You are on page 1of 5

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts

for publication in the IEEE ICC 2011 proceedings

Cooperative Game Strategy for IEEE 802.11s Mesh WLAN Power Management
Mohamed Esam
Faculty of Information Engineering and Technology German Uni. in Cairo (GUC) Cairo, Egypt Mohamed.Esameldin@guc.edu.eg
Abstract Mesh Wireless Local Area Network (WLAN) is a promising access network to enhance channels utilization, extend access coverage using simple and cheap network components. Power management affects mesh WLAN network coverage, topology, throughput, traffic routing and end to end delay. In this paper, the power management problem is formulated to show its impact on mesh WLAN network throughput. Then this formulation is used to propose a new ranking based cooperative game strategy. The proposed ranking of each access point depends on its cell congestion factor, number of hops to the gateway and SINR. This game strategy objective is to maximize the network average users data rate with considering mesh WLAN implementation requirements. Results show that the proposed strategy achieves higher average users data rate than non-cooperative game Nash equilibriums and comparable result with the optimum power assignment. Keywords: Mesh, Wireless, WLAN, IEEE 802.11s, Power, Management, Resource allocation, Cooperative, Game.

Mohamed Ashour
Faculty of Information Engineering and Technology German Uni. in Cairo (GUC) Cairo, Egypt Mohamed.Ashour@guc.edu.eg Wireless networks power management can be classified into up link or down link, cooperative or non-cooperative, centralized or distributed, closed or open or combined open and closed loop power management [3]. In uplink power management the concern is with power saving to support longer battery life, however downlink power management concerns with network performance enhancement like SIR optimization. Centralized power approaches [49] are based on collecting the wireless network nodes information by a central station and optimizing the transmitted power for each station. Then this central station sends the power assignment results to the network nodes. One of the advantages in centralized approaches is to obtain the optimum power assignment directly by solving one optimization problem. Finding a solution for such problems in a distributed manner may require using many iterative steps by each mesh AP. This makes the efficiency of centralized techniques higher than the distributed ones. Centralized implementations have the disadvantage of sending the whole network parameters to a central station which has a single point of failure problem. This reduces the network reliability and efficiency. Also it might not be feasible to implement centralized algorithms in distributed based technologies like mesh WLANs as mentioned in [12]. In distributed power algorithms, each wireless station should adapt its transmitted power; while performance enhancement objective can be for the mesh AP cell or for the whole system efficiency. Distributed management approaches [4, 6, 7, 912] do not have single point of failure problem and generate less signaling overhead. In case of power management, each node should assign its power level separately. This allows the network to have more rapid response toward the parameters variations. However distributed algorithms efficiency can hardly reach the centralized schemes output. This is because it is not easy for each node to determine the impact of its decision on the whole system performance. In [4-6] and [8] different centralized techniques for power allocation were presented. In these approaches the problem is modeled using linear and nonlinear programming [8], cooperative game theory [6], and KarushKuhn-Tucker (KKT) Lagrangian function conditions [4-6, 12].

I.

INTRODUCTION

IEEE 802.11s mesh WLAN is an extension to IEEE 802.11 WLAN standard for user terminal to Access Point (AP) communications as well as inter APs connections. In mesh WLAN at least one of the APs acts as the network gateway to the Internet or any other network technology. Power management is one of the main challenges for mesh WLANs [1] and [2]. This is due to its impact on users and APs SIR and so the transmission rate, network coverage, topology, traffic routing and end to end delay. Power increase enhances the received signal, network coverage and APs connectivity but increases the interference level. So SIR and users throughput may decrease. In this paper, we model mesh WLAN power management as a cooperative game. The game players are the distributed APs in mesh WLAN network and the payoff is the network average users data rate. The main contribution of this paper is proposing a new ranking based cooperative game strategy. This new power management strategy depends on applying a proposed ranking function for the mesh WLAN APs. The proposed ranking function is analyzed in section III. The idea of ranking in wireless mesh networks is applied in [15] but for centralized based channel assignment; without game strategy model.

978-1-61284-231-8/11/$26.00 2011 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

The objective functions used in these formulations includes minimizing the maximum load on mesh APs [6] or increasing the overall throughput upper bound [4] and [6]. In [6], a cooperative throughput maximization game is proposed for power assignment. In this case, each AP should randomly search in its profile space for power levels and channels that optimizes network throughput. The drawback in the proposed game in [6] is the space size where each AP should search for optimum channel and power level. This may require very long time to stabilize the network resource allocation. The game strategy proposed in this paper is down link, open loop and cooperative. In section III, the proposed strategy flow chart and ranking function are presented. The proposed strategy is then compared in section IV with optimization techniques and with Nash equilibriums of non-cooperative game modeling. Power control using a non-cooperative game is also used in [16] to maximize users' utility. In the next section, mesh WLAN power management problem is formulated. II. PROBLEM FORMULATION

Fig. 1: Mesh WLAN structure and formulation


M nm

Obj. : max DNet = Di,m


m =1 i =1
jG m

(1)

[C m C j ] Di , m = nm * m

(2)

m & j ; where K m + 1 = K j

Subj. to:

In this section, power management is represented as an optimization problem; where average users data rate is formulated as an objective function to be maximized. Fig. 1 depicts mesh WLAN scenario model, then formulas are introduced. The data rate sent by user i in cell m Di,m is the difference between the user cell AP transmission capacity Cm and sum the neighbor cells forwarded traffic capacities Cj over the number of users connected to the AP nm and random access congestion factor m . In Fig. 1, nm equals 2 for AP2. M is the number of user in the network. Kj is the number of hops for AP j to the gateway station and it is greater than Km with 1, as m is the next hop for j to the gateway. Optimization problem constraints limit the sum of receiving traffic capacity from neighbor APs to be less than the AP forwarding capacity (3). Also the users should be provided with minimum data rate that corresponds to a certain Quality of Service (QoS) level (4). Gm is the set of APs that are neighbors to access point m. So AP j belongs to Gm, but AP y does not (6). gm is the number of neighbors to AP m (the number of j nodes). The congestion factor is m such that (1/ m ) is the throughput deduction due to collisions. m is directly proportional with the sum of gm and nm [14]. This analysis applies Shannon limit as a reference for transmission data rate capacity. SINR is the received power by AP m from the neighbor AP j over the sum of the interfering signals on m by nodes of y plus the thermal noise power; as shown in (6). Each AP has different channel capacity based on the channel SIR. The relation of SINR in (6) is without applying channel assignment techniques, so all APs and users terminals are using the same channel. Increasing wireless channels capacities optimizes the average users data rate. To increase the channel capacity in (5), SINR should be enhanced. From (6) and (7), SINR is maximized with decreasing the interfering APs transmission power or increasing the neighbors APs power. Ch(m,j) in (7) is the channel path loss between APs m and j.

Cm > C j
jGm

(3)
(4) (5) (6)

Di ,m min Di ,m
Such that: Cm = B log 2 (1 + SINRm )
S IN R m , j = N
p

P rm (j) + P rm (y )
y= 1 M

Where: y {m Gm }

Prm (j) = Pt j * Ch(m, j)


Cm Cm m f (nm + g m ) III. PROPOSED COOPERATIVE GAME STRATEGY

(7) (8) (9)

The concern of this section is to explain the proposed distributed rank based cooperative game strategy for mesh WLAN power management. Firstly, we present the game strategy, and then the ranking function is analyzed.

A.

Game strategy In this strategy, all APs start with the maximum transmission power which is 20 dBm in WLAN APs. Then each AP has a rank for power reduction. From the flow chart of Fig. 2, each AP can know its own rank in the 4th block after obtaining the parameters in the 3rd block. The rank of the rest APs in the same mesh network can be determined by sending the rank of each AP included with the routing messages. So in this case the routing messages will include the APs addresses as well as their corresponding ranks. So there will be a rank table in each AP such as the routing table. In this case, each AP can know the top ranked AP after certain time (Tout). The

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

station that finds its self the top ranked AP should reduce its power level with 2dB. To keep network connectivity, this power reduction step should be undone if it leads to AP or users station isolation.

B.

Ranking function In this subsection, the proposed strategy ranking function is analyzed in order to maximize the average users data rate. The proposed ranking function assigns a rank to each mesh AP such that the top ranked node has the turn to reduce its transmitted power with 2 dB step with outage constraints consideration. After certain number of iterations, each mesh AP will be assigned to the minimum power level it should support. The impact of the proposed ranking function on the strategy performance is discussed in simulation results section. From the problem formulation in the previous section, to maximize the average data rate per user (2), the AP capacity should be maximized starting with the lower number of hops APs. To maximize the AP capacity, the SIR of the AP should be maximized (3). On the other hand, the congestion factor m should be minimized. This means that in the ideal case, APs with lower number of hops should have fewer users compared to higher number of hops APs, and the same number of hops APs should have the same number of users. This is because the lower numbers of hops APs have more traffic that should be forwarded toward the gateway via these APs. To achieve the above objective by distributed power management, the rank function considers both SIR maximization and collisions minimization. The top ranked AP is which has the turn for power reduction. The proposed ranking function is:

Fig. 2: Power management strategy flow chart


M D Dk = ( k-1 ) + (11) k-1 k-1 m =1 N paths, m NPaths,m are the number of paths between each of AP m with k hops APs and k-1 hops APs. Finally k-1 is the congestion factor due to CSMA/CA medium access control. Fig. 4 is determined by plotting (11) as a function of k. Assuming all traffic is toward or from the AP1 as a gateway, it is noted that the output curve can be approximated to (12) which is a negative exponential relation: Dk LGW * exp(-CR * k) (12) LGW is the gateway node load which is constant that does not affect the ranking function. This means that the load on each AP has an exponential decrease with increasing the number of hops to the gateway. So the rank function should have direct proportional with (12) which means to have negative exponential relation with the number of hops k such that lower k APs have higher rank. The results of Fig. 4 are calculated with assuming each AP has the same traffic Tcell due to the cell users load. For the same number of hops APs, the maximum AP load is plotted. For example AP3, AP5 and AP7 have the same number of hops to the gateway while AP5 is the most loaded, so it is the plotted one at number of hops equals 2.The values of Fig. 4 are calculated at Tcell = 40 Mbps. On the other hand to enhance the network throughput, the congestion factor k1 should be also minimized. k1 is based on the number of users in the cell who are competing for transmission. From [14] analysis on WLAN, the relation of the congestion factor m is represented in (13): 1 m = (13) Th(a,n,w,Nu ) Where Th is the AP throughput; which is a function of a, n and w. a is the probability of an idle station to receive a frame. n is the ratio of the frame transmission to contention window periods and w is the number of slots in the contention window.

Rank = m * e( CR*k + SIRm ) (10) CR is the ranking constant (equals 0.6 if the gateway node at the network edge and equals 1.2 if the gateway node in the network center). The rank function parameters can be obtained by each AP to know its own rank, which facilitates using distributed based power management. To prove the above rank function, we find the relation between the rank function and each of its parameters. The load on each mesh WLAN relay node is mainly due to the cell users load and the load of the neighbor APs. To formulate the AP load function, we assume the network diagram in Fig. 3. It is assumed that the area of Fig. 3 should be covered by dividing the total area over the average AP cell area and so 9 APs are required. By using Monte-Carlo estimation, the locations of the APs should be as shown in Fig. 3. If APs data routing cost is the number of hops to the gateway node, the load on each AP is determined by (11). In case multiple paths have the same cost, the traffic sent by a certain AP is divided equally over the number of the available paths to the gateway node. For example AP9 divides its traffic equally between AP6 and AP8. Where M is the Number of APs with k hops, and Dk is the load of the neighbor AP that has number of hops to the gateway equals k, D k1 is the load on the AP which has number of hops k-1 due to the cell users traffic.

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

Nu is the numbers of users share the same medium. The parameters of (13) are set such that their values fit 802.11g standard. The number of mini slots n is assumed to be 15 and w is 16. Since m represents the users congestion factor inside the AP cell, the most congested cell should have higher priority for power reduction. By connectivity constraints the out of coverage users should be switched to the neighbor APs which have less congestion and so higher transmitted power. To find the relation between the ranking function and the AP SIR SIRm,j, the main concern is to find the AP that its power reduction will result in higher SIR enhancement in the whole system; not in a specific AP cell. In this case, this AP should have higher power reduction opportunity than the residual access stations. By using (11), the total SIR SIRTotal is assumed as the summation of each AP SIR. SIRm,j is defined in (6) in the problem formulation section.
SIRTotal = SIRm
m =1 M

Fig. 4: Exact versus approximated load on the APs of Fig. 1 network

(14)

S IR T o ta l (15) Pm Before starting the power reduction, the strategy should select the best AP to start with. From SIR point of view, the best AP to start its power reduction is based on (15). In (15), the first derivative of the total SIR by each access station transmitted power Pm should be found. By this derivative, the impact of increasing the transmitted power of station m on the SIRTotal is calculated. From the mathematical definition of SIRTotal first derivative, it represents the impact of increasing the transmitted power by AP m with very small step on the SIRTotal. The minimum differentiation value with respect to AP m represents large reduction in total SIR due to power increase by AP m. So the minimum differentiation value should be selected to have highest SIRm with Pm decrease. The proposed power management strategy assumes that all APs transmit firstly with the maximum power and the challenge is to select the AP that its power reduction should start to maximize the sum of users data rate. Fig. 5 is the output of applying (14) and (15) on a randomly distributed APs mesh networks. As shown there is a relation between SIRm and SIRm; the minimum SIRm APs have the minimum SIRm. This has a logical explanation, as the highest interfered APs have the lowest SIR as shown in Fig. 6 for AP 2. So the power reduction of this AP leads to lower interference on many neighbor stations. Although this may harm the selected AP cell users, but the total SIR will be enhanced.

S IR m =

Fig. 5: Relation between Total SIR first derivative and APs SIR

Fig. 6: Highest interfered AP. Circles are APs and crosses are users

IV.

COMPARISON RESULTS

In this section, the concern is to test the proposed power management strategy and to compare it with the optimum power assignment and non-cooperative game modeling. The tests are run using MATLAB R6. The proposed strategy is tested using the parameters in Table I for uniform distribution randomly allocated APs and users. When the APs play a non-cooperative, they have selfish attitude [16]. In this case, each AP will decide its own power level to maximize its cell users' data rate regardless of this decision impact on the overall system performance.
TABLE I TESTING PARAMETERS Value Parameter 400 * 400 m2 Power reduction step 100 User stations sensitivity 5 Access Point sensitivity Muli-Path Gateway node Out door Carrier frequency 20 dBm Ptx min

Fig. 3: Network topology for load function formulation

Parameter Area No. of users No. of APs Channel Model Environment Ptx max

Value 2 dBm -40 dBm -44 dBm Station 1 2.4GHz 2 dBm

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2011 proceedings

Non-cooperative game shows more fairness than the cooperative and optimum techniques between mesh APs.

* * *

ACKNOWLEDGEMENT This research is achieved in cooperation with Prof. Paul Kuehn (IKR Inst. Stuttgart Uni. in Germany) and Dipl. Ing Bernd Gloss (Alcatel-Lucent Stuttgart Bell Labs. in Germany). We thank them very much for their comments and discussions about this paper work. REFERENCES

Fig. 7: Cooperative, Non-Cooperative and Optimum power management average users data rate

[1] [2]

Nash equilibrium is considered as the solution that makes all players satisfied; however from system point of view there is no guarantee that this is the optimum solution [16]. To find Nash equilibrium states of Table I network, a payoff matrix is built that includes the average user's data rate for all possible power combinations of all APs. Using MATLAB R6, this payoff matrix is built and Nash equilibrium combinations are determined. Nash states have maximum 58% and minimum 10% of the optimum average user's data rate. The optimum power management is the power assignment to the APs that optimizes the average user data rate Davr as an objective function. To find the optimum power assignment, it is more efficient to find a solution that makes use of the discrete levels of the Pm as optimization variable. This solution is to find all the available power levels combinations and filter out the combinations that result in user or AP outage and then to test the rest combinations by iterative search to find the power assignment that leads to maximum average users data rate. As shown in Fig. 7, the proposed rank based power management strategy iterations enhance the average users data rate until achieving comparable result (85%) of the optimum power assignment, however it has distributed and simpler implementation. Non-cooperative game has more fairness between the APs than the cooperative and optimum techniques. The minimum users data rate using noncooperative game is 1.2kbps; however it is almost 0.1kbps for cooperative game and optimum techniques. V. CONCLUSION

[3] [4]

[5] [6]

[7] [8]

[9]

[10]

[11]

In this paper, a down link rank based open loop distributed cooperative game strategy is proposed for mesh WLAN power management. The strategy meets the implementation requirements of mesh WLAN technology with an objective to maximize the average users data rate. Testing results show that the proposed strategy achieves a comparable result (85%) of the optimum power assignment average users data rate. However it has simpler implementation for mesh WLAN. The proposed cooperative strategy achieves higher performance than the selfish non-cooperative strategy for mesh APs power management. The results show that noncooperative strategy achieves maximum of 58% of the optimum result. And the average of Nash states achieves 24%.

[12]

[13]

[14] [15]

[16]

I. Broustis, J. Eriksson, S. V. Krishnamurthy, and M. Faloutsos, Implications of Power Control in Wireless Networks: A Quantitative Study, in PAM, pp. 8393, 2007. S. M. Faccin, C. Wijting, J. Kenckt, and A. Damle, Mesh WLAN networks: concept and system design, Wireless Communications, IEEE [see also IEEE Personal Communications], vol. 13, no. 2, pp. 10 17, 2006. Z. Han and K. J. R. Liu, Resource Allocation for Wireless Networks: Basics, Techniques, and Applications. New York, NY, USA: Cambridge University Press, 2008. M. Wiczanowski, S. Stanczak, and H. Boche, Performance and Interference Control in Wireless Ad hoc and Mesh Networks A Generalized Lagrangian Approach, IEEE Transactions on Signal Processing,@ARTICLE4435902, journal=Network, IEEE, year=2008, month=Jan.-Feb. , volume=22, , vol. 56, pp. 40394052, Aug. 2008. H. T. Cheng and W. Zhuang, Joint power-frequency-time resource allocation in clustered wireless mesh networks, Network, IEEE, vol. 22, pp. 4551, Jan.-Feb. 2008. Y. Song, C. Zhang, and Y. Fang, Joint Channel and Power Allocation in Wireless Mesh Networks: A Game Theoretical Perspective, Selected Areas in Communications, IEEE Journal on, vol. 26, pp. 1149 1159, September 2008. M. J. Neely, E. Modiano, and C. E. Rohrs, Dynamic Power Allocation and Routing for Time Varying Wireless Networks, in IEEE Journal on Selected Areas in Communications, pp. 89103, 2003. M. Haidar, R. Akl, H. Al-Rizzo, Y. Chan, and R. Adada, Optimal load distribution in large scale WLAN networks utilizing a power management algorithm, in Sarno Symposium, 2007 IEEE, pp. 15, 30 2007-May 2 2007. J. Yuan, Z. Li, W. Yu, and B. Li, A Cross-Layer Optimization Framework for Multihop Multicast in Wireless Mesh Networks, Selected Areas in Communications, IEEE Journal on, vol. 24, pp. 2092 2103, Nov. 2006. N. H. Zamora, J.-C. Kao, and R. Marculescu, Distributed powermanagement techniques for wireless network video systems, in DATE 07: Proceedings of the conference on Design, automation and test in Europe, (San Jose, CA, USA), pp. 564569, EDA Consortium, 2007. T. Holliday, A. Goldsmith, P. Glynn, and N. Bambos, Distributed power and admission control for time varying wireless networks, in Global Telecommunications Conference, 2004. GLOBECOM 04. IEEE, vol. 2, pp. 768774 Vol.2, Nov.-3 Dec. 2004. L. Chen and J. Leneutre, A Game Theoretic Framework of Distributed Power and Rate Control in IEEE 802.11 WLANs, Selected Areas in Communications, IEEE Journal on, vol. 26, pp. 11281137, September 2008. D. Avidor, S. Mukherjee, and F. Onat, Transmit Power Distribution of Wireless AdHoc Networks with Topology Control, in INFOCOM 2007. 26th IEEE International Conference on Computer Communications. IEEE, pp. 4652, May 2007. F. Gebali, Computer Communication Networks Analysis and Design. Victoria: Northstar Digital Design Inc, third ed., 2005. H. Skalli, S. Ghosh, S. Das, L. Lenzini, and M. Conti, Channel Assignment Strategies for Multiradio Wireless Mesh Networks: Issues and Solutions, Communications Magazine, IEEE, vol. 45, pp. 8695, November 2007. C. Saraydar, N. Mandayam, and D. Goodman, Efficient power control via pricing in wireless data networks," Communications, IEEE Transactions on, vol. 50, pp. 291-303, Feb 2002.

You might also like