Enhancing Wireless Networks with Directional Antenna and Multiple Receivers

Transcription

1 Enhancing Wireless Networks with Directional Antenna and Multiple Receivers Chenxi Zhu Fujitsu Labs of America 8400 Baltimore Ave., Suite 302 College Park, Maryland Tamer Nadeem Siemens Corporate Research 755 College Road East Princeton, NJ Jonathan Agre Fujitsu Labs of America 8400 Baltimore Ave., Suite 302 College Park, Maryland Abstract When directional antennas are used in based wireless LANs, higher network capacity is often accompanied with more collisions. This is due to the enhanced hidden node problem and the deafness problem that arise in the directional transmission/reception scenario. These problems are caused by an inconsistent view of the medium status by neighboring nodes. We have developed a new variation to the protocol called sectorized MAC (S-MAC), which employs a novel architecture consisting of multiple directional antennas and multiple receivers. With a new self-interference cancellation scheme, an S-MAC node can continuously monitor the channel status in all directions in order to avoid the aforementioned problems. It is fully compatible with the standard protocol and can inter-operate with omni-antenna-based nodes. It is applicable to both infrastructure mode and ad hoc mode. Simulation studies show that it achieves significant capacity gains, even if only used in part of the network. Keywords: wireless LAN, IEEE , directional antenna, OFDM. I. INTRODUCTION Wireless Local Area Networks (WLANs) based on the IEEE standard family [1], including a/b/g, have enjoyed tremendous popularity in recent years. A consequence of increasing WLAN deployment, coupled with limited number of channels and unlicensed spectrum usage, is that the interference between transmissions is becoming a serious problem. Directional antenna is a well known method to reduce the interference and to increase the range and the capacity for wireless networks. In general, network capacity is increased as a consequence of spatial spectrum reuse, and transmission range is extended due to the increased gain from directional transmission. Use of directional (or sectorized) antenna in cellular networks is the norm. Directional antennas have also been proposed to improve the performance of based wireless networks [2], [3], [4], [5], [6], [7], [8], [9], [10], [11]. However, the original protocol was not designed for directional antennas. It assumes that all the packets (RTS/CTS/DATA/ACK packets) are transmitted as omni-directional signals that are received by all nearby nodes. When directional antennas are used, new problems such as the enhanced hidden node problem and the deafness problem, arise. They increase the likelihood of collisions. Although many schemes have been proposed to solve these problems [4], [7], [14], [15], a satisfying solution completely within the framework has not yet been reported. In this paper we propose a new scheme, called sectorized MAC (S-MAC), which uses multiple directional antennas and multiple receivers to provide 360 degree coverage around a node. The scheme allows a node transmitting in some sectors while receiving in others. It addresses the hidden node problem and the deafness problem by continuously monitoring the channel in all directions (sectors) at all time. Compared with schemes that require only software change, S-MAC addresses the collision problem with both software and hardware change. With decreasing cost of hardware and increasing demand for throughput, we believe the extra harware cost is justified. The organization of the paper is as follows. We start with a description of the problems when directional antennas are used with the protocol. A summary of related works in given in Section II. The design of S-MAC protocol is presented in Section III and its operation in Section IV. S-MAC consists of two parts: 1) a self-interference cancellation scheme at the PHY layer, and 2) use of the correct channel status information at the MAC layer. Examples are used to show how S-MAC prevents the hidden node problem and deafness problem. A brief performance evaluation of S-MAC is presented in Section V. We conclude the paper in Section VI.

2 II. BACKGROUND AND RELATED WORKS Previous works [2], [3], [6], [10] have studied the effect of transmitting different messages with directional mode versus omni-directional mode in networks. It is found that transmitting all packets using the directional mode gives the highest capacity [10]. Besides allowing spectrum reuse, directional antennas also provide range extension. It is argued that all messages should be sent in directional mode to take advantage of this gain [7]. However, there is a trade off between more aggressive spatial reuse and more collisions. As space is segmented and the carrier is reused with directional antennas, the amount of MAC information increases. The original Distributed Control Function (DCF) protocol does not provide this extra information, and different nodes may have a different understanding of the medium status. Collisions arise as a consequence. When all the packets are sent directionally, several problems occur. These include the hidden node problems due to asymmetry in gain and due to unheard RTS/CTS, and the deafness problem. These problems are analyzed in great detail in [4] and only briefly summarized here. If a node cannot hear the RTS/CTS exchange because its antenna gain is not large enough in a certain direction (e.g., it is pointing its antenna to another direction), it may misinterpret the channel status in that direction and cause interference to other transmissions. The well-known hidden node problem then becomes more prominent. Several solutions that change between omni and directional modes have been proposed [2], [5]. If a node switches between directional mode and omni mode when transmitting different messages, or between transmission and receiving, two types of asymmetry arise: asymmetry in range and asymmetry in direction. A node receiving in omni mode may miss a RTS or CTS packet out side of its omni range, but when it transmits using directional mode, its energy reaches further and may cause interferenc. Similarly, a node receiving from a certain direction is less sensitive to a packet arriving from other directions. It may miss a RTS or a CTS. But if it starts a transmission in the other direction based on its perception of the channel, it may collide with other transmissions. Physical channel sensing is also less accurate with directional antennas. Another problem caused by directional antennas is deafness. Deafness occurs when a node A is engaged in a transmission with a node B in one direction, but this transmission is not known by a node C whose antenna is pointing at A from another direction. Node C tries to contact node A by repeatedly sending un-replied RTS packets. These unproductive RTS packets waste the network resources and lead to unnecessary back off and unfairness. If node C determines the link is lost because of this, it will also cause problems at higher layers including routing. Some schemes use out-of-band signals to perform carrier sensing and to prevent deafness [13], [14]. However, this makes them non-compatible with the family. III. ARCHITECTURE OF THE S-MAC PROTOCOL The S-MAC protocol applies to the distributed control function (DCF mode) of Because the hidden node problem and the deafness problem are caused by lack of correct channel state information in every direction at all time, to provide this information is the motivation of the S-MAC protocol. We assume that the coverage area of an S-MAC node is divided into M sectors with fixed wide beam directional antenna i covering sector i. The scheme is called sectorized MAC because of its resemblance to the configuration of a celluar base station with multiple sectors. The coverage areas of these antennas can overlap in their side lobe or back lobes, i.e. there is no limit regarding the antenna patterns. Each S-MAC node is equipped with one or more (at most M) transmission modules (T x) and M receiving modules (Rx). For simplicity the discussions below assume single T x. The extension to multiple Tx modules is straightforward. The T x and Rx modules are connected to a single MAC module. The T x module can switch to any one of the antennas and transmit in the corresponding sector. Rx i connects to antenna i and receives only from sector i. All the transmissions and receptions take place through directional antennas, i.e. there is no omni transmission or receiving mode. At the PHY layer, the M Rx modules are logically separate. When the T x module is not transmitting, all the M Rx modules are listening from their respective antennas. We further require that when the T x is transmitting in a sector (sector 1 for example), the remaining M 1 Rx modules continue to listen in their sectors. This requires the node to cancel the interference in Rx 2 through Rx M caused by its own T x. In sector 1 where the active transmission takes place, T x and Rx 1 still operate in TDD mode. Figure 1 shows the architecture of a S-MAC node with 2 T x and 3 Rx modules. Details of the selfinterference cancellation schemes are described below. A. Self-interference cancellation scheme Self-interference cancellation is done at the PHY layer. S-MAC requires Rx i to continue receiving cor- 2

3 Fig. 1. DUX Directional Antennas DUX DUX switching fabric RF RX RF TX RF RX 1 TX 1 RX 2 RX 3 TX symbol for self-interference cancellation TX 2 Base Band RX TX Separate queues S-MAC: SNAV=[NAV TX1,NAV TX2, NAV RX1, NAV RX2, NAV RX3 ] MAC and LLC Architecture of S-MAC node with 2 Tx and 3 Rx modules. rectly while T x is transmitting in another sector j. This can be done if Rx i cancels the interference caused by T x. Because Rx i and T x modules are part of the same PHY, Rx i knows the transmitted symbol. Therefore Rx i only needs to estimates the channel gain G ij between antennas i and j to cancel the Tx signal. Note that channel estimation is an integral part of all PHY layers (a,b,g). Every packet starts with a PLCP preamble (SYNC sequence) which is used for channel estimation by the receiver. A similar SYNC sequence is used here. Channel estimation from one antenna to another is called self-calibration because both the transmitter and the receiver are part of the same node. To estimate the channel properly, an S-MAC node needs to make sure that its self-calibration signal does not collide other signals. Below we introduce two schemes, both ensuring interference-free self-calibration. One works at the MAC layer and applies to all types of PHY, the other works at the PHY layer and applies to OFDM-based PHY layers like a/g. 1) MAC assisted self-calibration: An S-MAC node follows the normal carrier sensing/back off procedure before it starts self-calibration. It then transmits RTS packets circularly in every sector to silence its neighbors before it transmits the training symbols (SYNC) circularly in all sectors for channel estimation between pairs of antennas. As the RTS is sent in the M sectors one by one, the advertised duration field of the RTS packets may decrease gradually, but it needs to be sufficient for the neighbors to back off for the entire self-calibration period. After the circular RTS packets, the SYNC symbol is transmitted circularly. While it is sent in sector i, every Rx j, j i, in other sectors will estimate the channel gain G ij. Considering the channel symmetry G ij = G ji, one can average G ij and G ji to reduce the estimation error, or to send SYNC only in half of the sectors to reduce the overhead. How often this procedure is invoked depends on the dynamics of the radio propagation environment. Because the channel reservation is done at the MAC layer, it is applicable to all PHY layers. 2) Coded pilot assisted self-calibration for OFDM: OFDM is used in a for signaling. In the coded pilot assisted self-calibration scheme, the medium is not acquired first at the MAC layer. Therefore selfcalibration is vulnerable to interference from other transmissions. The coded pilot tones in the training symbols detects possible collisions at the PHY layer, and garbled channel estimations are discarded. Upon collision detection the self-calibration procedure can be repeated after the node determines the channel is available at a later time. The key to collision detection using the pilot is to generate a binary random sequence X of length 52 and to map this sequence to the 52 sub-carriers in a (excluding the DC tone) for channel estimation and collision detection. Let X = [x 26,..., x 1, 0, x 1,..., x 26 ], x i {0, 1}. Every 0 in the sequence is mapped to a null pilot (no signal transmitted), and every 1 in the sequence is mapped to a pilot tone and modulated with a BPSK symbol. The modulated SYNC word L is derived from the long training symbol of the PLCP preamble (SYNC) L in the a [1]: L = {l 26,..., l 1, 0, l 1,..., l 26 }, l i {1, 1}, and L = {l 26,..., l 1, 0, l 1,..., l 26 }, where l i = x i l i, 26 i 26. In other words, the long OFDM training symbol L of SYNC in a is punctured in the frequency domain by null tones whose positions are determined by the random sequence X. The coded SYNC word L is used for both channel estimation and collision detection. Additional long training symbols can be added to the SYNC sequence for better estimation. After following the standard channel sensing and backoff, the coded SYNC word L is sent through antenna i. Every receiver in other sectors Rx j, j i, will estimate the channel gain G ij using the pilot tones and detect possible collisions using the null tones. Therefore clear channel assessment (CCA) is carried out in the null tones at the same time as channel estimation. If collision is detected in the null tones, the channel estimation between antennas i and j is invalid. If no collision is detected, channel estimation between antenna i and j is obtained by interpolating between the pilot tones. Using 3

4 the randomly placed null tones makes it easy to detect collisions with self-calibration signals sent by another S- MAC node. A node can use its MAC address and current time as a seed to generate this pseudo-random sequence. This reduces the probability that two nodes choose the same sequence at the same time. Similarly to the MAC controlled self-calibration scheme, one can send SYNC frames in only half of the sectors. The M (or M 2 ) SYNC words can be sent contiguously or separately. It is possible that when a SYNC is sent, collisions take place only at some of the receivers. In this case the self-calibration process needs to be repeated only to fill in the parts that suffer from the collision. Without the MAC overhead, the channel usage for self-calibration is very light. With a the time required to send SYNC is only 16 us. So the overall overhead is very light, even considering repeated trials due to collisions. How often self-calibration is required depends on how often the environment (such as nearby reflectors) changes. In simulation below S-MAC is used at fixed access points in infrastructure mode, where the radio propagation condition does not change dramatically. With most WLANs used in static networks, self-calibration only needs to be executed with ralatively low rate and contributes little to channel congestion. B. Sectorized network allocation vectors and carrier Sensing S-MAC maintains and updates a network allocation vector NAV i for each sector i following the standard DCF procedure. This is similar to D-NAV proposed in some other schemes [5, 6]. The N AV table associated with the M sectors has the following structure: SNAV = [NAV T x, NAV 1, NAV 2,..., NAV M ], where NAV T x describes the allocation (busy time) of the Tx module, and NAV i describes the allocation of the channel in sector i. Note that detailed antenna pattern like direction and beam width information is absent. This is because the protocol is agnostic to the detailed beam pattern information. SN AV table is set as follows: NAV T x is updated when the S-MAC node is directly involved in a transmission, disregarding whether the transmission is initiated by which parties. In this case the length of NAV T x is set for the entire duration of the four way (or two way) message exchange. NAV i is updated following the standard DCF procedure, i.e. it is updated both for a transmission involving this node itself in sector i, and for a transmission between other pair of nodes which at least one of them is in sector i (Rx i receives RTS/CTS/DATA frames). Both NAV T x and NAV i are used to determine if the medium in sector i is available. Besides virtual carrier sensing, an S-MAC node needs to carry out physical carrier sensing (clear channel assessment) according to the standard [1]. In S-MAC physical carrier sensing is done on a per-sector basis with the respective Rx module. S-MAC can use the medium in sector i only if the physical medium is sensed clear by Rx i. IV. OPERATION OF S-MAC S-MAC protocol lends itself to both ad hoc mode and infrastructure mode, and can readily inter-operate with nodes with omni antennas. The reason is that from the view point of a nearby node, an S-MAC node behaves like a regular node so its S-MAC identity is transparent to the other node from protocol point of view. In infrastructure mode, most communications are between access points (APs) and stations (STAs), and sometimes between two STAs if the direct link protocol is used. It is more economic and more feasible to use directional antennas and S-MAC at the AP, while keeping the STAs with omni antenna. This makes every transmission between a STA and the AP a hybrid link. Capacity can be increased by only upgrading the APs. In ad hoc mode, S-MAC can be used in some or all of the nodes. Below we describe how S-MAC works in the different scenarios. A. S-MAC Operation and inter-operability S-MAC node initiating a transmission. After the S-MAC node determines that the channel in sector i is free with physical and virtual carrier sensing (checking NAV i and NAV T x ), it may initiate a transmission to another node in sector i by sending a RTS through antenna i. It sets NAV T x and NAV i accordingly. If it receives a CTS packet successfully with Rx i, the DATA/ACK exchange will follow through antenna i. If no CTS is received, it will reset NAV T x and NAV i. In the case that S-MAC can transmit to its destination through more than one antennas, S-MAC chooses the antenna with the best link quality. For each neighbor A, S-MAC maintains a location/link quality vector L A = {(i, qi A), (j, qa j ),...} where each pair (i, qi A ) means A can be reached through antenna i with link quality qi A (measured by received signal strength). Vector L A can be built and updated based on the signal received from node 4

5 A. Once deciding a sector, S-MAC will use that antenna for the entire four way handshake. S-MAC node responding to a RTS request. Suppose a four way handshake is initiated by another node to an S-MAC node, and the RTS packet is received by Rx i at the S-MAC node. As Rx i presents this RTS to the MAC, NAV i and NAV T x are checked. If both of them (and physical carrier sensing) are free, a CTS packet will be sent by T x through antenna i, and NAV i and NAV T x are updated with the duration of the transmission field in the RTS. Other nodes (omni or S-MAC) receiving the RTS or CTS will update their NAVs and respect this channel reservation. Having secured the channel, the node initiating the handshake starts transmitting the DATA packet and the S-MAC node will receive with Rx i. After successfully receiving the DATA packet, the S-MAC node sends its ACK through sector i. On the other hand, if the S-MAC node finds that either NAV i or NAV T x is busy, it will not respond to the RTS. Its SNAV remains unchanged. In this case the node sending the RTS will back off. If the RTS is received successfully by more than one Rx modules, say Rx i and Rx j, and the channel is available in all these sectors, S-MAC will send the CTS in sector i if it finds the best receiving quality in Rx i. This may happen if the other node is in the overlapping area of sector i and j. As the CTS is sent in sector i, NAV i, NAV j and NAV T x are all updated. The location/link quality vector L A for this node can be updated as well. It is possible that the S-MAC node receives more than one RTS packets from different neighbors successfully in different sectors. It can choose among the contenders based on channel availability in these sectors and link quality. It then sends CTS to the selected target through the corresponding antenna and ignores the rest. S-MAC node in the presence of transmission between other nodes In this case either the transmitter or the receiver (or both) of the communicating pair is one hop away from S-MAC. If Rx i of an S-MAC node receives a RTS or CTS or DATA packet transmitted between other pair of nodes, S-MAC sets its NAV i accordingly. It is possible that the S-MAC node receives the same RTS or CTS with more than one of its receivers. All the NAV in these sectors are set. It does not need to tell whether the source or the destination node of the transmission is omni or S-MAC. Overlapping beams of a S-MAC node. It is natural that an S-MAC node has overlapping sectors due to side or back lobes of antenna beam patterns. If a node is in the overlapping part of multiple sectors, all these sectors receive its RTS/CTS/DATA packet(s), and the NAV of all these sectors are set. S-MAC node does not need to correlate NAV component of different sectors to tell if they are set because of the same node or not. S-MAC to S-MAC transmission. From the view point of an S-MAC node, it cannot (and does not need to) tell whether the other party is an S-MAC node or a regular omni-directional node. The operation of the sender S-MAC node, and operation of the target S-MAC node, follow the same procedures as described previously. Regular (omni-directional) node. The directional operation of S-MAC is transparent to a regular omni node (or another S-MAC node). A regular (or S-MAC) node can operate with the belief that all nodes are omni and operate with the standard MAC. B. The hidden node problem and the deafness problem Note that with the S-MAC protocol, a node always uses its directional antennas for transmission or receiving. This eliminates the hidden node problem due to the difference in gain in terms of transmission range. This also allows the directional gain of the antenna to be fully explored. With the self-interference cancellation scheme, S-MAC continues to monitor the channel in other sectors while it is engaged in a transmission in a different sector. It will not miss a RTS/CTS/DATA packet. Consequently its SNAV always reflects the channel status in different sectors correctly and no hidden node problem will arise due to unheard RTS/CTS. So far no discussion has been given to the deafness problem. If an S-MAC node A is engaged in a transmission with a node B (S-MAC or omni) in a sector, its MAC information is not known to another node C (S-MAC or omni) in a different sector. This node C may believe node A is available based on its own MAC information and sends RTS to A. This is the same as the D-MAC protocol [5]. But there is a key difference between the two. With D-MAC, when C tries to contact A, C does not know that A is engaged in a transmission in another direction, and A cannot tell C is trying to contact it. With S-MAC, multiple Rx and self- 5

6 Fig. 2. An example scenario (adopted from [4]). 400m 250m AP 2 AP 0 AP 3 AP 1 Overlapping transmission regions interference cancellation provide node A with additional information to resolve this problem. Suppose that the RTS from C is received by Rx i at A. Node A may send a RTS to C after its transmission with B. This way C learns that A is still alive. Alternatively, if A decides the request from C is very important (for example, C has sent un-responded RTS packets a number of times), A may choose to drop its ongoing transmission with B and respond to C with a CTS packet. S-MAC avoids the deafness problem by continuously monitoring RTS packets in all sectors and responding properly. Note that when the RTS from C is not responded, the medium in sector i is not occupied for an unnecessarily long time (the duration of 4 way handshake advertised in RTS). The channel becomes clear to the other nodes after they do not hear the DATA packet after RTS and reset their NAVs. C. Examples We use same scenario as [4] (Figure 2) to show how S-MAC prevents collisions that would otherwise occur in DMAC. Suppose all the nodes have sector antennas and use S-MAC. Assume node B transmits a RTS to node F using its antenna pointing to F, and F responds with a CTS sent in the direction of B. This CTS reaches node E and A in their antenna pointing in the direction to F, so nodes E and A set in their SNAV for the corresponding sector. If A wants to transmit to E, by checking its SNAV it finds this sector is temporarily occupied so it will hold its packet to E. This way A does not interfere with the transmission between B and F. Assume E is transmitting to D. During this transmission B sends a RTS to F, and F responds with a CTS. The CTS from F reaches E in its sector point to F, so E sets its corresponding SNAV. When E finishes its transmission to D and wants to start a transmission to F, it has to wait till the B F transmission is complete. Assume node E is transmitting to B. As node C is not in the direction of E B, C is unaware of this transmission. If C sends a RTS to E, E will receive this Fig m Non-overlapping transmission regions A network of 4 cells. Each AP has 4 sectors and 40 STAs. RTS E by the sector pointing to C. After E receives a number of RTS from C, E can drop a transmission to B and replies a CTS to C. Deafness is thus prevented. V. PERFORMANCE EVALUATION We now present simulation-based studies on the performance of the S-MAC protocol. The simulations are done using the ns-2 simulator [16]. The transmission rate is 11 Mbps with transmission range of 250m, and the carrier sensing range is 550m. The coded pilot assisted self-calibration scheme is used. Due to space limit, only results from infrastructure mode is presented here. The network consists of several access points (APs), each of which is associated with its set of clients or stations (STA). The simulated scenario represents the case that in the network already deployed, the APs are upgraded with sector antennas from omni antennas, while omni antennas and the original MAC protocol are used at all the stations. Because the coverage area of each AP is fixed, we reduce the transmission power at the APs with directional antennas so an AP has the same transmission range before and after the upgrade. With S-MAC, each AP still has a single Tx. All the nodes are stationary. Network throughput is compared before and after the APs are upgraded S-MAC. The reason that that S-MAC is not compared with D-MAC is that the latter is designed for ad hoc networks with all nodes employing beamforming antennas and does not handle this scenario. A fair comparison between S-MAC and D-MAC will be presented in the future. Network traffic consists of UDP packets sent between the STAs to the APs. The average packet size is 1000 bytes. Four way handshake (RTS/CTS/DATA/ACK) is used. Each case is simulated for a duration of 50 seconds and each point shown is averaged from 5 independent runs. Figure 3 shows a network of 2x2 grid of overlapping 6

7 deafness problem with dedicated receiver in each sector and a new self-interference cancellation scheme between sectors. It is fully compatible with the standard protocol using omni antenna. It can be used to upgrade the access point in the infrastructure mode to enhance the network capacity, or used in some or all of the nodes in the ad hoc mode. Capacity gain over the omni MAC protocol is significant. The downside of S-MAC is that it requires more sophisticated hardware. However we believe with the cost of silicon going down and the need for capacity increasing, this tradeoff can be justified. Fig. 4. Total throughput for 2x2 grid APs under different traffic loads and S-MAC improvement. APs. Each AP has 4 sectors and 40 clients that are distributed uniformly in its coverage area. We conducted simulations under different network loads. Each STA sends packets to its AP at rates varying from 5 to 200 packets per second. Figure 4 shows the network throughput for both omni and S-MAC AP configurations with the enhancement percentage. The network capacity is enhanced by 110% when S-MAC is used at the APs. This demonstrates the benefits of employing directional antennas, even only at the network infrastructure side. Note that the number of Tx at the APs remains the same (1), so the capacity improvement does not arise from more transmitters, but from reduced interference with sector antennas and reduced collision from the S-MAC protocol. We also experimented with different number of sectors at the APs. The simulation results showed that the total throughput does not change significantly as the number of sectors increases from 2 to 4. Similarly no significant change was found as different antenna orientations were used. We believe this is partially due to the fact that the omni directional antennas at the STAs determine the interference between different transmissions (and limits the degree of spatial reuse), and partially due to the fact that each S-MAC AP has single T x. We also conducted simulations for different scenarios, including larger networks in infra-structure mode and ad hoc mode. The results showed that S-MAC can improve the network performance significantly, both in terms of transmission throughput and delay. Due to limited space, these results and a comparison of S-MAC with DMAC, will be presented in the future. VI. CONCLUSION We have developed a new scheme for fully exploring the potential of directional antennas with the protocol. It avoids the hidden nodes problem and the REFERENCES [1] IEEE Std Edition Wireless LAN Media Access Control (MAC) and Physical Layer (PHY) Specifications. [2] YB Ko, V Shankarkumar, NH Vaidya, Medium access control protocols using directional antennas in ad hoc networks, Proc. Of IEEE INFOCOM [3] A Nasipuri, S Ye, J You, RE Hiromoto, A MAC Protocol for Mobile Ad Hoc Networks Using Directional Antennas, Proc. Of IEEE WCNC [4] RR Choudhury, X Yang, R Ramanathan, NH Vaidya, Using directional antennas for medium access control in ad hoc networks, Proc. Of ACM MobiCom [5] M Takai, J Martin, R Bagrodia, A Ren, Directional Virtual Carrier Sensing for Directional Antennas in Mobile Ad Hoc Networks, Proc. Of ACM MobiHoc [6] T. ElBatt, T. Anderson, B. Ryu, Performance Evaluation of Multiple Access Protocols for Ad Hoc Networks Using Directional Antennas, Proc. of IEEE WCNC, [7] T. Korakis, G. Jakllari, L. Tassiulas, A MAC protocol for full exploitation of Directional Antennas in Ad-hoc Wireless Networks, Proc. Of ACM MobiHoc, [8] Akis Spyropoulos and C.S. Raghavendra, Energy Efficient Communications in Ad Hoc Networks Using Directional Antennas, Proc. Of IEEE INFOCOM [9] R. Ramanathan, On the performance of ad hoc networks with beamforming antennas, Proc. of ACM MobiHoc, October [10] C. Srisathapornphat, C.-C. Shen, Energy Consumption Behavior and Performance of Directional Virtual Carrier Sensing Schemes, Proc. Of IEEE WCNC, [11] C. Li, J. Li and X. Cai, Performance Evaluation of Access Delay of Efficient Media Access Schemes for WLAN with Smart Antenna, IEEE ICC [12] R. Ramanathan, J. Redi, C. Santivanez, D. Wiggins, and S. Polit, Ad Hoc Netwroking With Directional Antennas: A complete System Solution, Proc. Of IEEE WCNC, [13] Z. Huang, C. Shen, C. Srisathapornphat, and C. Jaikaeo, A busy-tone based directional MAC protocol for ad hoc networks, in Proc. Of IEEE Milcom, [14] RR Choudhury, and NH Vaidya, Deafness: A MAC problem in ad hoc networks when using directional antenna, Proc. Of IEEE International Conference on Network Protocols (ICNP), [15] H. Gossain, C. Cordeiro, D. Calalcanti and D. P. Agrawal, The Deafness Problems and Solutions in Wireless Ad Hoc Networks using Directional Antenna, IEEE Globecom [16] The Network Simulator ns-2, 7