A Game Theoretic Model for Wireless Medium Access Control in the Presence of Hidden Terminals

S.H.E. Mortazavi Najafabadi and C.C. Constantinou

School of Electronic, Electrical and Computer Engineering, University of Birmingham, B15 2TT, UK email: {sem928, c.constantinou}@bham.ac.uk

AbstractThis paper proposes a game-theoretic random channel access model, compliant with the IEEE 802.11 standard that can be integrated into the distributed coordination function. The objective is to design a game theoretic model that can optimize both throughput and channel access delay in each node in the presence of hidden terminals and thus optimize fairness. We propose a utility function that can decouple the protocols dynamic adaptation to channel load from collision detection. We demonstrate that our model can reach a Nash equilibrium resulting in a stable contention window, provided that a node adapts its behavior to the idle rate of the broadcast channel, coupled with observation of its own transmission activity. Simulation results show that this model is capable of achieving lower channel access delay and better throughput than the standard IEEE 802.11 distributed coordination function.

I. I NTRODUCTION The success and ubiquity of the IEEE 802.11 wireless local area network (WLAN) technology and in particular its medium access control (MAC) protocol [1] are well established and merit no introduction. The role of the MAC distributed coordination function (DCF) as the underpinning contention-based random channel access method has received much attention in the past decade [2][8] to quantify its performance and improve its throughput and fairness. In [9], [10] it is shown that the performance of DCF is highly dependent on the number of hidden nodes in the network: These studies show that the presence of hidden nodes causes signicant degradation in fairness and throughput due to the lack of time-synchronization between the hidden nodes [11]. As DCF is expected to continue to play a prominent role in 5th generation mobile communication technologies to address the issue of high bandwidth demand at base stations, eliminating its shortcomings remains a signicant challenge. We seek a channel access method that decouples the channel access probability, Pi , from the conditional collision probability, Pic , to minimize the effect of de-synchronization amongst hidden nodes. Thus, in order to achieve good short-time fairness and higher throughput, we propose a game-theoretic random access model based on CSMA/CA. Our protocol computes a suitable contention window size independently for each node in the network, resulting in an appropriate channel access probability for all nodes through local observation of consecutive idle slots between transmissions. The design of our game and its utility function is reverse engineered using a system control point of view from the desired operation point using the model in [2], [9]. We adopt a non-cooperative, static,

game-theoretic approach to model the behavior of contentionbased wireless medium access. In non-cooperative games, the players make rational, self-enforcing decisions considering only their individual payoffs. Due to the nature of the wireless channel, we use a static game in which all players make decisions without knowledge of the remaining competing players strategies. In designing and assessing utility and payoff functions, a 802.11 DCF network with a single access point (AP) in the presence of hidden nodes is assumed. We employ the idle sense method in [12] in designing our utility and payoff functions, to derive the optimal number of consecutive idle time slots between transmissions based on the channel modulation scheme and thus avoid or minimize contention. A gradient play solution is used as a strategy update mechanism to dynamically control the contention window size, by adjusting a current value of Pi gradually towards its optimal value only based on the locally observed number of idle slots between transmission attempts. We demonstrate that our model can improve network throughput and channel access delay and ultimately its short-time fairness, through detailed numerical simulations. II. A NALYSIS OF C HANNEL C ONTENTION The operation of the DCF channel access mechanism is well documented in [1] and will not be repeated here in the interest of brevity. We conne our remarks to the scenario depicted in Fig. 1. Taking X as the source for a particular transmission and AP as the destination, the back-off time measured in b slots is a random time chosen uniformly from a distribution [0, CW ], where CW is the contention window dened by the protocol. As shown in Fig. 2 when a source transmits a data frame to the destination, the covered nodes sense the channel as busy and defer their transmission by freezing their backoff timers. A packet collision between the source and a covered node occurs only if the covered node transmits its frame before it senses the transmission from the source. The possible collision period between two nodes is called the vulnerable period (VP) [9], [11]. This period for a covered node in the basic access method is the minimum period of time from the instant the source starts to transmit a data frame to the instant the covered node detects this ongoing transmission (dened as one slot time in [1]). On the other hand, when the source transmits a data frame to the destination, hidden nodes will sense the channel as idle until they receive an acknowledgement (ACK) frame from the destination. Any hidden node can transmit

measure) Pic (t), but not those of other nodes. It can then adjust Pi (t) based on a system control point of view, written as:
X

AP

Pi (t + 1) = Fi (Pi (t), Pic (t)),

(1)

Fig. 1. AP scenario in the presence of hidden nodes from the node X point of view: The dashed circle represents the carrier sense range of X, the solid black circles represent nodes covered by both X and the AP, whereas the open circles represent hidden terminals for X but not the AP.

where time is slotted, in units of a slot duration.The conditional collision probability depends on the channel access probability (i.e. the state of the broadcast medium), represented through a vector P (t) = (Pi (t), i N ): Pic (t) = Ci (P (t)). (2)

DIFS

SIFS Data Frame

Backoff

DIFS

Source

Destination

ACK Backoff

Here, Fi models the contention resolution algorithm and Ci models a mechanism which updates the conditional collision probability in DCF. We assume that Eq. (1) and (2) have an equilibrium (P , Pc ). The xed point of Eq. (1) denes an implicit relation between the equilibrium Pi (t) and Pic (t). Then, by the implicit function theorem [13], there exists a unique, continuously differentiable function Fi , such that, Pic = Fi (Pi ). (3)

Covered Stations

Backoff Timer is Frozen

Backoff VP

Hidden Stations

Backoff VP

Backoff Timer is Feozen

Under the assumption of a Poisson arrival process, in DCF the conditional collision probability in the presence of hidden nodes is dened as [9], Ci (P ) = 1 (1 Pi )nc 1 (1 Pi h )nh , i N.

(4)

Fig. 2. The vulnerable period (VP): The basic access method. For details of all quantities, c.f. [1]

another frame to the destination during this period (a data frame together with a short interframe space (SIFS) time duration). Hence, the vulnerable period for a hidden node in the basic access method equals the length of a data frame as shown in Fig. 2. As expected, the backoff period needs to be minimized to increase throughput, but maximized to reduce contention. From Fig. 2 it is evident that VP depends on the backoff delay and the data frame duration. We will explain how game theory is used to model the DCF in order to minimize the collision probability in the vulnerable period, without increasing CW as aggressively as IEEE 802.11 DCF does. To determine an accurate relationship between VP and the packet collision probability, we use [9] as the underlying analytical model to compute Pic and Pi . III. F ORMULATION OF G AME T HEORETIC R ANDOM ACCESS M ODEL Consider a set of wireless network nodes N with DCF as the contention protocol. Associated with each wireless node i N is its channel access probability Pi (t) and a certain contention measure observed by each node through the backoff mechanism. As stated in the description of DCF, each node in the network can observe its own channel access probability Pi (t) and its conditional collision probability (contention

where nc is number of the covered nodes including the transmitting node, and nh is the number of the hidden nodes. The total number of contending nodes in the network N is N = nc + nh and Pi h is dened shortly. The CW in DCF is formulated in terms of the node channel access probability, Pi . The CW easily maps to a backoff mechanism by the following correspondence between Pi and a constant contention window CW , that holds in saturation mode [2], [9] (the saturation mode implies that every node has a packet ready to transmit at all times) and is is independent of Pic : Pi = 2 . CW + 1 (5)

Moreover, it is also shown in [2], [9] that Pih can be written as a function of Pi as, Pi h =

V + 1

V (V + 1) 2CWmin

Pi ,

(6)

where V is vulnerable period [9]. We consider a slowly varying contention window in our utility function design (theoretically we treat it as a constant) in order to preserve the independence of Pi from Pic . As will become clear later, such a choice of constant contention window in the protocol plays a crucial role in guaranteeing the uniqueness of the Nash equilibrium and enabling the correct design of a payoff function. Therefore, we dene the utility function of each node i as:

Ui (Pi ) =

Pi

Fi (Pi )dPi .

(7)

vi

Based on the lack of dependency of Pi on Pic embodied in Eq. (5) and the utility function (7), a random access game G is dened as a triple G = {N, (Si )iN , (ui )iN }, where N is the set of players (nodes in the network). Player i N , adopts strategy Si = {Pi |Pi [vi , Pi ]} with 0 < vi Pi < 1, and a payoff function ui (P ) = Ui (Pi ) Pi Ci (P ), with a utility function Ui (Pi ) and price function Pi Ci (P ). Note that the throughput and fairness of node i depend on Pi (as throughput is proportional to Pi in the absence of collisions). Pi Ci (P ) is represents the contention price for node i, for a given Pi . Therefore, potentially each node can calculate precisely the cost of its action based on Pi and, with the knowledge that Pi and Ci (P ) are independent of each other, has the ability to maximize its payoff function regardless of having direct knowledge of any other players action at any specic time. A. Nash Equilibrium The objective of the noncooperative game is to nd the equilibrium solution for a wireless network with self-interested nodes in a distributed manner. A well-known solution for noncooperative games is the Nash equilibrium [14], which jointly maximizes all the players utilities. Based on the game dened in section III, the Nash equilibrium satises the following condition [14]:
ui (Pi , P i ) ui (Pi , P i ), i N, Pi Si ,

conditional collision probability), it follows that Ui (Pi ) is a non-decreasing function: This implies that Ui (Pi ) is strictly concave. Hence ui (P ) is continuous and concave in Pi and there exists a nontrivial Nash equilibrium for game G [14]. B. Utility Function Design We formulate our utility function for each node in the network so as to ensure the existence of a unique nontrivial Nash equilibrium for each node. Based on the idle sense method [12] we reverse engineer DCF from the desired operation point. Consider a node attempting to transmit in a given time slot. We denote the collision probability Pic as the probability that two nodes in the network have the same value of b and they transmit and collide at the same instant, or at least one hidden node transmits in the vulnerable period. Now let Ptr be the probability that there is at least one transmission in the considered time slot: Ptr (P ) = 1 (1 Pi )N . (11)

Thus it is straightforward to express the probability of a successful transmission in a given slot, Ps , if N nodes contend for the channel: Such an event requires a transmission attempt by a single node and the absence of all the others, including hidden and covered nodes: Ps (P ) = N Pi (1 Pi )nc 1 (1 Pi h )nh . Ptr (P i)

(12)

The collision probability in a slot can be expressed similarly: Ci (P ) = 1 Ps (P )Ptr (P ) Piidle (P ), where Piidle is the probability of an idle slot, given by: Piidle (P ) = (1 N Pi )(1 Pi )nc 1 (1 Pi h )nh .

(8)

(13)

where ui (.) is the payoff function of player i, and is a Nash equilibrium strategy vector for all the players except player i (P i (P1 , ..., Pi 1 , Pi+1 , ..., PN )). Eq. (8) states that each player i, in choosing P i is playing its best response to all the other players strategy choices. This statement is true if P i satises following condition [14]: ui (Pi , P i ) = 0, i N. Pi (9)

P i

(14)

Note that Eq.(9) does not guarantee that the equilibrium is unique and can admit solutions where the Nash equilibrium, Pi can take values at the boundaries of the strategy space, resulting in poor performance in terms of throughput and channel access delay. However, if it satises Eq. (10) then the equilibrium Pi is guaranteed [14] to be a unique nontrivial equilibrium: Ui (Pi ) = Ci (P ), i N. Pi (10)

In [12], a medium access method is proposed which uses the mean number of consecutive idle slots between two transmission attempts. Let us denote by n i the mean number of consecutive idle slots between two transmission attempts (successful transmission, collision or frame corruption). This can be written as a function of Piidle as: n i = Piidle (P ) . 1 Piidle (P ) (15)

To derive our utility function, we optimize the throughput by minimizing collision overhead and the time spent in idle slots. We express the throughput as a function of Pi (c.f. Eq.(5)) using [2], [9], as: S (Pi ) = Ps Ptr L . (16) (1 Ptr )Tslot + Ptr Ps Ts + Ptr (1 Ps )Tc

In order to meet the condition (10) the utility function Ui (Pi ) must be continuously differentiable and strictly concave. Since Eq. (7) is an integral, and since Pic = Fi (Pi ) and Pic 0, Fi (Pi ) is a continuous decreasing function (the larger the channel access probability, the smaller the

In this expression Tslot is a slot duration which is a xed protocol parameter. Tc and Ts are the average collision and transmission durations, respectively, and L is the packet size.

The average busy time after a collision by two mutually hidden nodes is denoted as Tc (hidden) and the average busy time after a collision by two mutually covered stations is denoted as Tc (covered) . Let Tint denote the time interval between the start times of two colliding transmissions [9]. Denoting = P HY(header) + M AC(header) , being the propagation delay and ACK(timeout) = SIF S + ACK + DIF S , we have: Ts = + L + + SIF S + ACK + + DIF S Tc (covered) = Tint (covered) + + L + + ACK(timeout) Tint (covered) = (1/2) Tslot Tc (hidden) = Tint (hidden) + + L + + ACK(timeout) Tint (hidden) = (1/2) ( + L) Tc = (nc /N ) Tc (covered) + (nh /N ) Tc (hidden) . Our objective is to calculate the optimal value Piidle , so we are setting the rst derivative of the throughput in Eq. (16) to zero, S (Pi )/Pi = 0, to obtain n opt for a given numbers i of competing and hidden nodes and for values of Piidle in [0, 1], corresponding to a variant of 802.11 summarized in Table II and Eq. (17). This leads to Pi . Eq. (14) and (15) give the optimal number of idle slots between two transmission attempts n opt and thus the optimal contention window size. i We summarize n opt for variable numbers of hidden nodes as i the lower triangular matrix i :

where Mi is dened based on the number of hidden nodes that node i sees, as well as the total number of nodes in network, i (k, j ) , i N. (20) 1 + i (k, j ) In order to reverse engineer the utility function from the desired point of operation, it can be easily shown that, Mi = Piidle (P ) = (1 N Pi )(1 Ui (Pi )), i N. (21)

(17)

where a prime denotes a derivative with respect to the function argument. According to Eq. (19) the utility function at its maximum should be Piidle (P ) = Mi . By applying Eq. (7) we obtain the utility function. This function turns out not to be a strictly non-decreasing function; hence it cannot guarantee the uniqueness of the Nash equilibrium. A convenient approximate choice that satises the uniqueness of the Nash equilibrium is, Piidle (P ) = Mi (1 + Pi ). (22)

Note that when N is large, the optimal channel access probability that maximizes the throughput is very small [9], [12]. Applying Eq. (7) we derive the utility function as follows,
Ui (Pi ) = Pi ln(N Pi 1)(Mi + N Mi ) Mi +1 + . (23) N N2

Therefore, the payoff function now becomes,

ui (P ) = Pi ln(N Pi 1)(Mi + N Mi ) Mi +1 + N N2 (24) Pi Ci (P ).

n opt i(2,0) n opt i(3,0) . . . opt n i(k1,0) n opt i(k,0)

n opt i(3,1) . . . opt n i(k1,1) n opt i(k,1) 0 . n opt i(k,2) .. n opt i(k1,j ) (18)

n opt i(k,j )

where k is the total number of active nodes in the network and j is the number of hidden nodes from each nodes point of view. If there are a total of 5 nodes in the network, a node with no hidden terminals should choose i (5, 0) and another node with one hidden node should choose i (5, 1) as its optimal n opt i . According to the above procedure, the opt variation of n i is highly sensitive to the vulnerable period. Therefore, when the total number of stations increases and the fraction of covered and hidden nodes changes, n opt varies i dramatically and does not converge to the same value across the network. In order to reach equilibrium, stations must mirror the fact that they observe differing numbers of consecutive idle slots (c.f. Eq. (18)) between two transmission attempts, based on their associated number of hidden stations. Any deviation from n opt will result in decreasing neti work performance in terms of both throughput, fairness and increased channel access delay. Therefore, without loss of generality, Piidle becomes, Piidle Mi , i N.

C. Dynamics of Random Access Game We proceed to investigate how interacting players converge to a Nash equilibrium. We consider repeated play of the random access game and seek an update mechanism in which players repeatedly adjust strategies in response to observations of other player actions, so as to achieve the Nash equilibrium. We adopt the gradient play [15] strategy update mechanism. Each node i N updates its strategy according to, Pi (t + 1) = Pi (t) + E (P )[Ui (P i (t)) Ci (P (t))]. (25) where E (P ) > 0 is the update step size. At stage t + 1, node i N chooses a channel access probability derived from stage t. Then each node i N updates its contention window CWi as follows, 2 Pi (t + 1) . (26) Pi (t + 1) The gradient play admits a nice economic interpretation, if we interpret the conditional collision probability Ci (P (t)) as contention price for node i. If the marginal utility Ui (P i (t)) is greater than the contention price, we increase the channel access probability. Conversely if the marginal utility is less than contention price, we decrease the access probability; then using Eq.(26) each node updates its strategy. The step size determines the convergence speed. CWi (t + 1) =

(19)

TABLE I MEDIUM ACCESS CONTROL DESIGN After each transmission{ / OBSERVE MEAN NUMBER OF IDLE SLOTS n i / sum sum + ni ntrans ntrans + 1 if ntrans maxtrans { n i sum/maxtrans / RESET VARIABLES / sum 0, ntrans 0 / COMPUTE CONDITIONAL COLLISION PROBABILITY / Ci (P ) =
1(1+ ni )Pi )(1+ (1N Pi ni )

TABLE II DSSS SYSTEM PARAMETERS AND ADDITIONAL PARAMETERS USED TO

OBTAIN SIMULATION RESULTS

Channel Bit Rate Physical Header ACK SIFS DIFS

11 Mbps 192 bits 112 bits 10 s 50 s

Slot Time MAC Header Packet Payload Propagation Delay

20 s 272 bits 12000 bits 1 s

70%
802.11b - N=20

60%

Game - Theoretic Model - N=20 802.11b - N=50 Game - Theoretic Model - N=50

50% Throughput

/ COMPUTE ITS DISTANCE FROM EQUILIBRIUM / dist [Ui (P i (t)) Ci (P (t))] / COMPUTE NEW CHANNEL ACCESS PROBABILITY / Pi (t + 1) = Pi (t) + E (P ) dist / ADJUST NEW CONTENTION WINDOW / CWi (t + 1) = }}
2Pi (t+1) (t+1) Pi

40%

30%

20%

10%

IV. M EDIUM ACCESS C ONTROL D ESIGN Using section III, we design a medium access model assuming that the AP can determine and periodically broadcast the number of active nodes. Then each station instead of using a binary exponential back-off scheme computes locally its displacement from equilibrium using Ui (P i (t)) Ci (P (t)) in a distributed manner, and uses it to adjust the size of its contention window (see Table I for a formal description of this scheme in the form of an algorithm). The variable maxtrans in Table I denotes the size of the observation window which only depends on the number of transmission attempts. As ntrans increases, n i approaches n i. A. Conditional Collision Probability Estimation In order to estimate accurately the conditional collision probability we have used the same method as [12]. Therefore, the conditional collision probability can be estimated via the observation of the wireless medium over several time slots and can be written using Eq. (4) and (14) as, Ci (P ) = 1 Piidle (P ) . 1 N Pi (27)

0% 0%

10%

20% 30% 40% Percentage of Hidden Terminals

50%

60%

Fig. 3. Saturation Throughput: DCF vs. Game-Theoretic Model

Combining Eqs. (15) and (27) yields with good accuracy, Ci (P ) = 1 (1 + n i )Pi . (1 N Pi )(1 + n i) (28)

V. S IMULATION R ESULTS In this section the performance of the proposed model is simulated and compared with the standard IEEE 802.11 DCF basic access method (i.e. not RTS/CTS). We use the OMNeT++ simulator [16] with the MiXiM modeling framework for mobile and xed wireless networks [17]. All relevant

simulation parameters are listed in Table II. We use IEEE 802.11b DSSS PHY layer [1] parameters as input to our game-theoretic random access model, as discussed in section III. maxtrans = 10, and the step size for gradient play is empirically xed as E (P ) = 0.025. To evaluate our model by simulation, we consider a hotspot scenario, where there is an AP at the center and stations are uniformly randomly distributed around the AP and with a perfect radio channel. All nodes are in saturation mode. All simulations run for 1,000 s to ensure convergence and stability of design. The saturation throughput achieved by our model against IEEE 802.11b in DCF mode is shown in Fig. 3 as the percentage of hidden nodes increases, the game-based design for total number of 20 and 50 nodes achieves a higher throughput than DCF. Fig. 4 compares the average channel access delay of different protocols for different percentage of hidden terminals in the network. it can be observed that game based design provide much better channel access delay than 802.11b. Fig. 5 compares the delay distribution histograms of the two protocols for different percentages of hidden terminals present. We see that the game theoretic based design provides much a better worse case channel access delay than 802.11b: The large histogram peak at the right end of the plots shows that 802.11 aggressively moves toward the maximum CWmax but the game theoretic model adjust its CW based on the Nash equilibrium in a much more measured way.

14 13 12 11 10 9 Delay(s) 8 7 6 5 4 3 2 1 0% 10%

802.11b - N=20 Game - Theoretic Model - N=20 802.11b - N=50 Game - Theoretic Model - N=50

access delay and better throughput which ultimately results in good short-term fairness. Our protocol is highly dependent on the estimation of consecutive idle slot times. Therefore, a wrong choice of maxtrans which denes the size of the observation window can lead to a collapse of the game. Also, the step size has to be chosen appropriately in order to converge to the desired point of operation with a reasonable convergence speed. This is an empirically arrived at compromise which balances convergence speed against protocol instability which arises by choosing a larger step size for the update mechanism in our design. R EFERENCES

20% 30% 40% Percentage of Hidden Terminals

50%

60%

Fig. 4. Average Channel Access Delay: DCF vs. Game-Theoretic Model

802.11b - N=20 0.8 Game - Theoretic Model - N=20 0.8 0.6 0.4 0.2 0

0.6 0.4 0.2 0

0.2

0.4

0.6

0.8

Delay (s)0%hidden nodes

0.8 0.8 0.6 0.4 0.2 0

Delay (s)20%hidden nodes

0.6 0.4 0.2 0

10

Delay (s)40%hidden nodes

Delay (s)60%hidden nodes

Fig. 5. Delay Distribution: DCF vs. Game-Theoretic Model (in presence of 0%, 20%, 40% and 60% hidden nodes)

VI. C ONCLUSION We have developed a game-theoretic model for contention control and propose a medium access method derived from CSMA/CA, in which each node estimates its state in a game and adjusts its persistence probability or contention window according to its displacement from the Nash equilibrium. The novelty of the protocol is that, to the best of our knowledge, it is the rst such protocol to explicitly account for the effect of hidden nodes and mitigate for it purely through local observations. The correct specication of per-node utility functions results in harmonized but independently determined performance objectives for all competing nodes. As wireless nodes can estimate the conditional collision probability by observing consecutive idle slots between transmissions, we can decouple contention control from the handling of failed transmissions. As shown in numerical simulations our gametheoretic medium access control design exhibits low channel

[1] Wireless lan medium access control (mac) and physical layer (phy) specications, IEEE Std 802.11, 1999. [2] G. Bianchi, Performance analysis of the ieee 802.11 distributed coordination function, Selected Areas in Communications, IEEE Journal on, vol. 18, no. 3, pp. 535547, 2000. [3] Y. C. Tay and K. C. Chua, A capacity analysis for the ieee 802.11 mac protocol, Wirel. Netw., vol. 7, no. 2, pp. 159171, Mar. 2001. [Online]. Available: http://dx.doi.org/10.1023/A:1016637622896 [4] F. Cali, M. Conti, and E. Gregori, Dynamic tuning of the ieee 802.11 protocol to achieve a theoretical throughput limit, Networking, IEEE/ACM Transactions on, vol. 8, no. 6, pp. 785799, 2000. [5] L. Gannoune and S. Robert, Dynamic tuning of the contention window minimum (cwmin) for enhanced service differentiation in ieee 802.11 wireless ad-hoc networks, vol. 1, pp. 311 317 Vol.1, sept. 2004. [6] H. Ma, X. Li, H. Li, P. Zhang, S. Luo, and C. Yuan, Dynamic optimization of ieee 802.11 csma/ca based on the number of competing stations, in Communications, 2004 IEEE International Conference on, vol. 1, 2004, pp. 191195. [7] L. Zhao, J. Zhang, K. Yang, and H. Zhang, Using incompletely cooperative game theory in mobile ad hoc networks, in Communications, 2007. ICC 07. IEEE International Conference on, 2007, pp. 34013406. [8] S. H. E. Mortazavi Najafabadi and C. C. Constantinou, Game theoretic approach to medium access control in wireless networks, in 2013 IEEE Wireless Communications and Networking Conference (WCNC): MAC (IEEE WCNC 2013 - MAC), Shanghai, P.R. China, Apr. 2013, pp. 878 883. [9] F.-Y. Hung and I. Marsic, Performance analysis of the ieee 802.11 dcf in the presence of the hidden stations, Computer Networks, vol. 54, no. 15, pp. 2674 2687, 2010. [Online]. Available: http://www.sciencedirect.com/science/article/pii/S1389128610001271 [10] H. Wu, F. Zhu, Q. Zhang, and Z. Niu, Wsn02-1: Analysis of ieee 802.11 dcf with hidden terminals, in Global Telecommunications Conference, 2006. GLOBECOM 06. IEEE, 2006, pp. 15. [11] A. Tsertou and D. Laurenson, Revisiting the hidden terminal problem in a csma/ca wireless network, Mobile Computing, IEEE Transactions on, vol. 7, no. 7, pp. 817831, 2008. [12] M. Heusse, F. Rousseau, R. Guillier, and A. Duda, Idle sense: an optimal access method for high throughput and fairness in rate diverse wireless lans, in Proceedings of the 2005 conference on Applications, technologies, architectures, and protocols for computer communications, ser. SIGCOMM 05. New York, NY, USA: ACM, 2005, pp. 121132. [Online]. Available: http://doi.acm.org/10.1145/1080091.1080107 [13] J. E. M. R. Abraham and T. Ratiu., Manifolds, Tensor Analysis, and Applications, second edition ed. Springer-Verlag, 1988. [14] D. Fudenberg and J. Tirole, Game Theory. Cambridge, MA: MIT Press, 1991, translated into Chinesse by Renin University Press, Bejing: China. [15] W. S. Zhu Han, Dusit Niyato and T. Basar, Game Theory in Wireless and Communication Networks: Theory, Models, and Applications. Cambridge/GB: Cambridge University Press, 2012. [16] OMNeT++ Community. OMNeT++ modeler website. [Online]. Available: http://www.omnetpp.org/ [17] OMNeT++MiXiM modeling framework for mobile and xed wireless networks. [Online]. Available: http://mixim.sourceforge.net/

Distribution

Distribution