Performance Evaluation of Multiple Access Protocols for ... - CiteSeerX

Performance Evaluation of Multiple Access Protocols for Ad hoc Networks Using Directional Antennas Tamer ElBatt, Timothy Anderson, Bo Ryu Information Sciences Laboratory HRL Laboratories, LLC 3011 Malibu Canyon Rd, Malibu, CA 90265, USA {telbatt,cellotim,ryu}@wins.hrl.com Abstract— In this paper, we introduce a novel reservationbased multiple access protocol for ad hoc networks using directional antennas. First, we investigate the limitations of the extreme reservation schemes, namely omni-directional and directional reservations. We highlight the trade-off between spatial reuse (favors directional reservation) and control/data packet collisions (favors omni-directional reservation). Next, we show that the so-called hybrid reservation schemes fail to balance the trade-off as well. Therefore, we introduce a novel algorithm that balances the aforementioned trade-off via sending reservation messages, that carry information about the required direction of transmission, in all unblocked directions. In addition, we introduce candidate techniques for handling new types of collisions inherent to directional antennas. Finally, we conduct a simulation study that shows considerable performance gains of the proposed scheme over the omni-directional, directional, and hybrid reservation paradigms.

I. Introduction The broadcast nature of omni-directional antennas is one of the major causes of excessive multi-user interference, which limits the spatial reuse of the shared wireless medium. The distribution of energy in directions other than the direction of intended receiver not only causes considerable levels of interference, but also limits the transmission range. This constitutes a major hurdle towards efficient utilization of the precious resources of wireless systems, namely bandwidth and energy. Directional antennas has a fundamental impact on the design of CSMA/CA multiple access protocols. This can be seen from the answers to these questions: What information is necessary for a neighbor to decide whether to attempt reservation in a specific direction or not? Is it based on hearing/not hearing RTS/CTS messages as in the omni-directional case? We attempt to answer these questions in the paper. The cellular capacity improvement reported in [1], [2] is due to the following unique features of directional antennas. First, focusing the electromagnetic energy towards the intended receiver and minimizing it in the directions of unintended receivers. Second, focusing the energy in a specific direction increases the range of the transmitter compared to the case of radiating the same amount of energy in all directions. Several MAC protocols for ad hoc networks using directional antennas have been proposed in

the literature [3]-[10]. The work in [3] introduces a performance analysis study that shows the impact of directional antennas on the performance of slotted aloha. In [4], the authors introduce a multiple access protocol that minimizes interference by using a group of tones to identify active neighbors. The reservation schemes in [5]-[8] attempt to modify the RTS/CTS handshake mechanism to exploit the interference reduction feature of directional antennas. These algorithms utilize combinations of omni-directional and directional reservation messages, and hence are classified as hybrid schemes in this paper. However, they fail to balance the trade-off between spatial reuse and collision rate as discussed in section III. In [10], very narrow beam-widths are assumed. This assumption may not be feasible in practice due to frequency band and data rate constraints. Our main contribution in this paper is two fold: i) Balance the trade-off between omni-directional and directional reservations and ii) Resolve new type of collisions inherent to ad hoc networks using directional antennas. Unlike previous work, we believe that the focus should be shifted from sending reservation packets to a subset of neighbors (due to their geographical location) to sending reservation packets carrying ”directional antennas information” (i.e. information about the directional antennas used during reservation and data/ACK transmission phases) to as many neighbors as possible. This is essential to better assist them in knowing their locations relative to the transmitter-receiver pair and hence decide whether to transmit on a specific beam or not. The paper is organized as follows: in section II we introduce the system model. Next, we illustrate the tradeoff via showing the limitations of the omni-directional, directional and hybrid reservation paradigms in section III. This is followed by a description of the proposed reservation scheme in section IV. Afterwards, simulation results are presented in section V. Finally, the conclusions are drawn in section VI.

II. System Model A. Antenna Model In this paper, we limit our attention to the class of switched beam antennas that consist of several highly directive, fixed, pre-defined beams, formed usually with arrays [11]. Considering more complex and expensive steerable antennas that support mobile tracking and direction of arrival estimation is out of the scope of this paper and is a subject of future research. We assume that each node is supported by a fixed number (B) of switched beams, each of width θ = 2π/B radians. We use an idealized model for transmission on a specific beam. We assume that the transmitted energy is distributed uniformly in a beam of width θ (thus we ignore the possibility of side lobe interference). In addition, we assume that nodes receive omni-directionally, i.e. signals received by all beams are combined via diversity combining techniques [11], in order to combat multi-path fading and scattering effects. We assume that beams are non-overlapping. Candidate techniques for relaxing this assumption are discussed in section V. B. Network Assumptions We consider an ad hoc network consisting of n nodes that communicate only via the wireless medium. In this study, we focus on the performance of MAC algorithms, hence, routing is not considered. Each node generates data packets of fixed length according to a Poisson arrival process with rate λ packets/sec. Each generated packet is intended for a single neighbor only according to a uniform distribution. Each node has a buffer for temporarily storing generated packets awaiting transmission. The size of each buffer is assumed to be arbitrarily large since our main focus in this paper is throughput rather than queuing delays. We assume that a node lying within the coverage of a neighbor’s directional data/ACK transmission, is not allowed to engage in any communications. This is essential to avoid the intolerable interference levels this node may suffer due to the omni-directional reception assumption. On the other hand, nodes lying outside this region may transmit on any directional beam except the one pointing towards the transmitter. Finally, we assume that each node has the following information via neighbor discovery (ND) schemes: i) The identities of all neighbors and ii) The identities of neighbors that lie within the coverage of each beam. Introducing ND algorithms that efficiently utilize the range extension feature of directional antennas is out of the scope of this paper and is a subject of future research. III. Channel Reservation Paradigms In this section, we motivate the work via illustrating the trade-off between omni-directional and directional reservations. In addition, we show that hybrid reservation schemes proposed earlier in the literature fail to balance

the trade-off. This is attributed to the fact that all three reservation paradigms rely on the mere event of hearing/not hearing a reservation message, in order to decide whether to proceed with a transmission in a specific direction or not. Although this criterion for deciding channel occupancy is efficient for omni-directional antennas, it turns out to be problematic for directional antennas. In case of directional antennas, a node that does not hear a directional reservation message of another node may still be a neighbor of that node and could cause collisions. In the case of omni-directional antennas, a node that does not hear a reservation message of another node is not a neighbor and, hence, is not expected to cause any threat to the ongoing transmission. We denote the set of nodes within the geographical area covered by the radiation patterns of all beams at node x as neighbors N(x). This set can be partitioned into two subsets, namely blocked neighbors and unblocked neighbors, depending on their awareness of the ongoing reservation process carried out by node x. A neighbor is said to be blocked if and only if at least one of its beams is blocked from transmission due to overhearing a reservation message. On the other hand, a neighbor is said to be unblocked if it does not hear any reservation message, and hence, is completely unaware of the ongoing reservation attempt. Unblocked neighbors could initiate a reservation request independently and hence cause collisions to the ongoing transmission. We denote the sets of blocked and unblocked neighbors as BN(x) and UBN(x) respectively. The objective is to optimally partition N(x) into blocked and unblocked groups depending on: i) Their location with respect to the transmitter, ii) Their required direction of transmission and iii) The direction to be used by the transmitter for sending data or ACKs. Using this framework, we discuss the limitations of the three reservation paradigms. A. Omni-directional Reservation In the communication scenario shown in Figure 1, we consider the conventional CSMA/CA protocol where the reservation messages are exchanged omni-directionally (ORTS/O-CTS) between S and D. It is straightforward to notice that the neighbor sets of nodes S and D are given by N(S)={1,2,3,4,5,6,D} and N(D)={1,4,5,6,7,8,9,S} respectively. Furthermore, BN(S)=N(S), UBN(S)=φ, BN(D)=N(D) and UBN(D)=φ. This is due to the fact that the reservation packets are sent in all directions and hence all neighbors become aware of the ongoing transmission and refrain from transmission. Although omnidirectional reservations minimize the possibility of collisions, they turn out to be too conservative due to the possibility of blocking neighbors unnecessarily. B. Directional Reservation In this section, we examine the other reservation extreme where reservation messages are exchanged between

Fig. 1. Omni-directional Reservation

Fig. 2. Directional Reservation

S and D in a directional manner (D-RTS/D-CTS) [7], [8] as illustrated in Figure 2. In this case, we do not consider the range extension feature of directional antennas, i.e. the set of neighbors reached omni- directionally is assumed to be equal to the union of the sets of neighbors reached by each beam individually. This is mainly due to the assumption that omni-directional transmissions are achieved by transmitting on all beams simultaneously. In this case, it is straightforward to notice that BN(S)={4,5,6,D}, UBN(S)={1,2,3}, BN(D)={4,5,S}, UBN(D)={1,6,7,8,9}. Clearly, this scheme is more aggressive since it initiates more simultaneous reservation attempts, yet, they are unaware of each other and are highly subject to collisions. Thus, there is a fundamental trade-off, between collisions caused by unblocked neighbors and the number of unnecessarily blocked neighbors which limits spatial reuse. Thus, a neighbor knowing about an ongoing transmission may degrade performance (e.g. node 2 sending to node 10, or node 8 sending to 11), while another neighbor not knowing about an ongoing transmission could degrade performance as well (e.g. node 2 sending to node S, or node 6 sending to D). C. Hybrid Reservation The reservation schemes proposed in [5], [7] can be classified as hybrid schemes. The common aspect among those schemes is that they use different combinations of omni-

Fig. 3. Hybrid Reservation

directional and directional reservation messages. However, none of them attempt to balance the above tradeoff. In this paper, we consider the reservation scheme in [5] as an example of hybrid schemes. It utilizes DRTS and O-RTS packets, along with O-CTS packets. According to Figure 3, BN(S)={4,5,6,D}, UBN(S)={1,2,3}, BN(D)= {1,4,5,6,7,8,9,S}, UBN(D)=φ. Clearly, the proposed RTS solution does not strike a balance between minimizing control packet collisions and number of neighbors that back-off unnecessarily. Furthermore, sending omnidirectional reservation messages (whether O-RTS or OCTS) to all neighbors, without distinguishing neighbors that lie within the coverage of the directional data/ACK transmission, may lead to further collisions. Therefore, we introduce a new concept that is motivated by the following key observations: 1. All neighbors of the source and the destination should be aware of the intended transmission, if possible. 2. Antenna blocking decisions should be based on the information included in the RTS/CTS packets, not on the mere event of hearing/not hearing a reservation packet. IV. Channel Reservation based on Directional Antennas Information In this section, we attempt to balance the aforementioned trade-off in three steps. First, we propose to send the RTS/CTS packets over all unblocked beams, even of one or more beams are blocked. This incorporates some aggressiveness to the reservation scheme. In addition, it notifies all neighbors, covered by unblocked beams, of the ongoing reservation. Second, we propose to add two fields, to the RTS/CTS messages, that carry the index of the directional beam currently being used during the channel reservation phase and the index of the directional beam intended to be used during data/ACK transmission. This directional information is critical for each neighbor to know its relative location with respect to the source-destination pair and hence take appropriate antenna blocking decisions accordingly. The prime mo-

tivation for adding the previous fields is to decouple the event of hearing a reservation packet from the decision of beam blocking. The decision to transmit on a specific beam, or not, depends solely on the required direction and the information included in the reservation packets. According to the proposed scheme, node S sends different RTS packets on all unblocked beams to inform neighbors that it wishes to initiate a transmission with node D. Thus, in Figure 1, node 5 would know that it lies within the intended direction of data transmission from S to D and therefore should not engage in any communications in order to avoid interference caused by S. On the other hand, node 2 would know that it is out of the intended direction of data transmission from S to D, and hence it blocks only the beam pointing towards node S. Moreover, node 2 would know that it can initiate a transmission with node 10 since it would use a different beam from the one pointing towards S, and thus it will not cause any interference to S and will not suffer any interference from S. In this case, the proposed scheme decides that node 2 belongs to the set UBN(S). Otherwise, if node 2 wishes to transmit towards node S, it should belong to the set BN(S). Finally, the following new type of collisions inherent to directional antennas remains as a challenge to the proposed scheme: a neighbor may miss reservation messages due to lying in the coverage of a blocked beam. This neighbor, who may be active or inactive, would be left unaware of the attempted reservation and may cause collisions later. One way to circumvent this problem is to use auxiliary channel(s) to transmit special reservation packets on blocked beams. Under such proposition, blocked beams transmit special RTS/CTS packets, on a different frequency, whereas regular RTS/CTS packets are still transmitted on the unblocked beams at the same time. This scheme eliminates the above type of collisions caused by inactive neighbors lying in the coverage of the blocked beam. This is gained at the expense of using more than one channel, adding more complexity to the radio transceivers, and the possibility of having relatively rare collisions among the special reservation packets on the auxiliary channel. However, the threat of suffering collisions from active users, after completing their ongoing transmission, still prevails. Therefore, we propose to send ”pending” RTS/CTS packets once the blocked beam becomes unblocked. The transmission from source to destination could be temporarily halted, such that the source can transmit the pending packet on the beam recently unblocked. Once the packet is sent, the source node resumes its original data transmission on the beam pointing towards the destination. This approach is only feasible due to the short time needed to send a control packet, which can be tolerated by the longer data packet transmission. However, it involves synchronization complexity due to the possibility of having multiple pending reservation packets. Another variation of this approach exploits the underlying radio’s ability to transmit different

packets on multiple beams at the same time. Therefore, the pending packet can be sent on the recently unblocked beam while at the same time continue sending data on the beam pointing towards the destination. In this paper, we employ the auxiliary channel technique for handling inactive neighbors and pending packets for handling active neighbors. Clearly, this problem involves a trade-off between protocol complexity and control overhead in one hand and throughput on the other hand. Finally, a third approach, that is undergoing research, involves adapting the aggressiveness of the proposed reservation scheme depending on the number of blocked beams per node. V. Results and Discussion In this section, we present performance results obtained using the ns-2 simulator. The results show the performance gains of the proposed scheme over the omni- directional, directional, and hybrid reservation schemes under a wide variety of network loads. In addition, we show the impact of varying the number of beams on performance. A. Simulation Setup In this paper, we consider small networks consisting of n = 50 nodes since it sufficiently captures the trade-off under investigation. We consider a rectangular area of dimensions 500 meters x 500 meters where the nodes are uniformly distributed in the area. Data packet size is assumed to be 500 bytes. In this study, we do not consider nodes’ mobility since the arguments made in this paper are independent of mobility. We believe that the addressed trade-off remains valid under mobility conditions. Throughout the simulations, omni-directional transmissions are achieved via using all directional beams. Therefore, the range of omni-directional transmissions (250 meters) is assumed to be equal to the range of directional transmissions. We assume B = 6 switched beams per node, each of 60o width. In order to simplify simulations, we abstract any aspect of ad hoc networks that is irrelevant to the main focus of the study. Thus, we do not simulate specific ND algorithms, we compute its output based on the geographical location of neighbors instead. Each simulation run is carried out for the duration of 900 sec. B. Simulation Results The performance metric used to compare the performance of the proposed algorithm to the aforementioned reservation paradigms is the multiple access throughput. It is defined as the long-run average number of bits that reach their respective neighbors successfully per second. This ratio is to be computed at the end of each simulation run. First, we compare the long-run average number of data packets transmitted per second under the four schemes. The importance of this experiment stems from the fact that this parameter reflects the ”aggressiveness” of the algorithms. It can be easily noticed from Figure 4

Next, we compare the long-run average number of data collisions per second under the four schemes as shown in Figure 5. This result reflects the price paid for the aggressiveness. It is straightforward to notice that aggressive algorithms (directional reservation) suffer more collisions than conservative algorithms (plain CSMA/CA). Again, the number of data collisions caused by the other two algorithms lie in between. Moreover, the proposed algorithm suffers more collision than the hybrid scheme due to its aggressiveness. However, the excessive number of collisions is compensated by the significantly higher number of data packets transmitted as shown later. The interaction between these two conflicting requirements is what determines the net MAC throughput. Figure 6 shows the MAC throughput achieved by the four schemes under investigation. First, we notice that despite the aggressiveness of directional reservation, it outperforms the omnidirectional scheme by a factor of approximately 22% at heavy loads. This result implies that aggressiveness is a desirable feature of MAC protocols for directional antennas. Second, the hybrid scheme is noticed to achieve an intermediate performance since it is a combination of both schemes. Third, our algorithm gives the best performance since it guarantees low levels of data collisions (less than half the data collisions caused by the directional reservation scheme), while being relatively aggressive (almost twice the number of data packets transmitted by the omni-directional scheme). It is worth noting that using an auxiliary channel was the main contributor to reducing collisions and, hence, improving performance. Transmitting pending reservation packets, as described in the previous section, had negligible impact on performance. This is attributed to the fact that the ”auxiliary channel” protects against multiple inactive neighbors which could cause collisions at any time. On the other hand, ”pending packets” protect against small number of active neighbors (typically one) which may cause collisions only when it completes the ongoing transmission. Finally, we notice

2000 Proposed Scheme Omni−directional Reserv. Directional Reserv. Hybrid Reserv

1800

Average Number of Data Packets Transmitted/sec

that classical CSMA/CA is the most conservative since it has least attempts of sending data packets. This is a direct consequence of unnecessarily blocking neighbors from transmission. On the contrary, the directional reservation scheme is the most aggressive since it sends reservation packets in the intended direction only. This, in turn, leads to initiating excessive number of data transmissions that are subject to frequent collisions with reservation packets transmitted by unaware neighbors. The other two algorithms lie in between. Our algorithm is more aggressive than the hybrid algorithm proposed in [5]. This is attributed to the fact that the proposed algorithm sends reservation packets over all unblocked beams, even if one or more beams are blocked. In addition, it provides the sufficient information for each neighbor to take the appropriate antenna blocking decision. This, in turn, enables our algorithm to achieve a balance between the two extremes.

1600

1400

1200

1000

800

600

400

200

0

0

10

20

30 40 50 60 70 Packet Arrival Rate/Node λ (packets/sec)

80

90

100

Fig. 4. Average Number of Data Packets Transmitted/sec (n = 50, B = 6, Av. number of neighbors = 13.84)

that as the offered load increases the MAC throughput increases for all reservation schemes. The proposed scheme outperforms the omni-directional scheme by a factor of 64% at heavy loads and decreases to about 40% at moderate loads. Furthermore, it outperforms the directional scheme by a factor of 33% at heavy loads and 25% at moderate loads. Finally, it outperforms the hybrid scheme by a factor of 52% at heavy loads and decreases to 35% at moderate loads. Thus, we conclude that the proposed reservation scheme achieves considerable performance improvement over a wide range of network loads. At light loads, the four schemes achieve almost the same performance due to the low number of simultaneous transmissions attempted, on the average, which rarely causes collisions. Finally, the MAC throughput is plotted versus the number of beams per node (B), as shown in Figure 7. In this experiment, the parameter B takes the values 1,2,4,6,8, where the beam-width is given by 2π/B radians. The significance of this experiment is to compare various reservation schemes as the directivity of the beams increases. First, note that the MAC throughput of plain CSMA/CA remains constant for all values of B since it is designed for omni-directional antennas. Second, the MAC throughput of the other three algorithms increase, as the number of beams increases, due to higher interference reduction. Furthermore, the proposed algorithm is shown to outperform the other two schemes for all values of B. Finally, we note that the performance improvement becomes negligible for more than B=6 beams/node which suggests that the performance gain achieved by highly directional beams could be outweighed by the design complexity and high cost associated with developing such antennas. In this paper, we adopted the assumption of nonoverlapping beams. However, real switched beam antennas experience some overlapping due to coverage constraints. This, in turn, may degrade the performance of the proposed reservation scheme since it involves transmitting different packets over multiple beams at the same

the required direction of transmission in the reservation packets sent in all unblocked directions. The rationale behind this approach is to notify the maximum number of neighbors of the ongoing reservation and, assist them in taking the best decision of whether to proceed or refrain from transmission. Third, we proposed candidate solutions for combating new types of collisions inherent to directional antennas. Finally, we conducted a simulation study that shows considerable performance gains over three reservation paradigms introduced in the literature.


Average Number of Data Packet Collisions/sec

800

700

600

500

400

300

200

100

3400 0

0

10

20


80

90

Proposed Scheme Omni−directional Reserv. Directional Reserv. Hybrid Reserv

100

3200

Fig. 5. Average Number of Data Collisions/sec (n = 50, B = 6, Av. number of neighbors = 13.84) MAC Throughput (Kbps)

time. This problem can be solved using one (or more) of the following techniques: i) Careful antenna design that minimizes beam overlap subject to a constraint on the beam coverage, ii) Physical layer algorithms that ”captures” the strongest signal from interference in the overlapped areas and iii) Carrying out reservations in a roundrobin fashion over the course of K phases, where K may take values between 2 and B, since beam overlap decreases significantly between non-neighboring beams.

3000

2800

2600

2400

2200

2000

1800

1

2

3

4 5 Number of Beams/Node (B)

6

7

8

Fig. 7. MAC Throughput (λ = 100 packets/sec, n = 50 and Av. number of neighbors = 13.84)


3500

References [1]

MAC Throughput (Kbps)

3000

2500

2000

1500

1000

500

0

0

10

20


80

90

100

Fig. 6. MAC Throughput (n = 50, B = 6, Av. number of neighbors = 13.84)

VI. Conclusions In this paper, we introduce a novel CSMA/CA-based multiple access protocol that improves the throughput of wireless ad hoc networks using switched beam antennas. First, We showed the limitations of omni-directional, directional, and hybrid reservation schemes proposed in the literature. Our contribution in this paper is three fold. First, incorporating aggressiveness in the reservation scheme. Second, modifying the channel occupancy criterion to depend on information carried by the RTS/CTS messages, rather than their mere reception as in IEEE 802.11. Thus, we proposed to include information about

J. Ward and R. Compton, ”Improving the Performance of a Slotted ALOHA Packet Radio Network with an Adaptive Array,” IEEE Transactions on Communications, vol. 40, no. 2, pp. 292-300, 1992. [2] F. Shad, T. Todd, V. Kezys and J. Litva, ”Dynamic Slot Allocation (DSA) in Indoor SDMA/TDMA Using a Smart Antenna Basestation,” IEEE/ACM Transactions on Networking, vol. 9, no. 1, pp. 69-81, 2001. [3] J. Zander, ”Slotted ALOHA multihop packet radio networks with directional antennas,” IEEE Electronics Letters, vol. 26, no. 25, 1990. [4] T-S Yum and K-W Hung, ”Design algorithms for multihop packet radio networks with multiple directional antenna systems,” IEEE Transactions on Communications, vol. 40, no. 11, pp. 1716-1724, 1992. [5] Y. Ko, V. Shankarkumar and N. Vaidya, ”Medium Access Control Protocols Using Directional Antennas in Ad Hoc Networks,” IEEE INFOCOM, April 2000. [6] A. Nasipuri, S. Ye, J. You and R. Hiromoto, ”A MAC Protocol for Mobile Ad Hoc Networks Using Directional Antennas,” IEEE WCNC, 2000. [7] M. Sanchez, T. Giles and J. Zander, ”CSMA/CA with Beam Forming Antennas in Multi-hop Packet Radio,” Swedish Workshop on Wireless Ad hoc Networks, March 2001. [8] R. Ramanathan, ”On the Performance of Ad Hoc Networks with Beamforming Antennas,” ACM Mobihoc, Oct. 2001. [9] M. Takai, J. Martin, R. Bagrodia, and A. Ren, ”Directional Virtual Carrier Sensing for Directional Antennas in Mobile Ad Hoc Networks,” ACM Mobihoc, June 2002. [10] C. Balanis, J. Aberle, J. Capone, T. Duman, S. El-Ghazaly, A. Spanias and T. Thornton, ”Smart Antennas for Future Re-configurable Wireless Communication Networks,” Arizona State University Technical Report, April 2000. [11] R. Janaswamy, ”Radiowave Propagation and Smart Antennas for Wireless Communications”, Kluwer Academic Publishers, Boston, 2001.