Medium Access Control Protocols. using Directional Antennas in Ad Hoc Networks. (Preliminary Version) Young-Bae Ko and Nitin H.

Transcription

1 Medium ccess ontrol Protocols using Directional ntennas in d Hoc Networks (Preliminary Version) Young-ae Ko and Nitin H. Vaidya Department of omputer Science Texas &M University ollege Station, TX fyoungbae, vaidyag@cs.tamu.edu Technical Report May 11, 1999 bstract Using directional antennas can be useful for mobile ad hoc networks consisting of a collection of wireless hosts. To best utilize directional antennas, a suitable Medium ccess ontrol (M) protocol must be designed. urrent M protocols, such as the IEEE standard, do not benet when using directional antennas, because these protocols have been designed for omnidirectional antennas. In this technical report, we attempt to design new M protocols suitable for mobile ad hoc networks based on directional antennas. 1 Introduction mobile ad hoc network (MNET) is an autonomous system of mobile nodes which are free to move arbitrarily and typically assumed to be equipped with omnidirectional antennas 1 [3]. However, it is also possible to use directional antennas [5] or adaptive antennas [9] to improve the ad hoc network capacity. In this report, we consider directional antennas, but the ideas can Research reported is supported in part by Texas dvanced Technology Program grants and National Science Foundation grants D n omnidirectional antenna transmits in all directions (360 degrees) 1

2 be applied to adaptive antennas as well. Using directional antennas may oer several interesting advantages for mobile ad hoc networks. For instance, routing performance could be improved using a directional antenna (for route discovery [7] or for data delivery). To best utilize directional antennas, a suitable Medium ccess ontrol (M) protocol must be used. urrent M protocols, such as the IEEE standard, do not benet when using directional antennas, because these protocols have been designed to exploit omnidirectional antennas. In this technical report, we propose new M protocols using directional antennas. 2 Related Work lthough work on M protocols for directional antennas is limited, some researchers have previously suggested use of directional antennas for a packet radio system. For example, Zander [13] has proposed the use of directional antennas in slotted LOH multihop packet radio networks whose broadcast radio channel is shared by means of some random time division multiple access (RTDM) scheme. More recently, a way of using adaptive directional antennas for the Mobile roadband System (MS) has been proposed [5]. In addition, [5] argues that conventional M protocols are not suitable with directional antennas and suggests a dynamic slot assignment (DS) protocol for directional antennas. Other researchers have also suggested using directional antennas for packet radio networks [8, 11, 12]. 3 Network Model We assume that all mobile hosts in a region share a common channel and communicate on that shared wireless channel. Each host is assumed to be equipped with multiple directional antennas for its transmitter and an omnidirectional antenna for its receiver 2. directional antenna can transmit over a smaller angle (e.g., 90 degrees), and several directional antennas may be used to cover all directions. Therefore, mobile hosts can receive from all directions simultaneously, and transmit in any chosen direction(s). We assume that communication between two hosts is bidirectional, although this assumption may not be always true in reality due to interference and obstacles [10]. Each host has a xed transmission range and two hosts are said to be neighbors if they can communicate with each other over a wireless link. Initially, we assume that each node knows its neighbors' location as well as its own location. t the end of this paper, we briey consider the case when location information is not known accurately. The physical location information may be obtained using the global positioning system (GPS) [1]. ased on location of a receiver, a sender may select an appropriate directional antenna to send packets to the receiver. Most of the current M protocols, such as IEEE M standard [2], use a handshaking mechanism implemented by exchanging small xed-size control packets: Requestto-Send (RTS) and lear-to-send (TS) packets. The successful exchange of these two control packets reserves the channel for guaranteeing undisturbed transmission of the longer data packet 2 n adaptive antenna could also be used instead. 2

3 and a short acknowledgement (K) packet. In our initial discussion, the K packets are not considered, i.e., the M protocols would not send Ks. In Section 5.2, we will consider how the K aects proposed protocols. 4 The urrent RTS/TS Mechanism Figure 1 illustrates a M protocol, similar to M [6] and the IEEE standard, for omnidirectional antennas that uses RTS and TS control messages. In this protocol, any node that wishes to transmit data must send a RTS packet before it can start data transmission. For example, in Figure 1, node broadcasts a RTS packet for its intended receiver, node (please see caption of Figure 1 for an explanation of the gure). If hears the RTS successfully, it replies with a TS packet so that can start transmitting data packets upon receiving the TS. Note that both RTS and TS packets contain the proposed duration of data transmission. Since nodes are assumed to transmit using omnidirectional antennas, all nodes within radio range of and will hear one or both of those control packets (nodes and D in Figure 1) { these nodes must wait for the duration of data transmission before they can transmit anything themselves. Thus, the area covered by the transmission range of sender (node ) and receiver (node ) both is reserved for the data transfer from to, to prevent collisions. This characteristic of RTS/TS mechanism overcomes the hidden terminal problems in wireless LN environments. However, it is easy to see that this mechanism can waste a large portion of the network capacity by reserving the wireless medium over large area. For instance, even though node D has data packets for node E while and are communicating with each other, node D has to defer the transmission to E until the transmission from node to completes. 5 Proposed Directional M Mechanisms 5.1 Directional M Protocol #1 To illustrate how M protocols for directional antennas may be designed, we consider a modication of the above protocol. The modied M protocol utilizes a directional antenna for sending the RTS packets in a particular direction, whereas TS packets are transmitted in all directions (either using multiple directional antennas, or using an omnidirectional antenna). Figure 2 and Figure 3 show how wireless bandwidth eciency of the previous M protocols in Section 3.1 can be improved by using a directional M protocol. In Figure 2, when node has data packets for node, it sends a directional RTS (DRTS) packet including the physical location information of, in the direction of node. Thus, node does not receive the DRTS from node even though node also exists within 's transmission range. If node receives the DRTS packet from successfully, it then returns an omnidirectional TS (OTS) reply. Two location informations are included in the OTS packet: one for its own location (for instance, node 's location in Figure 2) and the other for the sender of the corresponding DRTS packet (for instance, node 3

4 D E RTS RTS TS TS DT DT Figure 1: M protocols with omnidirectional RTS/TS mechanism: The circle centered at each node shows its transmission range. In the lower half of the gure, time progresses from top to bottom. The gure shows messages sent by various nodes. lack bars below nodes and D indicate that these nodes are not allowed to transmit in the duration covered by the bars (to avoid interference with the transfer from to ). 4

5 in Figure 2). fter the successful exchange of DRTS and OTS packets the actual data packet can be sent by node using a directional antenna. Now, during the proposed length of transmission between and, node D, which is a neighbor of node, wishes to transmit data to node E. ecause node D can use directional antennas, and knows location of nodes and, it can determine that data transmission to E cannot interfere with the data transfer from to. Therefore, node D can send a directional RTS packet towards node E. Thus, if node D can verify that its data transmission to node E would not interfere with data transfer from to, D is allowed to send a DRTS control packet to E. s a result, our modied M protocols for directional antennas can improve performance by allowing simultaneous transmissions that are disallowed when using only omnidirectional antennas. D E DRTS () OTS (, ) OTS (, ) DRTS (D) DT OTS (D, E) OTS (D, E) DT Figure 2: Example of using a directional M #1 protocol: This gure uses notation similar to Figure 1. Letters in parentheses, such as DRTS() or OTS(,), denote the physical location information included in the message. The white box below node D denotes that node D may transmit to node E in the corresponding duration, unlike when using omnidirectional antennas. Similarly, in Figure 3, node is allowed to transmit to node F while transmission between and is taking place. This is possible because node does not receive the DRTS from node, so node is not blocked from transmitting DRTS to node F. Note that, with standard omnidirectional RTS/TS mechanisms, node in Figure 3 must defer transmission to node F until the transmission from node to nishes, causing performance degradation. 5

6 F D DRTS () DRTS () OTS (, F) OTS (, F) OTS (, ) OTS (, ) DT DT Figure 3: nother Example of using directional M mechanism #1 6

7 Now consider some other node X in Figure 3 whose location is covered under the directional antenna of pointing towards node. learly, node X may also receive the DRTS from when node sends the DRTS packet to node, as shown in Figure 4(a). With this scenario, we suggest two options for how node X may react. a) Node X MUST Defer Transmission The rst alternative does not allow node X to initiate any data transmission when it has already received the DRTS from someone else (node in Figure 4(a)), until the data transfer indicated in that DRTS is complete. This is because any attempt at data transmission by node X may cause unexpected interference, as in Figure 4(b). In Figure 4(b), node Y's OTS packet following the DRTS by X may cause a collision at node with data packets from node. b) Node Y MY Defer Transmission In the second approach, node X sends a DRTS packet to node Y even though it has already received a DRTS from node. Thus, when X needs to send a DRTS to Y after receiving node 's DRTS, it will check to see whether its directional transmission to node Y can interfere with the ongoing data transfer from to. If it can interfere, node X will not send DRTS to node Y. Otherwise, X will send the DRTS. When a node Y, gets a DRTS packet from some node X, Y may or may not reply with the OTS to X. In order to decide whether or not to reply, Y checks if it has received an OTS packet from any neighbors previously and if the data transfer indicated in that OTS is nished. If node Y has never received any OTS from its neighbors or the proposed data transfer time in the previously receiving OTS is over, node Y will send the OTS back to node X, being followed by the actual data transfer from X to Y. Otherwise, Y will not reply with OTS to node X. For instance, in Figure 4(d), node Y can know the fact that its returning OTS packet for DRTS from node X may cause a collision with data packets from to. Therefore, it will not reply to the OTS for node X. 5.2 Directional M Protocol #2: Impact of K packets In the current IEEE M protocol standard, immediate link layer acknowledgements 3 are employed to determine if the data packet was successfully received. Thus, RTS-TS-DT- K exchange mechanism is used to enhance reliability of data transmission. Note that, in our proposed directional RTS mechanism explained above, no K packet is assumed. Now, let us consider the case when we include the K packet immediately after the DT. Figure 5 shows the case when adding the K packet aects our proposed protocol. In both Figure 5(a) and (b), the K packet is returned after successful data transmission from to. During the transmission session between nodes and, assume that node G whose transmission range includes both nodes and wants to send data packets to node by using a directional antenna in Figure 5(a). lso, let us assume that the directional antenna used for 3 Here, an K is treated as a control packet sent by a M layer. Therefore, no RTS is sent for the K. 7

8 OTS(X,Y) DRTS() X Y DT DRTS(X) DT X Y OTS(X,Y) (a) (b) DRTS(, ) Y DT Y X X DRTS(X,Y) (c) (d) Figure 4: How to react when node receives DRTS? 8

9 sending the DRTS from G to covers node as well, meaning that node also gets the DRTS sent by node G and may experience a collision with the returning K. Node cannot know the fact that is communicating with so it responds with the OTS packet to G, following which node G transmits data. Since this data packet transmitted by node G reaches node as well as, it may cause a collision at node with the K packet returned by node. Of course, the OTS from node can also collide with the K. These collisions harm reliability of data transmission from to. Unlike Figure 5(a), Figure 5(b) shows no impact of K packets because data transmission from node G to node H does not aect node (i.e., no collision at node ). K OTS (,G) DRTS(G) DRTS OTS DT K OTS D DRTS OTS DT K OTS D G DT G DRTS OTS DT K H "ollision" (a) (b) Figure 5: Impact of K packets in the directional M #1 protocol: oth nodes and are covered by the same directional antenna of node G. (a) DRTS and DT from node G destined for would also reach node, potentially causing collisions with a returning K from node. Similarly, OTS from node can collide with the K. (b) No collision occurs in this case. In order to tackle the case of Figure 5(a), our proposed directional M protocol #1 needs to be modied as follows: Use omnidirectional RTS: To take K packets into account, using omnidirectional RTS (ORTS) packets has to be considered. Recall that the proposed directional M mechanism #1 discussed above consists of directional RTS, omnidirectional TS, and directional DT (DRTS-OTS-DT). In Figure 6, unlike the rst directional M protocol, node broadcasts an ORTS for data transmission to node. The ORTS would then reach nodes and G which are within radio range of. In the current RTS/TS mechanism with omnidirectional RTS and TS packets explained in Section 4, nodes and G have to wait for the duration of data transmission from to. This is because any attempt at data transmission by them may cause a collision at node. However, in our proposed directional M protocol #2, nodes receiving the ORTS are allowed to transmit data in 9

10 some cases (explained further below). This is possible because we assume the use a directional antenna for data transmission. For example, node G will be allowed data transfer to node H in Figure 6(b), but not to node in Figure 6(a). In Figure 6(b), collisions may still occur at node between the K from node and the ORTS from node G. However, there will be no collision with the long data packets. heck if any nodes receiving an ORTS is allowed for data transfer: When node X receives an ORTS packet from node Y, which is destined for another node Z, X notes node Y's on-going transmission to node Z. If node X now wants to begin data transfer to node Q, it may or may not be allowed to initiate an ORTS transmission. The decision is based on where node Q is, i.e., node Q's location information. In the case when Q is close enough to node Y so a directional antenna of node X in the direction of node Q also covers node Y, the decision of ORTS transmission from X will be negative. Otherwise, the decision will be positive. Figure 6(a) belongs to the case when the above decision is negative. Nodes and exist within the transmission range of node G. In addition, both nodes are covered under the same directional antenna of node G. Therefore, when G sends directional DT packets towards node, the packets also reach node. ecause node G knows that DT for node can collide at node with an K returned by node, G will not send the ORTS to. Instead, node G stays silent for the intended transmission duration from to. In contrast, Figure 6(b) shows the case where node G may perform an ORTS packet transmission. With this scenario, G concludes that a directional DT for node H will never collide with the returning K packet from. Of course, in order for the data transmission between G and H to occur, node H has to send an OTS back to G. Node H will not send OTS for the corresponding ORTS packet if it has heard an OTS packet for a data transfer that is not guaranteed to have completed yet. Otherwise, H will send the OTS. onict-free K packets To avoid the collision of an K with the long DT packets, our directional M protocol #2 requires omnidirectional RTS, omnidirectional TS, directional DT, and directional K (ORTS-OTS-DT-DK). However, as shown in Figure 6(b), the probability of K collisions with small control packets, such as ORTS or OTS, still remains non-zero. To remedy this problem, we present here some approaches. a) Use Two hannels To guarantee no conicts of K packets, the single common channel is split into two separate channels: one for DT and K packet transmission, and the other for ORTS and OTS packet transmission. M-level acknowledgement requires the receiving node of data packets to respond with an K immediately, without exchanging RTS/TS control packets. This implies that K packets are generated by the M layer and they are sent on the data channel which has been used for the corresponding DT reception. 10

11 ORTS () ORTS () G ORTS () OTS (,) DT () K OTS (,) D ORTS (G) ORTS () G ORTS () OTS (G,H) DT (G) K ORTS (G) ORTS () OTS (,) DT () K OTS (,) D ORTS (G) (a) H (b) Figure 6: The proposed directional M mechanism #2: Dotted lines indicate an angle of directional antennas. Observe that, in (a) and (b), node G takes dierent actions when it has data packets for node and H, respectively. Since K packets are transmitted on a dierent channel than other control packets (RTS/TS), conict-free transmission for the K can be guaranteed. b) Exchange nother RTS/TS for K packets We can think of another possible solution for conict-free K packets: RTS/TS exchange for the K itself. single common channel is assumed here. The basic idea is that an K packet is considered as another data packet requiring a successful RTS/TS exchange. Unlike immediate M-level acknowledgement mechanisms described above, K packets are generated by an upper layer such as logical link control (LL). To send the K successfully, another successful exchange of RTS and TS packets is required. Of course, this additional RTS/TS exchange mechanism would decrease bandwidth ef- ciency due to overhead. Thus, there exists a trade-o between reliability of data transmission and the control packet overhead. c) Utilize the Duration Information Omnidirectional RTS packets contain a duration eld that denes the period of time that the channel is to be reserved to complete the actual data transmission and the returning K packets. Therefore, if all nodes within the reception range of an ORTS avoid transmitting the ORTS at the time corresponding OTS or K are expected to be transmitted, collision with Ks will not occur. For the scenario in Figure 6(b), Figure 7 compares the current RTS/TS mechanism in the standard with the second directional M protocol for conict-free Ks in terms of deferred access time at node G. Node G in Figure 7(b) has greater possibility of accessing the medium, compared to that in Figure 7(a), since node uses a directional antenna to transfer data in Figure 7(b). 11

12 Node DIFS RTS SIFS SIFS DT SIFS Node TS K Node G NV (RTS) DIFS ontention Window Defer ccess (a) The urrent RTS/TS Mechanism Node DIFS ORTS SIFS SIFS DT SIFS Node OTS K Node G DIFS ontention Window Defer ccess Defer ccess (b) The Directional M #2 Mechanism Figure 7: onict-free K packets by using duration information (for the topology in Figure 6(b)). 12

13 6 Optimization: dd directional Wait-To-Send (DWTS) packet We showed in Section 5.1 that a simple modication of using directional RTS packets (DRTSs) can potentially improve performance of mobile ad hoc networks. Let us consider another scenario for using DRTS packets. In Figure 8, nodes and communicate with each other for some duration of time, similar to Figure 2. However, unlike Figure 2, where node D has data packets for node E during that period of time, now node E wishes to transmit to node D. When using the rst directional M mechanism, node E sends a DRTS in the direction of node D and expects a OTS packet to be returned from D. Node D may know the fact that node is receiving data packets from node so its OTS reply for node E can disturb node 's data reception from node. Therefore, node D will be silent despite a DRTS from node E until the proposed transmission between and is done. This can cause unnecessary retransmission of directional RTS from E to D (See Figure 8). This situation would happen in the current IEEE protocol as well. D? E DRTS() OTS (, ) OTS (, ) DRTS(E) DT retransmitted DRTS (E) DRTS(E) Figure 8: Useless retransmission of RTS packets One solution to prevent this situation is to introduce a short control packet, a directional Wait-To-Send (DWTS). DWTS messages can be used for preventing useless retransmission of RTS packets by telling how much time to wait before retrying the RTS packets. Thus, a DWTS packet contains a duration eld that indicates the period a node must wait for transmission. When a node receives a directional RTS (DRTS) packet from its neighbor while it is aware of another ongoing transmission, it replies with a DWTS packet to the neighbor that sent the 13

14 DRTS packet. Figure 9 illustrates this mechanism. In the gure, a DRTS packet from node E follows an omnidirectional TS (OTS) packet from node. Upon receiving the DRTS, node D returns a DWTS packet back to node E because D cannot reply with a OTS packet for node E at this time. Using directional WTS packets can avoid retransmission of DRTS packet by node E until the time specied in the duration eld of DWTS packet has elapsed. When E sends the next DRTS (after waiting appropriate duration), node D replies with an OTS (See Figure 9). The main idea of using DWTS packet is to let node E know about how much to wait before retrying the DRTS packet. There may also be other situations where using the DWTS packet is useful. D E DRTS() OTS (, ) DT OTS (, ) DRTS (E) DWTS (D) DRTS(E) OTS (D, E) OTS (D, E) DT Figure 9: Example of adding directional Wait-To-Send (DWTS) packets 14

15 7 Location Information The assumption in the above discussion is that a mobile node knows its own location and neighbors' location accurately { this information is necessary to determine which directional antenna to use either to send DRTS or DT. When the nodes are mobile, it is hard to know precise location of a node at all times. mobile node may inform its location to its neighbor periodically using beacons. lso, the location information could be included in other messages (such as RTS and TS). However, due to node mobility, the location information can become stale. Since we suggest using directional antennas for DRTS and/or data, it is useful to consider how the protocol should be modied when location information is not known accurately. When a node X wishes to send data to node Y, it may send DRTS or ORTS, using our protocols. Of course, for sending ORTS, node X need not know Y's location. However, to send a DRTS, X needs to know the location. If X does not have any location information for Y, then the DRTS may be replaced by ORTS, without loss of correctness. On the other hand, if node X does know, potentially out-dated, location of node Y, then X can transmit the DRTS in the appropriate direction. reply may not received to the rst DRTS, due to various reasons, such as transmission errors or because the out-dated location information resulted in the use of a directional antenna that does not cover the current location of node Y. To deal with causes such as errors, node X may retransmit the DRTS after a suitable back-o interval. However, to recover from out-dated location information, an ORTS must be transmitted. Thus, in general, node X may retransmit the DRTS upto a specied threshold, and then default to using an ORTS. It is important to note that using an ORTS instead of a DRTS does not aect correctness of the M protocol. When sending the data as well, node X uses a directional antenna. Since an RTS/TS exchange precedes data transmission, and since location information of node Y can be included in the TS message, node X has accurate location information. Node X can use this information to choose the appropriate directional antennas. There is always a (small) probability that host Y moves out of scope of the chosen directional antenna during the data transfer. This may result in the loss of the data packet, and may be handled similar to a loss due to transmission errors. 8 Performance Evaluation To evaluate proposed M protocols, we plan a simulation study using a modied version of the ns-2 simulator [4]. In addition, we will also attempt a mathematical analysis. 9 onclusions The current M protocols using omnidirectional Request-to-Send (RTS) and lear-to-send (TS) can waste wireless bandwidth by reserving the wireless medium over large area. To improve bandwidth eciency of the previous M protocols, we propose a new approach, named \directional M", utilizing the directional transmission capability of a directional antenna. 15

16 We considered several possible cases and argued that our directional mechanisms can improve performance by allowing simultaneous transmissions that are not allowed in the current M protocols. We also discussed an optimization using directional Wait-to-Send (DWTS). We are presently doing a simulation study to evaluate our protocols. References [1] \Iowa State University GPS page." Web site at [2] \Wireless LN medium access control (M) and physical layer (PHY) specications," Draft Standard IEEE , P802.11/D1: The editors of IEEE [3] S. orson and J. Macker, \Mobile ad hoc networking (MNET): Routing protocol performance issues and evaluation considerations (Internet-Draft)," Mobile d-hoc Network (MNET) Working Group, IETF, Oct [4] K. Fall and K. Varadhan, \ns Notes and Documentation," ugust Web site at [5] M. Horneer and D. Plassmann, \Directed antennas in the mobile broadband system," in Proc. of IEEE INFOOM '96, pp. 704{712, [6] P. Karn, \M - new channel access method for packet radio," in Proc. of RRL/RRL mateur Radio 9th omputer Networking onfere nce, September [7] Y.-. Ko and N. H. Vaidya, \Location-aided routing (LR) in mobile ad hoc networks," in M/IEEE the 4th nnual Intl. onference on Mobile omputing and Networking (Mobi- om'98), October [8]. Lau and. Leung, \ slotted LOH packet radio networks with multiple antennas and receivers," IEEE Transactions on vehicular technology, vol. 39, no. 3, pp. 218{226, [9] G. T. Okamoto, Smart ntenna Systems and Wireless LNs. Kluwer cademic Publishers, [10] R. Prakash and M. Singhal, \Impact of unidirectional links in wireless ad hoc networks," in The DIMS Workshop on Mobile Networking and omputing, Mar [11] N. Pronios, \Performance considerations for slotted spread-spectrum random access networks with directional antennas," in Proc. of IEEE GLOEOM '89, Nov [12] T.-S. Yum and K.-W. Hung, \Design algorithms for multihop packet radio networks with multiple directional antennas stations," IEEE Transactions on communications, vol. 40, no. 11, pp. 1716{1724, [13] J. Zander, \Slotted LOH multihop packet radio networks with directional antennas," Electronics Letters, vol. 26, no. 25,