[ieee 2010 ieee globecom workshops - miami, fl, usa (2010.12.6-2010.12.10)] 2010 ieee globecom...
TRANSCRIPT
![Page 1: [IEEE 2010 Ieee Globecom Workshops - Miami, FL, USA (2010.12.6-2010.12.10)] 2010 IEEE Globecom Workshops - Novel power and time allocation algorithm for a Dynamic TDMA slot assignment](https://reader031.vdocuments.site/reader031/viewer/2022020615/575095071a28abbf6bbe3f66/html5/thumbnails/1.jpg)
Novel Power and Time Allocation Algorithm for a Dynamic
TDMA Slot Assignment Multiuser Transmission SchemesYosra Abbes, Fatma Abdelkefi and Hichem Besbes
Research Unit TECHTRA Sup’Com
University of the 7th November at Carthage, Tunisia
Emails: [email protected], [email protected] and [email protected]
Abstract—In the literature, many design parameters have been utilizedin network optimization but till now little attention has been paid onthe buffer management in the base station. In this paper, we propose
a new power allocation algorithm suited for Dynamic Time DivisionMultiple Access (TDMA) scheme, that is based on the minimization ofthe total transmitted power under the constraint of ensuring the bufferstability since the buffer overflow is one of the important causes of trafficoutage and so the throughput degradation. We show that this optimization
problem can be efficiently solved througth using the convex optimizationtechniques. The global optimal minimization solution can be computednumerically since the objective and the constraint functions are non
linear. We can ensure the power control by adding an upper bound onthe transmitted power. The new formulated problem can be used foradmission control. Since the optimization problem is no longer convex,
we propose a simple heuristic algorithm to solve it. Simulation results aregiven to support our claims and to illustrate the good performance at the
level of power allocation obtained by the proposed allocation schemes.
Index Terms—Power allocation scheme, time allocation, buffer stability,
power control and admission control.
I. INTRODUCTION
High data rates, high spectral efficiency, flexibility and low power
consumption will be important features in the next generation of
wireless communications systems. An efficiently resource allocation
scheme is needed to distribute the available radio resources among
multiple users in the presence of limited resources (buffer’s length,
bandwidth and power).
The problem of resource management was investigated in many
previous works [1], [2], [3], [4]. In [1], a multiuser OFDM sub-
carrier, bit and power allocation algorithm was proposed in order to
minimize the total transmitted power using Lagrangian relaxation.
Despite its significant gains over the fixed allocation scheme, the
algorithm is computational intensive and is difficult to implement.
Linear programming was used in [2] to solve the subcarrier allocation
by linearizing the function of rate in terms of power. However, the
linearization can not generally be applied to all types of modulations
and it still needs further iterations to solve the linear programming
problem. In [3], an iterative water-filling algorithm was suggested
to solve a weighted sum power minimization problem. Authors in
[4] proposed a set of problem formulations that allocate network
resources to optimize the Signal to Interference Ratio (SIR), to
maximize throughput and to minimize delays. Such formulations
belong to non linear optimization problems, but can be efficiently
solved through the usage of convex optimization methods.
When the transmission demand is increased, buffers will eventually
start to become overflowed. This results in QoS degradation in the
network system. In [5], authors focus on the buffer management in the
base station. The resource allocation scheme aimed to maximize the
total packet throughput subject to individual user’s outage probability.
This strategy permits to allocate traffic channels according to uses
channel conditions as well as traffic patterns. In [6], authors suggest
This work was supported in part by the Qatar National Research Fundunder the project NPRP 08 - 577 - 2 - 241.
resource allocation algorithm based on queue-balancing and ensuring
stability of the system and establishing threshold rate of arrival
traffic.Notably, most of the existing works, in the field of power
allocation, doesn’t consider the problem of buffer stability as a
constraint of the power allocation scheme. This paper investigates
power allocation scheme under the constraint of maintaining buffer
stability in a dynamic Time Division Multiple Access (TDMA) mode.
We propose an efficient power allocation strategy that minimizes
the total power consumption and assigns to the different users the
required number of time-slots in a dynamic way.
We start by formulating the minimization problem, then showing
that this problem is convex and we propose to solve it under
the Karush-Khun-Tucker (KKT) conditions. The optimal solution
is computed numerically using an appropriate initialization and an
algorithm deduced from Fixed Point Method in the simulations part
of this paper.
Due to the limitation of power resource, minimum power require-
ments for all users may not be satisfied simultaneously. This motivates
the investigation of admission control where users are not automati-
cally admitted into the network. The constraint of power control will
be taken in account by reformulating the initial optimization problem
through the addition of an upper bound on the power allowed to
each user. However, the minimization problem is no longer convex.
In order to solve it, we suggest an heuristic suboptimal algorithm that
utilizes the outcome of the initial problem.
The remainder of this paper is organized as follows. In Section
II, we describe the system model and we formulate the optimization
problem. In Section III, we prove the uniqueness and existence of
the global minimum solution and we formulate the KKT conditions
used in the derived results. In Section IV, we represent the power
allocation algorithm with admission control. In Section V, we states in
a detailed way the considered simulation parameters and we represent
the obtained results. Finally, conclusions are given in Section VI.
II. SYSTEM MODEL AND OPTIMIZATION PROBLEM
A. Dynamic TDMA system
The TDMA scheme is a technique for resource management based
on the basic idea of time-sharing. In this paper, we denote the total
signal occupancy period as T = N Ts, where Ts is the time-slot
duration and N is the total number of time-slots. The time of channel
occupancy of each user is equal to:
Tk = tk T,
where tk denotes the percentage of spectral occupancy of the user
“k” such that 0 ≤ tk ≤ 1 and
K∑
k=1
tk ≤ 1.
We consider a single cell downlink Dynamic TDMA system with
K users. The Channel State Information (CSI) is assumed accurately
known at the transmitter and the K receivers.
IEEE Globecom 2010 Workshop on Broadband Wireless Access
978-1-4244-8865-0/10/$26.00 ©2010 IEEE 824
![Page 2: [IEEE 2010 Ieee Globecom Workshops - Miami, FL, USA (2010.12.6-2010.12.10)] 2010 IEEE Globecom Workshops - Novel power and time allocation algorithm for a Dynamic TDMA slot assignment](https://reader031.vdocuments.site/reader031/viewer/2022020615/575095071a28abbf6bbe3f66/html5/thumbnails/2.jpg)
In this paper, we assume that the channel is “slow flat fading
channels”. In fact, it is not straightforward to consider a more realistic
channel model and we point out that the frame length is smaller than
the channel coherence time. The received signal at the user “k” side
is given by:
yk = hk.s+ nk
where s is the transmitted signal, hk is the channel gain, assumed
constant during the period T and nk is an Additive White Gaussian
Noise (AWGN) having a variance equal to σ2k.
We consider that the channel gain to noise ratio of a user “k” has
the following expression:
γk =|hk|
2
σ2k
. (1)
Remind that the channel capacity for the kth user is given by
ck = B log2(1 + Pkγk)
where Pk denotes the amount of power allocated to the user “k”.
As we consider in our work a dynamic TDMA systems, it results
that the rate offered to the user “k” is equal to:
rk = tkck. (2)
We assume that the stability condition is verified by each user [7],
this is equivalent to say that when it is supposed a finite time horizon
n, the average offered traffic tends to the arrival data rate λk. This
means that rk = λk. Also, we must emphasize that λk is known
at the receiver side. Therefore: λk = tkck and as
K∑
k=1
tk ≤ 1, we
deduce the following stability condition[6]:
K∑
k=1
λk
ck≤ 1. (3)
However considering the stability approach in such a system, we
believe that it is a new approach which implies one of the main
contributions of this paper.
B. Problem formulation
Our main purpose consists in the minimization of the total
transmitted power when the buffer is maintained stable. Therefore,
we formulate our optimization problem by applying the stability
condition already presented in (3). Since our goal is to minimize
the total transmitted power, the transmitted rate rk must be closed
to the arrival data rate λk in order to reduce the waste of power. To
define the optimum power allocation for each user “k”, we propose
in (4) to minimize the total energy consumed during the period T ,
given by ET =∑K
k=1 PkTk =
K∑
k=1
λkPk
ckT . Since T is constant,
our optimization problem is formulated by:
minPk
K∑
k=1
λkPk
B log2 (1 + Pk γk)
subject to
K∑
k=1
λk
B log2 (1 + Pk γk)≤ 1 (C1)
(4)
where we assume that Pk ≥ 0, since Pk < 0 is not a feasible
solution.
Although the problem (4) is a non- linear optimization problem,
it will be shown in the next section that it has a unique and global
solution.
III. UNIQUENESS OF THE SOLUTION
Lemma 1: The non-linear optimization problem introduced in (4)
has a unique optimal solution.
Proof: The proof will be presented in three steps.
Step1: Introduction of variables change
In order to simplify the resolution of the minimization problem (4),
we introduce new variables xk and Ak given by xk = ln (1 + Pkγk)and Ak = λk ln(2)
B. The objective function is given by
f(X) =
K∑
k=1
Ak (exp (xk)− 1)
γkxk
(5)
subject to
g(X) =
K∑
k=1
Akγk
xk
− 1 (6)
where X = [x1, . . . , xK ]T .
Hence, the optimization problem (4) can be equivalently rewritten
in a standard form as:
Pbeq :
minX∈R
n
+
f (X)
subject to g (X) ≤ 0(7)
It is interesting to note that f which was given in (5) and g that
was given in (6) are convex and continuous functions, so Pbeqis a
convex programming problem.
Step 2: existence and uniqueness of the solution
Let L = {X ∈ Rn+; g (X) ≤ 0} a set of the feasible solution of
Pbeq . Since f and g are convex, L is a convex and closed set of
Rn+. Furthermore, lim
‖X‖→+∞f (X) = +∞ ∀X ∈ L and X > 0 .
We deduce that it exists a unique and global minimum, solution of
Pbeq [8].
Using the fact that there exists x̄ solution of Pbeq and noting that f
et g are differentiable at x̄, we can deduce that Karush-Khun-Tucker
(KKT) conditions are satisfied [8]. So, there exist a scalar α ≥ 0 such
the following conditions hold while considering a single constraint:
∇f (x̄) + α∇g (x̄) = 0
(α∇g (x̄)) = 0
g (x̄) ≤ 0
α ≥ 0
Consequently, we obtain the following system:
{
exp (xk) (xk − 1) + 1− αγk = 0, k = 1, ..,K
α(
∑K
k=1Ak
xk
− 1)
= 0 α ≥ 0(8)
Note that g is an active constraint if g (X) = 0. Assuming that g
is inactive, it means that g (X) 6= 0 and α = 0, then (8) becomes
exp (xk) (xk − 1) + 1 = 0 and it results that xk = 0 , k = 1, ..,K.
This is impossible because xk> 0 .
Hence, g is an active constraint, g (X) = 0 and α > 0.
As a result, (8) leads to the following system{
∑K
k=1Ak
xk
− 1 = 0
α = exp(xk)(xk−1)+1γk
> 0 , k = 1, ..,K.(9)
Finally, solving the system (9), we finds that
xk = W
(
α.γk − 1
e
)
+ 1 (10)
where W is the LambertW function 1.
1
825
![Page 3: [IEEE 2010 Ieee Globecom Workshops - Miami, FL, USA (2010.12.6-2010.12.10)] 2010 IEEE Globecom Workshops - Novel power and time allocation algorithm for a Dynamic TDMA slot assignment](https://reader031.vdocuments.site/reader031/viewer/2022020615/575095071a28abbf6bbe3f66/html5/thumbnails/3.jpg)
Step 3: The obtained solution
Let us define the function
h (α) =
K∑
k=1
Ak
W(
α.γk−1e
)
+ 1− 1. (11)
It is important to mention that h is a decreasing function of α.
Furthermore, the Lambert W function is positive for α > 0 and
is strictly concave, then limα→+∞
h (α) = 0 and limα→0
h (α) = +∞. We
can conclude that it exists a unique α solution of h (α) = 0.
We deduce from the system (9) that
P∗k =
exp(
W(
α.γk−1e
)
+ 1)
− 1
γk. (12)
We recognize the expression of the capacity and the percentage of
the channel spectral occupancy (see Section II.1):
ck = B ln 2
(
W
((
α.γk − 1
e
)
+ 1
))
,
tk =Ak
(
W(
α.γk−1e
)
+ 1) .
It is clear that an explicit expression of α cannot be readily
obtained. However, the equation (11) can be solved numerically.
For finding the root of (11), we consider an iterative algorithm
deduced from the Fixed Point Method algorithm (see Algorithm 1)
with an appropriate initialization α0 defined as:
α0 =eAeqexp(Aeq) + 1
gb(13)
where Aeq = W (α0gb−1e
) =∑K
k=1 Ak and gb =∏K
k=1(γk)1K
.
We start the iterative process by this initial value (13) and we
consider the function defined as
en =K∑
k=1
Ak(
W(
αn γk−1e
)
+ 1) − 1. (14)
This iterative process stops once en = 0.
Algorithm 1 alpha search algorithm
Initialize α0,eps,e0and µ
en = e0;
αn = α0;while (en > eps)αn+1 = αn(1 + µen);calculate en+1 using (14);
en = en+1;
end
The appropriate initial value α0 further reduces the number of
iterations and the convergence is mostly obtained less than 10iterations.
IV. POWER ALLOCATION WITH ADMISSION CONTROL
As it was mentioned before, since the power resource at the
Base Station (BS) is limited, all user’s power requirement may not
be satisfied. An admission control mechanism should be employed
to determine which users to be admitted into the network. This
scenario is important for real-time multimedia application in order
to achieve the required QoS performance.In [9] authors proposed a
The Lambert W function is the inverse function of f (w) = wew whereew is the exponential function and w is any complex number.
cross-layer model based on adaptive resource allocation scheme by
developing different admission control and power time-slot allocation
algorithm. The joint of power allocation and admission control has
been considered in [10]. The global problem is posed as a two-
stage optimization problem. In the first admission control stage, we
define the set of admitted users. Then, we determine the amount of
power needed by these admitted users in the power allocation stage.
However, the admission control problem is combinatorial hard, which
introduces high complexity for practical implementation. Therefore,
a low-complexity solution approach for the joint admission control
and power allocation problem is highly desirable.
A. Problem formulation
We reformulate the initial optimization problem (4) by adding the
constraint of power control for each user,
minPk
K∑
k=1
λkPk
B log2 (1 + Pk γk)(C1)
subject to
K∑
k=1
λk
B log2 (1 + Pk γk)≤ 1
subject to 0 < Pk ≤ Pmax
(15)
where Pmax is the maximal power budget allowed to each user.
As we can see, the reformulated problem (15) is a no longer convex
and its resolution is not straightforward. Therefore, we resort to a
suboptimal heuristic algorithm to solve it.
B. Solution and proposed algorithm
In the following, we propose a suboptimal algorithm which over-
comes the main difficulty of finding a straightforward solution to the
considered power allocation problem (15). We propose a reduced-
complexity iterative algorithm that is constituted of three main steps.
At the beginning of this algorithm, we consider the solution obtained
in the initial optimization problem (12) We sort the initial power
allocated Pk in a decreasing order so P1 > P2 > · · · > PK . In some
cases, the second constraint is ignored, i.e Pk > Pmax. Therefore,
we propose to reallocate a new value of power to the corresponding
users. The power value that can be assigned can be equal to 0 or equal
to Pmax. In fact, we allocate to a given user, the amount Pmax only
if he can fulfill a number of bit bk = ⌊log2 (1 + Pmaxγk)⌋ 6= 0so tk is changed to tk = λk
B log2(1+Pmax γk), otherwise we assign
to him an amount of power equal to 0 and tk = 0. However, the
constraint of buffer stability cannot be respected since the value of
tk is changed. In order to solve this problem, we suggest a power
reallocation. In fact, after fixing the amount of power assigned to the
kth user and the time needed tk, we substitute this user from the
list of user K and we consider a number of user equal to K − 1.
The constraint is no longer equal to 1 but to 1 − tk. The proposed
algorithm is explained in Figure 1.
V. SIMULATION RESULTS
A. Parameters
We describe in this part the different parameters used in our simu-
lations which are given in Tab.I. We consider a single cell with 1kmradius. In the performed simulations, each user’s location is randomly
generated and evenly distributed over the cell. Users’data rate arrive
according to a Poisson process with rate λk. We consider the path
loss, shadowing and flat fading in the channel propagation model. For
path loss, we use the modified Hata urban propagation model. The
shadowing component follows a log-normal distribution with mean
826
![Page 4: [IEEE 2010 Ieee Globecom Workshops - Miami, FL, USA (2010.12.6-2010.12.10)] 2010 IEEE Globecom Workshops - Novel power and time allocation algorithm for a Dynamic TDMA slot assignment](https://reader031.vdocuments.site/reader031/viewer/2022020615/575095071a28abbf6bbe3f66/html5/thumbnails/4.jpg)
Figure 1. Flowchart of power allocation with admission control
value of 0 and standard deviation of 8dB. We perform Monte-Carlo
simulations over 103 realizations. For each configuration, we consider
T = 104 time-slots to simulate the outcome of the proposed power
allocation schemes. Although we derived an optimal power allocation
scheme, it is not straightforward to apply this result in a more realistic
scenario. In order to make the optimization problem more tractable,
we use simplified simulations parameters.
Table ISIMULATION PARAMETERS
Bandwidth B = 5MhzCarrier frequency fc = 2Ghz
Noise power σ = −174 dBm/Hz
Average of arrival rate λ = 100kbit/sTime symbol duration Ts = 200 ns
B. Results
Illustrating the optimal scheme given in Section Section III, Figure
2 represents the time allocated to each user in terms of channel
conditions (4). We notice that the time allocated to each user is
decreasing when the radio channel conditions become better. In
Figure 3, we plotted the total throughput achieved and the arrival data
rate in terms of number of users. We remark that the total throughput
is closed to the total arrival data rate to ensure the buffer stability
and to minimize the total allocated power. We fix Pmax = 1dBm.
Figure 4 shows the power allocated with admission control in terms
of channel conditions. As it is shown, we can identify from this
figure “active” users which can be accepted by the admission control,
in order to guarantee the required system performance and buffer
stability. Figure 5 depicts the total power allocated in terms of the
average of the arrival data rate when the number of user is 15 and
30. It is clearly noticeable that when the number of users increases,
the total power allocated get higher. In fact, when the number of
users increases, the arrival traffic increases and the system becomes
more crowded and the demand in power will increase. Applying the
admission control mechanism (explained previously), we can remark
a minimization in the total consumed power (remain less then 1dBm)
since the user that has a bad channel conditions and who requires
the largest amount of power are dropped.
-20 -10 0 10 20 30 400
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Normalized radio channel conditions (dBm)
pourc
enta
ge o
f tim
e a
llocate
d (
%)
Figure 2. The impact of gain channel to noise on the time allocation
5 10 15 20 25 300.5
1
1.5
2
2.5
3
Number of users
rate
(M
bit/s
)
total throughput
arrival data rate
Figure 3. total throughput versus number of users
-20 -15 -10 -5 0 5 10 15 20 25 300
0.5
1
1.5
2
2.5
3
3.5
radio channel state (dB)
Pow
er
allo
cate
d (
dB
m)
Power allocated without admission control
Power allocated with admission control
Figure 4. The impact of admission control on power allocation
VI. CONCLUSIONS
In this paper, we have proposed a non linear power and time
allocation schemes in Dynamic TDMA system to minimize total
transmitted power while ensuring buffer stability. We had suggested
an optimal outcome for the initial optimization problem. As in any
827
![Page 5: [IEEE 2010 Ieee Globecom Workshops - Miami, FL, USA (2010.12.6-2010.12.10)] 2010 IEEE Globecom Workshops - Novel power and time allocation algorithm for a Dynamic TDMA slot assignment](https://reader031.vdocuments.site/reader031/viewer/2022020615/575095071a28abbf6bbe3f66/html5/thumbnails/5.jpg)
10 20 30 40 50 60 70 80 90 1000.32
0.34
0.36
0.38
0.4
0.42
0.44
0.46
0.48
0.5
mean of arrival data rate (kbits/s)
tota
l pow
er
consum
ed (
dB
m)
15 users
30 users
Figure 5. The impact of the variation of the arrival data rate on the totalpower consumed
practical transmission situation, we are constrained by the individual
user power budget, we reformulated the initial derived optimization
problem considering the inclusion of the constraint of power control.
The joint admission control and power control algorithm aims to
minimize the total consumed power and providing better QoS.
Finally, we proposed a reduced-complexity heuristic algorithm that
results in the efficient resolution of this optimization problem.
REFERENCES
[1] C. Y. Wong, R. S. Cheng, K. B. Letaief, and R. D. Murch, ”Mul-tiuser OFDM with adaptive subcarrier, bit and power allocation,” IEEE
J.Select. Areas Commun., vol. 17, pp.1747-1757, Oct. 1999.[2] Inhyoung Kim, Hae Leem Lee, Beomsup Kim and Lee, Y. H., ”On the
use of Linear Programming for Dynamic Subchannel and Bit allocationin Multiuser OFDM ,” IEEE Global Telecommunications Conference,vol. 6, pp. 3648 -3652, 2001.
[3] D. Yu, J. M. Chioffi, ”Iterative water-filling for optimal resource alloca-tion in OFDM Multiple-Access and broadcast channels,” in Proceedings
of IEEE GLOBECOM, Nov. 2006.[4] D. Julian, M. Chiang, D. O’Neill, and S. Boyd, ”QoS and fairness
constrained convex optimization of resource allocation for wirelesscellular and ad hoc networks,”. In Proceedings of IEEE INFOCOM 2002,vol. 2, pp. 477-486, June 2002.
[5] G. Li and H. Liu, ”Dynamic Resource Allocation with Finite BufferConstraints in Broadband OFDMA Networks,” in Proceedings of the
IEEE Wireless Communications and Networking Conference, New Or-
leans, vol. 2, pp. 1037-1042, March 2003.[6] S. Najeh and H. Besbes, ”Queue-Balancing Resource Allocation for
Rate-Stable OFDMA Systems,” in Proceedings of the IWCMC 2010,July 2010.
[7] S. Kittipiyakul and T. Javidi, ”A Fresh Look at Optimal SubcarrierAllocation in OFDMA Systems,” in Proceedings of the IEEE Conference
on Decision and Control, vol.3, pp. 3289- 3294, December, 2004.[8] R. Fletcher, Practical Methods of Optimization. John Wiley and sons,
1987.[9] J.Tang and X.Zhang, “Cross-Layer based modeling for quality of service
guarantees in mobile wireless networks,” IEEE Commun. Magazine,pp.100-106, Jan.2006.
[10] E. Matskani, N. D. Sidiropoulos, Z.-Q. Luo, and L. Tassiulas, “Convexapproximation techniques for joint multiuser downlink beamforming andadmission control,” IEEE Transactions on Wireless Communications,vol. 7, no. 7, pp. 2682–2693, 2008.
828