On the Performance of Power Controlled Random Multiple Access in Wireless Networks Arash Behzad and Julan Hsu CS218 Final Project Professor Mario Gerla University of California (UCLA) Los Angeles, CA 90095-1594 {abehzad, jlhsu}@ee.ucla.edu Abstract— In this paper, we develop and investigate two novel schemes to enhance the throughput of the slotted ALOHA medium access control (MAC) mechanism. We analyze the sensitivity of the randomized power control algorithms for slotted ALOHA MAC with respect to the parameters of power distribution. In particular, we prove that, independent of the power distribution, (the upper bound for) the aggregate throughput of the network is proportional to the square of the coefficient of variation of the power distribution. We show that this result can be directly utilized in the process of designing efficient randomized power control algorithms. We reconfirm the validity of our theoretical results through simulation experiments, which are partially conducted in QualNet 3.6 environment. I. INTRODUCTION Although in random multiple access the collision of packets results in low throughput and energy dissipation which makes it unacceptable for many applications, random access in an indispensable approach in wireless networks due to its ease of implementation and low overhead. ALOHA is a widely studied and deployed medium access control (MAC) protocol: almost all deployed cellular systems use ALOHA (or one of its variants) as a mean to request network access for mobile users [4]; ALOHA is commonly used in the control channel of ad hoc wireless networks [7]; Furthermore, ALOHA is used as a multiple access protocol in local wireless communications [6]. ALOHA was first introduced by Abramson [1], where it was determined that the maximum channel utilization under “pure ALOHA” is 18 percent of the channel bandwidth. The slotted variation was then introduced by Roberts, showing its capacity is doubled (over unslotted ALOHA), to 36 percent [2]. In classical slotted ALOHA it is assumed that a transmission is successful if and only if no other user attempts to use the channel during the same slot ([2], [3]). This is a reasonable assumption when all the packets are received under nearly power levels. Clearly, in the ground radio environment such an assumption is somewhat pessimistic as the propagation loss and multipath fading can make the received power of packets at access point differ by an order of magnitude. Such a condition allows taking advantage of the capture effect, the phenomenon that the packet with the strongest power is successfully received even in the presence of other interfering packets [5]. In this paper, we develop and investigate two novel schemes to enhance the throughput of the slotted ALOHA medium access control (MAC) mechanism based on the capture effect. In our primary scheme, we first analyze the sensitivity of the randomized power control algorithms for slotted ALOHA MAC mechanism with respect to the parameters of power distribution. In particular, we prove that independent of the power distribution, (the upper bound for) the aggregate throughput of the network is proportional to the square of the coefficient of variation of the power distribution. This result can be directly utilized in the process of designing efficient randomized power control algorithms, as it is discussed later in the paper. Furthermore, we provide an analytical bound for the performance of the class of randomized power control algorithms (including the optimum algorithm if it exists). In our second scheme, we present a closed form formula for the aggregate throughput of the network as a function of the probability of transmission. We note that the latter formula is based on the capture effect under the Physical Interference Model, which takes the specific coordinates of the nodes in the system into consideration. Based on the derived closed form, the optimum probability of transmission for every given (symmetric/asymmetric) topology can be computed. The rest of the paper is organized as follows. In section II, we present the system model and the underlying assumptions. The mathematical analysis is elaborated in section III: in part A we provide our mathematical analysis regarding the optimal performance of the randomized power control algorithms. We elaborate our asymptotic results on the optimal probability of transmission under the capture effect in part B. Simulation results and conclusions are presented in section IV and section V, respectively. II. SYSTEM MODEL The nodes of the network are assumed to strive to communicate with a central entity, which we refer to as the access point (AP).1 All nodes operate in the same channel and transmit under a fixed data rate R. Nodes are equipped with identical half-duplex radios and omnidirectional antennas. During the period of operation under consideration in this paper, we assume nodes to be immobile and to be located based on an arbitrary distribution, unless otherwise is specified. The channel access protocol is slotted ALOHA. That is, the time axis is divided into identical synchronized time slots whose duration is assumed to be equal to the transmission time of a packet (which is assumed to be fixed) plus some overhead duration that includes the maximum propagation delay. Each user can transmit its packet only at the beginning of a slot. It is assumed that all nodes always have packets waiting for being transmitted (i.e., heavy traffic load). Conventionally, a newly arrived packet is transmitted in the first slot after its arrival; packets which are not received successfully are buffered and retransmitted after a random delay. However, similar to many other studies in the literature (e.g., [4],[9]), we assume that the newly arrived packets and packets awaiting transmissions are treated identically. That is, in every slot, each node transmits a packet according to a Bernoulli process with parameter q, 0 q 1 , whether it is a new packet or retransmitted packet. Let us denote the thermal noise and the transmission power from node k to the AP by N and Pk , respectively. We represent the distance from node k to the AP by dk. Without loss of generality, throughout this paper we assume that if there are M simultaneous transmissions in a slot, then the set of simultaneous transmitters is equal to {node 1, , node M } . A transmission from node k to the AP is received successfully if the signal-tointerference and noise ratio (SINR) at the AP is not less than the minimum required threshold , i.e. 1 The central node is also sometimes referred to as a base station or backbone node, according to the underlying application. Pk / d k , k 1,2,..., M , M N P /d i (1) i i 1 ik where M is the number of simultaneous transmissions in the underlying slot, and represents the path loss exponent. This model is commonly known as the Physical Interference Model [15]. Similar to many other papers in the literature, we assume to be always greater than one, which is the case in most practical scenarios. It is straightforward to show that at most one packet can capture the channel if 1 [7]. III. MATHEMATICAL ANALYSIS A. Power Control and Capture Effect Let consider a symmetric topology, whereby the distances of all nodes and the AP are identical. We denote such a distance by d. Every node may adjust its transmit power in a slot-by-slot fashion. The set of available transmit power levels (for each node) is subject to the underlying circuitry design. Such a set might include a continuous spectrum of power levels (e.g. the Prism chipset) or only a handful of discrete power levels (e.g. the Cisco/Aironet cards). The aggregate throughput of the network (in packets per slot) can be calculated as the following, TH Pr{sin gle transmission} Pr{capture} n q(1 q) n 1 Pr{capture}, 1 (2) where n is the number of nodes in the network. When a node attempts to transmit in a time slot, it selects its transmit power level randomly and independently (from the other involving nodes in the system) based on the probability density function (p.d.f.) f(P). We refer to such an algorithm and the associated p.d.f. as the randomized power control algorithm and the power distribution, respectively ([10], [11], [12], [13]). To support fairness, we assume that the underlying power distribution is identical for all the nodes in the network. Since the probability of transmission (q) is independent of the power distribution, the latter does not have any effect on the first term of equation (2). Consequently, the only effect of the power distribution on the aggregate throughput is crystallized in terms of probability of capture. Clearly, from throughput point of view, it would be optimal to select a power distribution that maximizes the probability of capture.2 In the following theorem, we address the effect of the power distribution on the aggregate throughput of the network. In particular, we demonstrate the impact of the mean and variance of the power distribution on the probability of capture. Pk / d k Pr{ node k captures M trans.} Pr{ } M N P r /d r r 1 r k M Pr{ Pk P r Nd }. (7) r 1 r k Therefore, Theorem 1.Consider a slotted ALOHA medium access control with a randomized power control algorithm whose power distribution is f(p). Suppose that the mean and the variance of the power distribution are equal to Pr{node k captures M trans.} M Pr{ Pk and , respectively. Then, independent of the power distribution, the aggregate throughput of the network is always bounded as follows: n [ M q M [1 ( M 1)]} M Pr{| Pk P [1 (M 1)] | Nd r [1 ( M 1)]}. (8) r 1 r k n TH q(1 q) n 1 1 2 r r 1 r k 2 n P [1 (M 1)] Nd (1 q) n M M 2 [1 ( M 1)]M By considering the Chebyshev’s inequality3 and noting that 2 [ ( Nd (1 ( M 1)))]2 ]. (3) M E[ Pk P ] [1 (M 1)] (9) r r 1 r k Proof. The probability of capture can be represented by conditioning on the number of simultaneous transmissions in the same slot, i.e. and M Var[ Pk P ] [1 r 2 ( M 1)] 2 (10 ) r 1 r k n Pr{capture} Pr{capture M trans.} Pr{ M trans.}. (4) M 2 (since the transmit power independently), we have Since at most one node can capture the channel, we have Pr{node k captures M trans.}. are selected Pr{ node k captures M trans.} M Pr{capture M trans.} levels (5) k 1 1 2 ( M 1) 2 . (11) 2 [ Nd (1 ( M 1))] By considering relations (4), (6), and (11), we have Considering relation (5) and due to the topological symmetry, we have Pr{capture } n [1 2 ( M 1)]M [ q M (1 q) n M . (12 ) M [ ( Nd (1 ( M 1)))]2 M 2 n 2 Pr{capture M trans.} M Pr{node k captures M trans.}, (6) where node k is an arbitrary node, k {1, , n} . Considering relations (3) and (12), we conclude that, independent of the power distribution, the aggregate We have 3 Let the random variable X have a distribution of probability about which we assume only there is a finite variance 2 . Then for every 2 Note that the energy optimization is out of the scope of this paper. k > 0, Pr{| X | k} 1 / k 2 [14]. throughput of the network is always bounded as the following, n TH q(1 q) n 1 1 n 2 n [ M q M (1 q) n M M 2 [1 2 ( M 1)]M [ ( Nd (1 ( M 1)))]2 ], Clearly, the latter result provides a significant insight into the design of the power distribution of the randomized power control algorithms. Corollary 1.2. If the optimum randomized power control algorithm exists, then it cannot increase the probability of capture to more than n which completes the proof. ■ Theorem 1 indicates that (the upper bound for) the aggregate throughput is proportional to the variance of the power distribution. This is intuitively very correct as we increase the variance of the power distribution (assuming that the mean of the distribution is fixed), there is a higher chance for capture. In the extreme, as the variance of power distribution becomes zero, it would be impossible to have a capture, which is consistent with relation (12). On the other hand, Theorem 1 indicates that (the upper bound for) the aggregate throughput is inversely proportional to the mean of the power distribution. This result is also intuitively plausible as we increase the mean of the power distribution (assuming that the variance of the distribution is fixed), we expect to have a less chance for capture. The latter is due to the fact that what really matters for capture is the ratio of powers. We note that in most practical cases, Nd is dominated by (1 ( M 1)) in relation (3). Therefore, by eliminating the former term in relation (3), the upper bound is not increased considerably. Consequently, we have n TH q(1 q) n 1 1 n n [1 2 ( M 1)]M ( )2 [ q M (1 q)n M ], (13) M 2 M [ ( ( M 1) 1)]2 which leads to the following corollary: Corollary 1.1.Independent of the power distribution, (the upper bound for) the aggregate throughput of the network is proportional to the square of the coefficient of variation4 of the power distribution. min{ 2 n [ M q M 2 M (1 q) n M [1 2 ( M 1)]M [ ( Nd (1 ( M 1)))]2 Example: Let us consider a wireless LAN consisting of 20 nodes, which their distance from the AP is 20 meter. Suppose that the underlying MAC is slotted ALOHA with the probability of transmission q=0.05. The minimum required SINR is 9.5dB. Every node is equipped with a wireless card that allows it to transmit under one of the following power levels: 1mW, 10mW, and 100mW. Based on Theorem 1, a randomized power control algorithm that selects each of the transmit power levels with the same probability (i.e. 1/3) cannot increase the probability of capture to more than 0.0804. This implies that the maximum throughput under such a mechanism cannot be larger than 0.4482 packets per slot. B. Optimum probability of transmission under the capture effect In this section, nodes are assumed to be distributed based on an arbitrary random pattern. Furthermore, all transmissions are assumed to be performed under an identical power level P (i.e., no power control mechanism is considered). The problem that we address in this section is to find the optimal probability of transmission (q), which is assumed to be identical for all nodes in the network, such that the aggregate throughput of the network is maximized. Lemma 2. Let us consider an arbitrary topology with large number of nodes operating under a slotted ALOHA medium access control with probability of transmission q. Based on the Physical Interference Model, the probability of success for transmission from node k can be calculated in a closed form as S k Q( f k (q)), where Q(.) is the Q-function, and 4 i.e. the ratio of the standard deviation to mean of the distribution. ,1}. (14 ) (15) N / P 1 / d k n r 1 r k f k (q) n q(1 q) /d n Y /d q /d r r , q 0, q 1. (16 ) 2 Proof. Based on the Physical Interference Model, the probability of success for transmission from node k to the AP in an arbitrary slot can be presented as P / dk }, n N Y P / d r n (ir , AP ), r 1 r k q(1 q) /d 2 (ir , AP )], (20 ) r 1 r k where Q(.) is the Q-function, and N / P 1 / d k f k (q) Relation (17) can be written as r /d r ] N / P 1 / d k }. q /d n q(1 q) /d r , q 0, q 1 .■ 2 r r 1 r k Theorem 3. Let us consider an arbitrary topology with large number of nodes operating under a slotted ALOHA medium access control with probability of transmission q. Based on the Physical Interference Model, the aggregate throughput of the network as a function of q can be calculated in a closed form as follows, (19 ) r 1 r k n r 1 r k r 1, if node k transmits in the underlying slot Yk . (18 ) 0, otherwise n q /d S k Q( f k (q)) , where Yk ' s, k 1, , n are independent Bernoulli random variables with parameter q, defined as follows: [Y r 1 r k (17 ) r 1 r k S k Pr{ n (ir , AP ) ~ N [ where N[A,B] is a Gaussian distribution with mean A and variance B. Consequently, relation (19) can be written as r r 1 r k S k Pr{ n TH ( q) qQ( f k (q)), ( 21) r 1 n [Y r /d r ] is a linear combination of large number r 1 r k of i.i.d random variables. Therefore, based on a generalization of the Central Limit Theorem (i.e. the Lindeberg-Feller Central Limit Theorem5) we have where Q(.) is a Q-function, and f k (q) is defined as before (relation (16)). Proof. The aggregate throughput of the network (in packets per slot) as a function of q can be calculated as the following, n TH (q ) E[ X r ], (22 ) r 1 5If the independent random variables Z1 , Z 2 ... satisfy the Lindeberg condition, then for all a < b, where X r ' s, r 1, , n are variables defined as below: dependent Bernoulli n [Z E(Z )] i lim P(a n i i 1 b) (b) (a), 1, if node r transmits successfully in the underlying slot Xr . (23) 0, otherwise n Var (Z ) i i 1 where is the normal distribution function. Note that in most practical cases the Lindeberg condition is satisfied. Therefore, relation (22) can be written as n TH (q) P( X r 1 r 1). (24 ) It can be easily shown that 1 2 P( X r 1) qS k . (25) . Considering relation (24), relation (25), and Lemma 2, we conclude TH ( q ) qQ( f k ( q )), r 1 and the proof is complete. ■ As a consequence of Theorem 3, we can analyze the aggregate throughput of the network as q converges to 1. Interestingly, based on relation (21), it can be shown that lim TH (q) q 1 n 1, if k : 1 / d (ik , AP) N / P q /d (ir , AP) . (26 ) r 1 r k 0, otherwise Relation (25) illustrates the fact that if exactly one of nodes is relatively close to the AP with respect to the other nodes (Fig. 1), then the aggregate throughput becomes equal to one; otherwise, throughput becomes equal to zero, which is intuitively very correct. IV. SIMULATION RESULTS For our simulation experiments, we developed and incorporated the slotted ALOHA MAC scheme for the QualNet environment. Our simulation study is partially based on QualNet 3.6. and the power control algorithm is not yet incorporated. In the simulation environment, 20 nodes are randomly and uniformly distributed, unless otherwise is specified. Noise power is set to be equal to 90dBm and the path loss exponent is assumed to be equal to 4. The minimum required SINR is assumed to be 10dB. Every node has an infinite reservoir of packets for transmission (i.e. heavy traffic load) and is assumed to be immobile during the period of simulation. In Fig. 2 we illustrate the effect of power control on the aggregate throughput of the network. The randomized power control algorithm selects each of the two available power levels (i.e., 1mW and 100mW) with equal probabilities. The alternative scheme always transmits under 50mW. As expected, even a very simple 6 3 AP Fig. 1. Illustration of a random multiple access, whereby node 4 is relatively close to the AP with respect to the other nodes. power control operation over the two power levels can improve the aggregate throughput somewhat considerably. When the involving nodes transmit with a probability equal to one (q=1), still the aggregate throughput under the randomized power control algorithm is strictly positive. Furthermore, we observe that q=0.05 (that is 1/n) does not provide the optimal probability of transmission. This is due to the fact that q=0.05 only maximizes the probability of a single transmission in the system (i.e. the first term in relation (2)). However, q=0.08 optimizes the sum of the probability of a single transmission and the probability of capture (see relation (2)). We note that due to the capture effect induced by the underlying asymmetric topology, the aggregate throughput for the slotted ALOHA (with no power control) under q=0.05 is significantly larger than 1/e. In Fig. 3 we examine the fairness of the latter two algorithms. As it has been illustrated, the minimum throughput (among all the nodes) under the power controlled slotted ALOHA is larger than that under the slotted ALOHA with no power control. Therefore, the randomized power control operation not only increases the aggregate throughput of the network, but also supports fairness (in the context of MaxMin). No power control (P=50mW) Throughput (packets/slot) n 5 4 With power control (P1=1mW;P2=100mW) 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0.05 0.08 0.3 0.5 0.8 1 Probability of Transmission (q) Fig. 2. Illustration of the aggregate throughput (in packets per slot) for slotted ALOHA with no power control and power controlled slotted ALOHA as a function of probability of transmission. No power control (P=50mW) r With power control (P1=1mW;P2=100mW) Throughput (packets/slot) 0.05 AP 0.04 R’ R 0.03 0.02 0.01 0 0 5 10 15 20 25 Node ID Fig. 3. Illustration of the throughput per node (in packets per slot) under slotted ALOHA (with no power control) and the power controlled slotted ALOHA. We compare our theoretical results for calculating the optimal probability of transmission with q=1/n in Fig. 4. For this experiment we assume that the distances between all the nodes and the AP are identical, which makes q=1/n an optimal choice. Clearly, the theoretical q converges to q=1/n as n becomes sufficiently large. In our next experiment, we assume nodes are randomly and uniformly distributed in an annulus defined by concentric circles of radius R and R’ (Fig. 5). Clearly, as r = R - R’ becomes smaller, the LindebergFeller Central Limit Theorem converges to the conventional Central Limit Theorem (CLT) and the speed of convergence (as a function of n) becomes faster. In Fig. 6 we show the sensitivity of our theoretical results for finding the optimal q with respect to small number of nodes. We assume 20 nodes are randomly and uniformly distributed in an annulus with parameter r, where R = 250m. For each r we find the optimal probability of transmission (q*) by running the simulation in QualNet over different values of q. We then compare the theoretical q (q_theoretic) derived according to section III with q*. Fig. 5. Illustration of an annulus representing the geometric locus of location of nodes. We observe that as r decreases the difference between the theoretical q and the optimal q converges to approximately zero, even when there are only 20 nodes in the network. Of course, when r is large, our asymptotic analysis will not remain valid for only 20 nodes. V. CONCLUSIONS In this paper, we developed and investigated two novel schemes to enhance the throughput of the slotted ALOHA medium access control (MAC) mechanism based on the capture effect. In our primary scheme, we analyze the sensitivity of the randomized power control algorithms for slotted ALOHA MAC with respect to the parameters of power distribution. In particular, we prove that independent of the power distribution, (the upper bound for) the aggregate throughput of the network is proportional to the square of the coefficient of variation of the power distribution assuming a symmetric topology. In our second scheme, we present a closed form formula for the aggregate throughput of the network as a function of the probability of transmission. Based on the derived closed form, the optimum 0.5 1 |q*-q_theoretic| Theoretical q q=1/n 0.8 0.6 0.4 0.2 0.4 0.3 0.2 0.1 0 0 0 20 40 60 80 Number of nodes (n) Fig. 4. Illustration of the convergence of the theoretical q and q=1/n for large values of n. 0 50 100 150 200 250 300 Parameter of annulus (r) Fig. 6. Illustration of the difference between the optimal q and the theoretic q when the number of nodes is small. probability of transmission for every given (symmetric/asymmetric) topology can be computed. A more comprehensive simulation study of the latter schemes with emphasis on their stability and fairness features is in progress. Moreover, generalization of the derived results in this paper to mobile multihop wireless networks is part of our ongoing research. ACKNOWLEDGMENT The authors wish to thank Professor Mario Gerla for discussions and insightful comments leading to improvement of quality of this paper. REFERENCES N. Abramson, “The Aloha System – Another Alternative for Computer Communications,” in AFIPS Conference Proceedings, vol. 36, 1970, pp. 295-298. [2] L. Roberts, “Aloha Packet System with and without Slots and Capture,” Comput. Commun. Rev., no. 5, 1975, pp. 28-42. [3] T. N. Saadawi and A. Ephremides, “Analysis, Stability, and Optimization of Slotted ALOHA with a Finite Number of Buffered Users,” IEEE Trans. Auto. Control, vol. AC-26, no. 3, June 1981, pp. 680-689. [4] A. B. MacKenzie and S. B. Wicker, “Stability of Multipacket Slotted ALOHA with Selfish Users and Perfect Information,” in Proc. IEEE INFOCOM, 2003. [1] T. Liu, J. Silvester and A. Polydoros, ”A General Performance Model for Mobile Slotted ALOHA Networks with Capture,” in Proc. IEEE ICC, 1995. [6] S. A. Al-Semari and M. Guizani, “Channel Throughput of Slotted ALOHA in a Nakagimi Fading Environment,“ in Proc. IEEE ICC, 1997, pp. 605-609. [7] W. Luo and A. Ephremides, “Power Levels and Packet Lengths in Random Multiple Access,” IEEE Trans. Info. Theory, vol. 48, no. 1, January 2002, pp. 46-58. [8] J. J. Metzner, “On Improving Utilization in ALOHA Networks,” IEEE Trans. Comm., vol. 24, no. 4, April 1976, pp. 447-448. [9] T. C. Hou and V. O. K. Li, “Transmission Range Control in Multihop Packet Radio Networks,” IEEE Trans. Commun., vol. 34, no. 1, January 1986, pp. 38-44. [10] R. O. LaMaiire, A. Krishna and M. Zorzi, ”On the Randomization of Transmitter Power Levels to Increase Throughput in Multiple Access Radio Systems,” Wireless Networks, vol. 4, 1998, pp. 263-277. [11] C. C. Lee, “Random Signal Levels for Channel Access in Packet Broadcast Networks,“ IEEE Jounal on Selected Areas in Communications, vol. 5, no. 6, July 1987, pp. 1026-1034. [12] H. Takanashi, H. Kayama, M. Iizuka and M. Morikura, “Enhanced Capture Effect for Slotted ALOHA employing Transmission Power Control Corresponding to Offered Traffic,“ in Proc. IEEE ICC, 1998. [13] F. Berggren and J. Zander, “Throughput and Energy Consumption Tradeoffs in Pathgain-Based Constrained Power Control in ALOHA Networks,” IEEE Communications Letters, vol. 4, no. 9, September 2000, pp. 283-285. [14] R. V. Hogg and A. T. Craig, Introduction to Mathematical Statistics, 1995, Prentice-Hall, Inc. [15] P. Gupta and P. R. Kumar, “The Capacity of Wireless Networks,” IEEE Trans. Inform. Theory, March 2000, pp. 388-404. [5]