, where M = K/N ,
and M is assumed to be an integer.
The group-based allocation problem is formulated from (2)

0-7803-8521-7/04/$20.00 (C) 2004 IEEE 504

0-7803-8521-7/04/$20.00 (C) 2004 IEEE

with one additional condition as following V. S IMULATION R ESULTS
d(n) c(a(n1)M +1 ) = c(a(n1)M +2 ) = = c(anM ), In order to show that the proposed algorithm has low com-
(3) putational load and comparable performance to the conven-
which means that the same number of bits is allocated to tional subcarrier-based algorithm, simulations were performed
all the subcarriers in the same group. Then the total transmit in the conditions of an adaptive, uncoded OFDM system with
power has the following form 64 subcarriers using HyperLAN/2 channel model B. The adap-
tive modulator of the OFDM system allows ck to take values in


K
f (c(k))
N
the set D = {0, 1, 2, 4, 6, 8, }, that is, BPSK and square
= G(n)f (d(n)), (4)
|H(k)|2 QAM. Target BER is set to be 103 . The simulations are
k=1 n=1
carried out with M = 1, 4, 16, 32. Also the simulation results
where G(n) is defined as of the proposed algorithm (case 1) are compared with those


nM
1 of the simple group-based algorithm (case 2) which groups
G(n) := . (5) subcarriers without sorting and then their indices set of the
|H(ak )|2
k=(n1)M +1 nth group is {(n 1)M + 1, (n 1)M + 2, , nM }.
The overall transmission rate becomes Figure 2 shows the BER performance according to through-
put. There is no BER performance loss due to grouping


K

N
c(k) = M d(n). (6) subcarriers because the transmit power per each subcarrier is
k=1 n=1
determined to satisfy the target BER respectively.
Figure 3 shows the computational loads of the proposed
As a consequence, under the condition (3), (2) can be
algorithms. The algorithms were coded in MATLAB and the
reformulated as
platform was an Intel Pentium-4 2.4 GHz personal computer.


N
The CPU time for the proposed allocation algorithm with
min G(n)f (d(n))
d(n)D M=16 is less than almost a hundredth of that of the conven-
n=1
(7) tional subcarrier-based algorithm.


N
RT Figure 4 and Figure 5 depict the transmit power divided
subject to d(n) = .
n=1
M by noise power. In order that the bit and power allocation
with M=16 obtains the same throughput as M=1, the transmit
Conventional allocation algorithms proposed in [1], [3] and
power must be increased more than 2 dB if subcarriers are
[4] can be easily modified to find the optimal solution for
not sorted (case 2), but less than 0.5 dB if subcarriers are
(7). The modified algorithms for (7) have less computational
sorted (case 1) respectively. These results show that sorting
complexity than the original algorithms for (2) because K/M
subcarriers considerably reduces the performance loss due to
decision variables exist in (7) while K variables exist in (2).
grouping.
In this paper, we employ the greedy approach. The
computational complexity of the proposed group-based VI. C ONCLUSION
allocation algorithm is O(RT K/M 2 + K log K) while that This paper have described a bit and power allocation prob-
of the optimal greedy algorithm is O(RT K). The procedure lem for multi-carrier transmission schemes, in which overall
of the proposed algorithm is as following transmission power is minimized with the constraints of a
fixed data rate and bit error rate. We proposed a new bit
Initialization: and power allocation algorithm, which divide all subcarriers
Sort all the subcarriers by their channel gains in into groups according to their channel gains, and the same
ascending order. The sorted subcarrier indices are number of bits is allocated to the subcarriers in the same group.
denoted as {a1 , a2 , , aK } with elements satisfying Simulation results demonstrate that the proposed algorithm
|H(a1 )| |H(a2 )| |H(aK )|. has significant computational advantages over an optimal
d(n) = 0, n allocation algorithm, while offering comparable performance.
G(n) = k=(n1)M +1 1/|H(ak )|2 , n
nM

q(n) = (f (1) f (0))G(n), n R EFERENCES

Repeat the following RT /M times:
n = arg minn q(n)
d(n) = d(n) + 1
q(n) = (f (d(n) + 1) f (d(n)))G(n)
Finish:
Bit allocation: c(k) = d(k/M  + 1), k
Power allocation: p(k) = f (c(k))/|H(k)|2 , k
[1] D. Hughes-Hartogs, Ensemble modem structure for imperfect transmis-
sion media, U.S. Patents Nos. 4,679,227, July 1987; 4,731,816, Mar.
1988; and 4,833,796, May 1989.
[2] P. S. Chow, J. M. Cioffi and J. A. C. Bingham, A practical discrete
multitone transciever loading algorithm for data transmission over spec-
trally shaped channels, IEEE Trans. Commun., vol. 43, pp 773-775,
Feb./Mar./Apr. 1995.
[3] B. S. Krongold, K. Ramchandran, and D. L. Jones, Computationally effi-
cient optimal power allocation algorithms for multicarrier communication
systems, IEEE Trans. Commun., vol. 48, pp. 23-27, Jan. 2000.
[4] S. K. Lai, R. S. Cheng, K. B. Letaief and R. D. Murch, Adaptive trellis
coded MQAM and power optimization for OFDM transmission, Proc.
VTC 1999, vol. 1, pp. 290-294, May 1999.

0-7803-8521-7/04/$20.00 (C) 2004 IEEE 505

0-7803-8521-7/04/$20.00 (C) 2004 IEEE

[5] R. Grunheld, E. Bolinth and H. Rohling, A Blockwise loading algorithm
for the adaptive modulation technique in OFDM systems, Proc. VTC
2001 Fall, vol. 2, pp. 948-951, Oct. 2001.
[6] S. Zhou and G. B. Giannakis, Adaptive modulation for multi-antenna
transmissions with channel mean feedback, Proc. ICC 2003, vol. 4, pp.
2281-2285, May 2003.
[7] L. Zhen, W. Weihua, Ahou Wenan and S. Junde, A simple bit and power
allocation algorithm for OFDM systems, Proc. ICCT 2003, vol. 2, pp. 25
1169-1172, 2003.
[8] C. Y. Wong, R. S. Cheng, Khaled Ben Letaief and R. D. Murch,
Multiuser OFDM with adaptive subcarrier, bit, and power allocation,

Transmit power (dB)

IEEE J. Select. Areas Commun., vol. 17, no. 10, Oct. 1999.
20
1
10
M=1
M=4 15
M=16
M=32
2 Sorting, M=4 M=1
10
Sorting, M=16 10 Sorting M=4
Sorting, M=32 Sorting M=16
BER

Sorting M=32

3
100 150 200 250 300
10 Data rate (bits/OFDM frame)

Fig. 4. Data rate vs. Transmit power (Case 1)

4
10
100 150 200 250 300
Data rate (bits/OFDM frame)

Fig. 2. Data rate vs. BER

2
10
M=1
Sorting, M=4
1 Sorting, M=16
10
Sorting, M=32 25
0
CPU time

10
Transmit power (dB)

20
1
10

2 15
10

3 M=1
10 10
100 150 200 250 300 M=4
Data rate (bits/OFDM frame) M=16
M=32
Fig. 3. Data rate vs. CPU time
100 150 200 250 300
Data rate (bits/OFDM frame)

Fig. 5. Data rate vs. Transmit power (Case 2)

0-7803-8521-7/04/$20.00 (C) 2004 IEEE 506

0-7803-8521-7/04/$20.00 (C) 2004 IEEE