Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Channel state
Abstract—In the multiuser OFDM system, various adaptive information
subcarrier-bit-and-power allocation algorithms have been inves-
tigated based on the assumption that the number of supportable
users is given. This paper proposes a supportable user set Subcarrier 1
Data of user 1 Adaptive
estimation in order to adopt those allocation algorithms in the Subcarrier 2
Supportable subcarrier,
multiuser-OFDM system. The number of supportable users is Data of user 2
user bit,
estimated in an iterative way by calculating the required power OFDM
set and
…
…
…
approximately and comparing it with an available power after estimation power
Subcarrier N
determining the number of subcarriers to be allocated. Simula- allocation
tion results show that the proposed user set estimation provides
a significant complexity reduction for the adaptive subcarrier-
bit-and-power allocation in the multiuser-OFDM system. Ke
Index Terms—Adaptive subcarrier-bit-and-power allocation, Fig. 1. Block diagram of the downlink multiuser OFDM transmitter with
multi-user OFDM. an adaptive subcarrier-bit-and-power allocation.
I. I NTRODUCTION
problem was introduced in the part of algorithm design but
A N ADAPTIVE subcarrier-bit-and-power allocation for
the multiuser OFDM system has been investigated in
[1] where a Lagrangian method of optimization is used to
not sufficiently dealt with.
This paper proposes a supportable user set estimation in
minimize the total transmit power under the constraints of order to adopt an adaptive subcarrier-bit-and-power allocation
the users’ QoS requirements. Although the algorithm achieves algorithm which assumes the number of supportable users
a dramatic gain in power efficiency, the high computational is given. The amount of required power is calculated ap-
complexity renders it impractical. Hence, many other adap- proximately after determining the number of subcarriers to
tive subcarrier-bit-and-power allocation algorithms have been be allocated. Comparing it with the total available power,
proposed to reduce the complexity of allocation algorithm [2]- the supportable user set is determined in an iterative way.
[6]. Estimating the supportable user set, the adaptive subcarrier-
Focusing on the complexity issue, most researches have bit-and-power allocation can be implemented efficiently with
proposed various allocation algorithms based on the assump- a significantly reduced complexity.
tion that the number of supportable users is given [1]- [5].
Assuming that K users are to be supported, the power
minimization problem is defined and solved under the users’ II. S UPPORTABLE U SER S ET E STIMATION P ROBLEM
QoS constraints. However, the exact number of supportable
Fig. 1 shows the block diagram of a downlink multiuser
users can not be determined before the allocation is finished
OFDM transmitter employing the proposed supportable user
[5]. For example, suppose that an algorithm allocates the
set estimation. Using the channel state information from the
subcarriers, bit, and power assuming that K users could be
receiver, the transmitter applies the combined subcarrier-bit-
supportable, but it is found that the available power is not
and-power allocation algorithm in order to assign the different
enough to satisfy the K users’ QoS requirements. In this
subcarriers to the different users and the number of bits to be
case, the allocation algorithm has to be repeated after dropping
transmitted on each subcarrier. Depending on the number of
some users. Even though the algorithm is designed to solve
bits assigned to a subcarrier, a complex symbol is generated
the allocation problem efficiently, this repetition causes an
using a corresponding modulation scheme, and the transmit
unacceptably high computational complexity. Such a repetition
power is adjusted according to the allocated power level.
problem can happen in most allocation algorithms which
Applying OFDM modulation to N complex symbols, OFDM
assume the number of supportable users is given. In [5], this
symbol is generated and transmitted. Prior to applying the
Manuscript received January 22, 2007; revised August 5, 2007, January 1, allocation algorithm, the supportable user set is estimated by
2008, and February 1, 2008; accepted February 23, 2008. The associate editor the proposed iterative search method.
coordinating the review of this letter and approving it for publication was S.
Hanly. This work was supported in part by Samsung Electronics Co., Ltd., Let PT be the total available power at the transmitter and pk
Korea. be the required power for user k to transmit Rk bits per unit
The authors are with the Dept. of Electrical and Electronic En- OFDM signal. Assuming that all the users are sorted according
gineering, Yonsei University, Seoul, Korea, 120-749 (e-mail: {ace4y,
jsseo}@yonsei.ac.kr). to the priorities by the scheduler, the user set estimation
Digital Object Identifier 10.1109/TWC.2008.070090. problem is to find the maximum number of supportable users
1536-1276/08$25.00
c 2008 IEEE
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 4, 2009 at 10:21 from IEEE Xplore. Restrictions apply.
2046 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 6, JUNE 2008
Ke with total power constraint solution n∗ is not required in solving (3) if the current solution
is enough to calculate Preq for comparing with PT . Hence,
Ke = arg max K
the standard direct search method is adopted in which the
K
(1) optimum solution is obtained iteratively in the feasible set
s.t. Preq = pk < PT .
[9].
k=1
From the qth current solution n(q) , the (q + 1)th solution
Since calculating the required power by means of the adaptive n (q+1)
is updated as
subcarrier-bit-and-power allocation requires complex opera-
tions, (1) is solved in an iterative way by calculating pk n(q+1) = n(q) + α · d (4)
approximately and comparing it with PT . Let nk denote the (q) ∗
where d is a direction vector from n to n and α is a
number of subcarriers to be allocated for user k. Assuming that
distance from n(q) to n(q+1) . At the current solution n(q) , let us
nk can be relaxed to be a real number and that the channel
assume that (1 + t) constraints are active among the (1 + 2K)
gain on every subcarrier of user k is the same as the average
constraints in (3), i.e., 1T n = N and #S(n(q) ) = t, where
channel gain Hk , pk (nk ) can be calculated as (q) (q)
S(n(q) ) = {k : nk = Rk or nk = 1, k = 1, 2, · · · , K} and
nk Rk #S denotes the cardinality of the set S. Then, the direction
pk (nk ) = f (2)
Hk nk vector d = [d1 , d2 , · · · , dK ]T at n(q) is obtained by projecting
where f (dk ) = Γ(2dk − 1) is the required power for a reliable the steepest descent direction onto the plane composed of the
reception of dk bits/subcarrier per symbol when the channel active constraints as
gain is unity and Γ is the SNR gap to account the reduction d = −W∇p(n(q) ) (5)
of SNR with respect to capacity. For M-QAM modulation, Γ
is defined as [7] where W is a K × K projection matrix determined
2 by the active constraints and −∇p(n) is a K × 1
1 −1 SER
Γ= Q steepest descent direction vector given by ∇p(n) =
3 4
[ ∂n
∂
1
p1 (n1 ), ∂n
∂
2
p2 (n2 ), · · · , ∂n∂K pK (nK )]T . Denoting I and
where Q function is defined as ik an identity matrix of size K and a kth row of I, respectively,
∞ the projection matrix W in (5) is expressed as
1 2
Q(x) = √ e−t /2 dt
2π x W = I − CT [CCT ]−1 C (6)
and SER is the target symbol error rate. For a given K,
Preq =
K where C is (1 + t) × K matrix representing the (1 + t)
k=1 pk (nk ) is compared with PT and it is de-
cided that K users are supportable if Preq is smaller than constraints at n(q) , of which the first row is 1T and the other
PT . Since pk (nk ) is calculated approximately under the flat rows are ik (k ∈ S(n(q) )). Applying the vector multiplication
fading channel assumption without the subcarrier assignment and addition, (6) can be simplified as
and power allocation, there exists a difference between the 1
W = diag(w) − wT w (7)
estimated number of supportable users Ke and the maximum K −t
number of supportable users K ∗ which can be obtained by
where w = 1T − k∈S(n(q) ) ik .
the adaptive subcarrier-bit-and-power allocation algorithm.
Having determined the direction d at n(q) , the step size
III. S UPPORTABLE U SER S ET E STIMATION α is calculated using the fact that the slope of the objective
function at n∗ is zero. In order to determine α that ensures
For a given K in the supportable user set estima- n(q+1) will lie in the feasible set, let αk = {(Rk − 1)/|dk | +
tion, the number of subcarriers to be allocated to user (Rk − 2nk + 1)/dk }/2, (k ∈ / S(n(q) )), which represents a
k is determined in order to calculate the total re- distance from n(q) to the kth rate constraints in (3). Then,
quired power. Let n = [n1 , n2 , · · · , nK ]T and p(n) = the maximum step size αU from n(q) is determined as αU =
[p1 (n1 ), p2 (n2 ), · · · , pK (nK )]T be K × 1 vectors which rep- min αk . Given n(q) and αU , 1T p(n) in (3) is a function of α.
resent the number of allocated subcarriers and required power Defining g(α) = 1T p(n(q) + αd) to be the objective function
to transmit R = [R1 , R2 , · · · , RK ]T bits, respectively. Since at n(q) + αd, the slope of g(α) is given by
the required power function f (dk ) in (2) is convex, it can
be easily shown that the derivative of p(nk ) is nondecreasing d T
g (α) = {1 p(n(q) + αd)} = ∇p(n(q) + αd)T d (8)
(p (nk ) ≥ 0) for nk > 0 and therefore p(nk ) is convex for dα
nk ≥ 0 [8]. Hence, Preq = 1T p(n) in (1) is convex for n ≥ 0 and the step size is determined as α = αU if g (αU ) < 0. If
and n is obtained by solving a convex optimization problem g (αU ) > 0, a simple bisection method is applied to find α as
with linear constraints shown in Fig. 2 since the minimum of 1T p is in the interval
min 1T p(n) [0, αU ].
Given the average channel gain Hk , the required data rate
s.t. 1T n = N (3)
R, and total available power PT , the number of supportable
1≤n≤R user Ke is obtained as follows.
where 1 is a column vector of size K with all entries equal Step 1) Set K = 2.
to one. Since (3) is formulated to calculate Preq , the optimal Step 2) Set n(0) = 1N T R R.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 4, 2009 at 10:21 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 6, JUNE 2008 2047
g (a) = 1T p(n ( q ) + α d)
feasible region
g '(αU ) < 0
p '(n∗ ) ≈ 0 g '(0) > 0
n ( q ) + αU d n∗ n( q )
TABLE I
N UMBER OF OPERATIONS REQUIRED TO SOLVE (3)
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 4, 2009 at 10:21 from IEEE Xplore. Restrictions apply.
2048 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 6, JUNE 2008
100. With the proposed method, not only the number of and the number of supportable users is increased if Preq < PT
total computations to obtain Ke but also the number of and decreased if Preq > PT until K ∗ is determined. In Fig. 5,
total computations to find the number of supportable users the number of total computations required to obtain K ∗ from
by repeating the resource allocation algorithm is counted. Ke is compared to the number of total computations required
Without the proposed method employed, the number of total to obtain K ∗ from Kmax , in which the ratio is ∗calculated as
the number of total computations to obtain Ke and K (f rom Ke )
computations required to obtain K ∗ from Kini is counted. In the number of total computations to obtain K ∗ f rom Kmax ×
obtaining K ∗ , the required power Preq is calculated using the 100. It is shown that the number of computations is reduced by
adaptive subcarrier-bit-and-power allocation algorithm and the 50% as compared to the approach in [5]. Since the maximum
number of supportable users is decreased if Preq < PT until number of supportable users (K ∗ ) obtained from Ke is the
K ∗ is found as in [5]. Since the number of total computations same as that obtained from Kmax , the average number of
without the proposed method is greatly affected by the initial supportable users and transmission data rate will be the same
state Kini , Kini is set to Kmax = max K ∗ , the largest as those without employing the proposed method in Fig. 4.
value among the maximum number of supportable users (K ∗ s) Hence, by using Ke as Kini , the proposed user set estimation
which are generated while evaluating the average number of could be combined with the resource allocation algorithm
supportable users in Fig. 4. In Fig. 5, it is shown that the without the performance degradation while still providing the
ratio of the total computation is less than 20% regardless of complexity reduction effect.
N , which means that only 20% of the total computations is In [4], BABS has been proposed as a part of the
needed when the proposed method is employed in the adaptive subcarrier assignment algorithm but it can be used in the
subcarrier-bit-and-power allocation. Also, we can find that proposed method in which n is determined by BABS
the proposed method provides the complexity reduction effect first and Preq is calculated from n in step 3). In Fig.
even though the resource allocation process is repeated when 5, the required number of total computation to solve (3)
PT is not enough to support Ke users. This is because that by the proposed method and BABS is also compared
the repetition happens less than 4% of time and Ke is not in which the ratio of total computations is calculated as
the number of total computations to solve (3) by the proposed method
so far from K ∗ when Ke is greater than K ∗ as shown in the number of total computations to solve (3) by BABS ×
Fig. 3. Note that if the more effective (but more complex) 100. Compared to BABS, the proposed method provides
allocation algorithm such as MAO in [1], enhanced subcarrier more than 50% of the reduction which comes from the fact
swapping in [6], or RCG in [4] is adopted as the adaptive that the proposed method tries to find n in the feasible set,
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 4, 2009 at 10:21 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 6, JUNE 2008 2049
using the standard direct search method in the feasible set, the
computational complexity to obtain the number of supportable
users is reduced. Numerical results show that the supportable
user set estimation reduces the allocation complexity signif-
icantly when it is combined with the adaptive subcarrier-bit-
and-power allocation algorithms.
ACKNOWLEDGEMENTS
The authors would like to thank the Associate Editor and
the reviewers for their helpful comments and review. Their
suggestions greatly improved the presentation of this paper.
R EFERENCES
[1] C. Y. Wong, R. S. Cheng, K. B. Letaief, and R. D. Murch, “Multiuser
OFDM with adaptive subcarrier, bit, and power allocation,” IEEE J.
Select. Areas Commun., vol. 17, pp. 1747–1758, Oct. 1999.
Fig. 5. Ratio of total computations with and without the proposed method [2] C. Y. Wong, C. Y. Tusi, R. S. Cheng, and K. B. Letaief, “A real-
employed. time subcarrier allocation scheme for multiple access downlink OFDM
transmission,” in Proc. IEEE VTC’99, vol. 2, pp. 1124–1128, Fall 1999.
[3] T. C. H. Alen, A. S. Madhukumar, and F. Chin, “Capacity enhancement
of a multi-user OFDM system using dynamic frequency allocation,”
hence min( KN
, K) iterations are not always needed to solve IEEE Trans. Broadcasting, vol. 49, pp. 344–353, Dec. 2003.
[4] D. Kivanc, G. Li, and H. Liu, “Computationally efficient bandwidth
(3) while BABS iterates N times with K function evaluations allocation and power control for OFDMA,” IEEE Trans. Wireless
and K comparisons in each iteration to obtain a feasible Commun., vol. 2, pp. 1150–1158, Nov. 2003.
solution n. [5] H. Yin and H. Liu, “An efficient multiuser loading algorithm for OFDM-
based broadband wireless systems,” in Proc. IEEE GLOBECOM, vol.
1, pp. 103–107, Dec. 2000.
V. C ONCLUSIONS
[6] Y. Chen, J. Chen, and C. Li, “A fast suboptimal subcarrier, bit, and
This paper proposed the supportable user set estimation in power allocation algorithm for multiuser OFDM-based systems,” in
order to adopt the adaptive subcarrier-bit-and-power allocation Proc. ICC’04, vol. 6, pp. 3212–3216, June 2004.
[7] Ana Garcia-Armada, “SNR gap approximation for M-PSK-based bit
algorithms which assume that the number of supportable users loading,” IEEE Trans. Wireless Commun., vol. 8, pp. 57–60, Jan. 2006.
is given. In order to estimate the number of supportable users, [8] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge
the amount of required power is calculated and compared University Press, 2004.
[9] A. D. Belegundu and T. R. Chandrupatla, Optimization Concepts and
with the available power iteratively after determining the Applications in Engineering. Prentice Hall, 1999.
number of subcarriers allocated to each user by solving a [10] W. Cheney and D. Kincaid, Numerical Mathematics and Computing.
linear constraint optimization problem. By solving the problem Brooks/Cole Publishing Company, 1999.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 4, 2009 at 10:21 from IEEE Xplore. Restrictions apply.