You are on page 1of 9

Medium Access Control Protocols Using Directional Antennas in Ad Hoc Networks

Young-Bae KO, Vinaychandra Shankarkumar, and Nitin H. Vaidya Department of Computer Sciencc Texas A&M University College Station, TX 77843-3 112, USA y oungbae,vOs5080,vaidya @ cs.tamu .edu
for djrcctional antennas. Othcr researchers have also suggested using directional antcnnas for packet radio nclworks [ 1 I], [13], [191. Recently, usc of adaptive antennas has been considcrctl in packct-switched systeins. For example, [ 171 and [ 1111 showcd that using adaptive antennas can make the performance nf a slotted ALOHA packet radio iictworks to improve. In thcir study. the adaptivc antenna basestation is allowcd to receive multiple packets simultaneously, resulting in ihc pcrformance improveI. INTRODUCTION ment. This ability of inulliplc reception at the same time is A wireless, mobile ad hoc network is an autonomous system known as space division niultiple access (SDMA), SDMA has nf niabilc nodcs which wc typically assumed to hc quipped been treatcd as R technology to incrcasr: thc capacity of cclluwilh oninidirecriomd anterrnm [5]. However, i t i s also possible Iar mobile coinmunication systems and has latcly been studied to use dirwtiond antelinus 171 or crhprive aiiteiinus [ 121 10 in>- in [14], [15]. kpccially, in (141, a CSMA/SDMA protocol has prove tlic ad hoc nctwork capacily. Using dircctional antennas kccn presented to mitigate the cffects of tiiddeti terminal probinay o h several interesting advantngcs for ad hoc nctworks. lems. For instancc, routing pcrrortnance cuuld be jmprflvcd by using The hidden terminal problem arises due to the possibility that a directianal antenna (Tor route discnvcry [IO] iir for data deliv- transmissions from two nodcs which cannot hear each nlhcr, cry). may interfere at a third node. Unfortunately, many packet raTo best utiliec directional antennas, ii suitable medium access dio network eiivironmcnts suifer from packet corruption due to control (MAC) protocol must he uscd. Current MAC protocols, this probicin. While modern MAC protocols For nmnidircctional such as thc E E E 802.1 1 standard 121, do not hcncfit when us- antennas havc taken this problein into account 141, [SI, [ 161, i t ing dircctional antclinas, hecniisc these protocols haw hccn de- is not atlequately consitlcred in previous studies of using direcsigned tn exploit onmidircctiotial antcnnas. In this paper, we tional antennas in packct radio networks. propose ncw MAC pro(ocols using directional antcnnas. Physical size limitations may prohibit thc usc of directional 1 1 1 . PHELIMINARIES antemias on handheld devices. However, it i x practical to use thcm nn vehicle-niounteddeviccs. Also, usc oihigher frequency A. Network Model bands rcduces the physical size or directional antennas. We ilSSllmc that d l hosts i n a rcgiou share a wireless channcl
Absrmc~-Using dlrectlnnal antcnntts can be I~enefldalfor wireless nd linc iictworkq coiwisting of R cotlcction of wirclcss hnsh. To bcst iitillze dlrcctiorial antennns, n snituhle Mcdiiini Access Control (MAC) protocolmust lic dcsigned. Currcnt MAC protocuIs,sucli ns thc IEEE 802.11 standard, do not Iwncfil wlicn d n g dircctional anteiinos, hccntLse thcsc protocols havc heen designed for ninnidircctlntinl nutcnnas. 111 this pnpcr, we ottcmpt to dcslgih ncw MAL prolocals sidtnblc for nd hoc nctrvorks based on dircct1or1nl nntcnnas. Key~vi~rrlr-hlACprotocols, Directionnl Antennns, Ad Hoc Nctworks.

and communicatc on that shared chantiel, Each host is assumed to be equippcd with multiple directional antcnnas. A directional Although work 011 MAC pIotocols for dircctional antennas antenna can transinit ovcr R small angle (e.g,, 90 dcgrees), and has becn limited, some researchers havc previously suggested several directional antennas inay be uscd togclhcr to cover all usc ordirectional antennas for packet radio networks. For exam- directions2. plc, Zander 1201 has proposed Ihe use of directional antcnnas in Wc assume that transmissions by two different nullcs will inslullcd ALOHA multihop packet radio nctworks whosc hroad- terferc at some node X, cvcn if diflcrcnt directional antennas cnst radio channel is shared hy means of thc random timc divi- a i nodo X receive these two rransmissions. This assumpiion is sion multiple access (RTDMA) scheme. Morc rccently, a way of somewhat pcssimistic, and rcmoving this assumption will imusing adaptive dircctional antennas for Mobilc Broadband Sys- prove the performance af the proposed protocols. This w u m p tems (MBS) has been proposcd [71. In [71, the authors argue t i r i r i is justified, for instancc, in thc case where signals rcccived that convcntional MAC protocols arc not suitahlc Tor directional by all antcnnas arc combined hcforc scnding to the receiving ciriintcnniis and suggest a dynamic slot assignineiit (DSA) protocol cuitry. We alsti assume that siinullancous (imsmissinns hy ttic sainc node to differcnt tlircctions fire not allowcd. I his mscarcti is supporlcd ill piu7 by National SFir~cc~0iiiidiirii)~grnntgANI11, RELATED WORK
997662h,AN1-99731 Slnnd 9972235,
Ari omnidireclional niitcniiulransriiits in

all directions [i.c.,360dcgmCS).

aAlicmaiivcly, an omnidirecrin~lnl ainciina m n y be used.

0-7~03-~880-$/001$10.00 (c) 2000 ILEE

13

lEBE [NFOCOM 2000

Each host has a fixed transmission range and two hosts are said to be neighbors if thcy can communicate with each other over a wireless link. Initially, we assume that each node knows ils neighbors' location as wcll as its own location. At the cnd of this papcr, we briefly considcr thc case when location infonnation is not known accurately. The physical location information may be obtained using thc global positioning system (GPS) [ I ] . Basccl on location of the receiver, the sender may select an appropriate directional antenna tn send packets to the receiver. Most of the current MAC protocols, such as IEEE 802.11 MAC standard 121,use a handshake mechanism implemented by cxchangiiig small control packcts namcd Reqraest-to-Send [RTS) and C1car-ro-Serad(CTS). The successful exchange of these two cotitrol packets rcscrvcs the channel for transmission of thc. potentially longer, data packct and a short acknowledgement (ACK) packet.
B. RTSKTS Mechanism in IEEE 802.I 1 MAC Prorocol
Figure 1 illustrates the IEEE 802.11 MAC protocol [2] for omnidircctional antennas that uses RTS and CTS control messages. In this protocol, any node that wishes to transmit data must send a RTS packet beforc it can start data Iransmission. For example, in Figure 1 , node B broadcasts a RTS packct for its intcndcd rcccivcr, node C. (Please see caption of Figure 1 for an explanation of the figure). I f C rcccives thcRTS successfully, it rcplies with a CTS packet so that I3 can start transmitting a data packet upon receiving the CTS. When node C successfully reccivcs the data packet, it immediately sends an ACK tu node I 3 (thc ACK has a priority over any other transmission by any node in the vicinity of nodes U and C ) .Notc that both RTS and CTS packcts contain the proposed duration of data transmission. Since nodes are assumed to transmit using omnidircctional antcnnas, all nodes within the radio range of 3 and C will hear one or both of those control packcts (nndes A and D in Figure I ) these nodes must wait for thc duration of data transmission before they can transmit themsclvcs. Thus, thc area covcrcd by the transniissioii range or both the sender (node B ) and the rcccivcr (node C ) is reserved for the data transfer from B to C, to prevcnt collisions. This characteristic of RTS/CTS mechanism ovcrcomes the hidden terminal problems in wireless LAN environmcnls. HOWCWF, it is easy to see that this mechanism can waste a large portion of llie network capacity by reserving the wireless medium over a large aren. For instance, even though node D has data packets for node R while B and C are communicating with each other, node D has to defer the transmission to E uiilil Ihe transmission from node B to C is completed.
Pig. 1. MAC prosocol wirh omnidirecriunai RTS/CTS niechuuirm: The circle centeredui emknude shows its trrinsmissiuii range. I n tlie lower halfof the figure, rhie progressesfwn top io boifom. The j g u r e shows message.rsent by varionsnodes. Black bois below nodes.4 mid ll rnriiclrte t h f these nodes are not rillowed tu transmit iri kfie durutiori covered by the bars (to avoid interference with the trnwferfrom B IO C).

is in progress. The directional MAC protocols apply a similar logic, but on B per-antcnna basis. In brief, if antenna T a t node X has received an RTS or CTS rclatcd to an on-going transfer hetween two other nodes, then node X will not transmit anything using antenna T until that other transfcr is completed. Antenna T would be said to be "blocked" for the duration of that transfer - the duration of transfer is included in cach RTS and CTS packet (as in IEEE 802.11), therefore, each nodc can dclcrmiiic when a blocked antenna should become unblocked. The key point to note above is that, whcn using directional antennas, while one directional antcnna at some node may be blocked (as dcrincd above), uiher directional antcnnas at thc same node may not be blockcd, allowing transmission using thc unblocked antennas. This property results in perfonnitice improvement when using directional antenuas. Omnidirectional transmission of n packet in D-MAC schemes requires the use of all thc dircctional antennas. Therefore, an omnidirectional transmission can be perforrncd if and only i f none of the directional antennas are blockcd.

A. Scheme I : Using DRTSpackets


Directional MAC (D-MAC) scheme 1 utilizes a directional antenna for sending the RTS packcis in a particular direction, whcrcas CTS packets are transmitted in all dircctions. Figure 2 and Figure 3 show how wireless bandwidth efficiency o f the 802.1 I MAC protocol can be improved by using a directional MAC protocol. In Figure 2, assume that node B has a data' packet for node C, and also assume that no other data transfers are in progress (so none of the antennas are blocked). In this case, node B sends a directional HTS (DRTS) packet including the physical location infortnation of R , in the dircctioxi o f nodc C. Thus, node A docs not receive the DRTS from node B

Iv. 1)IRECTIONAI.

MAC @-MAC) SCHEMES

The pro[)osed Directiotial MAC (13-MAC) schemes are similar to ll4X 802.1 I in many ways. Thc dircctiona1 MAC schemes also send an ACK immediately a h the DATA, as in 802.1I - however, in D-MAC schemes, the ACKs ni'e sent using a directional antenna, instead of an omnidirectional antenna. In 802.1 I, if a nodc X is aware of an on-going transmission between some other two nodcs (due to the receipt of an RTS or CTS from those nodes), nodc X will not participate in a transfer itself - that is, X will not send a11 RTS, or send reply to an RTS from another nodc. whilc the transfer between other two nodes

0-7803-5880-5/00/$10.00 (c) 2000 IEEE

14

IEEE lNFOCOM 2000

wards iindc C. Clearly, node X will also rcceivc the DRTS from B when nodc J3 sends the DRTS packet to nodc C, as shown in Figure 4(a) and (b), Therefnrc, the directional antenna iil nodc

X tliot points towards node B will be blrickccl for thc duration


even though node A also exists within E's transmission range. If node C rcccives the DRTS packet from B successfully, it then returns an urnnidirectiod CTS (OCTS) reply. Tho location informations are included in the OCTS packet: location of the node sending OCTS (node C's Iocafion in Figure 2) and location of the sender of the corresponding DRTS packct (node B in Figure 2). After the successful cxchange of DRTS and OCTS packets, a data packet is sent by nodc B using a directional antenna. When node C receives the data packet, it immediately sends an ACK to node B using n dircctjonal antenna. Now, during tlie proposed length of transmission between B and C. assume that node D, which is a neighbor of node C, has data to transmit to node E. Nntc that thc ditectional antenna of nodc D that points towards node C is blocked, since nodc D would have received on this dircctionnl sntcnna the OCTS sent by node C to node B.However, tlic blocked antenna is diffcrcnt from thc directicinal antenna timl points towards nodc E. Therefort, node [ > can send a dircctional RTS packct towards node E. Usscntially, i f node 13 knows that its data transmission to node E would not interfere with the other on-ping data transfer from I3 to C, D sends a DKI'S control packet to E. As a rosult, our modified MAC protocol for dircctional antennas can irnprvvc pcrformancc by allowing simultaneous transmissions that are disallowed when usirlg only trmnidirectional antenilas. Similarly, in Figure 3, no& A is aflowcd to transmit to nodeP while Lranstnission between B and C is taking place. This is possible hccause node A tloes not receive the DRTS froin node B, so nodc A is nut hlocked from transmitting [he DRTS to node E Note that, with standard oinnidirectional RTS/C'TS meclinnisms, nodc A in Figure 3 must defer transmission to nodc F until thc transmission from B to C finishes, causing pcrformancc degradation. Lct us now consider some other nndc X in Figure 3 whose loctltion is covcrcd by thc dircclional atitcnna of B pointing too i transfer h a m B to C. With this scenario, in scheme I , node

X is still allowcd to initiate its +ta transinission to some other node Y as long as the directional antenna at node X that points
towards Y is not blocked by the receipt for DRTS from nodc D (or, by the receipt of DRTS or OCTS from some olhcr node). When a node Y gets a DRTS packet from node X,Y tnny or may not send an OCTS to X, dcpenrling on the status of its directional antennas. Since an omnidilcctionnl CTS (OCTS) packet fransmission requires thc iisc of all directional antennas, an OCTS cannot hc sent if m y of thc dircctiond antennas are blocked. Thercforc, node Y transmits an O C l S it) rcply to thc DRTS from node X if and only if none of its dircctional sntennas are blocked. Thus, in Figure 4(a), nndc Y may scnd an OCTS to node X, huwcvcr in Figure 4(b). nodc Y may not scnd thc OCTS. Since, in case (b) of Figure 4, node Y would have rcccivcd an OCTS fiom node C blocking its dircciional anlcnna For the duratioii cif Iransfer from B to C. l h c appendix briefly presents pseudo-codc ror oiir 11-MAC scheme I .
H. Schenre 2: W s h g bvth DRTS ond OXTS pclckezs

In our first directional MAC prutucol to improve nctwork performancc, a directional RTS (DRTS) packcl is trensmittcd in the direction of thc intended receivcr prior to the transmission of thc actual data packets - U-MACscheme 1 docs not use oiiinidirccpackets. Using DRTS, instead of omnidirectional RTS (ORTS) tional RTS, may increase thc probability of control packet collisions in sonic cascs. We consider one such scemrio in Figiire 5 . In Figure 5 , assume that nodc B has initiated n packet transFer to iiodc C. Node A is unaware of this iransfcr, sincc node E's DRTS to node C has not reccived by node A. Now, node A wants to send a packet to node B, while B's trani;fcr to norlc C I S still it1 progress. Transmission of a IIRTS by node A to node l 3 inay i n tcrferc with the rcception of OCTS or ACK control packcls sent

0-7803 -5 8SO-j/OO/$I0.00 (c) 2000 113%

15

IEEE lNFOCOM 2000

,'-

. . ... .

-~ ..

. - - - ..

DRTS,our D-MAC scheme 2 is identical to scheme I .

by node C to node TI3. Note that node A does not defer ifs attempt to communicate with node B because A has not receivcd node B's DRTS packet directed to nodc C. This siluntiun cannot happen in thc current omnidirectional RTS/CTS exchange mechanisms. Since the size ofcontrol packets is typically much sinaller than 1hc data packets, thc probability of collisions described above is not vcry high, although i t is higher than rhat in casc of IEEE 802.I 1 MAC. To reclucc the probability of collisions between control packets, we. proposc directional MAC schcme 2, In D-MAC schetne 2, thcrc are two types o l RTS packets: direclional HTS (DRTS) and ornnidirecdional RTS (OKTSJ. In schcme 2,when a nodc, say node X, wishes to initiate a data transfer, it may send ORTS or DRTS as per two rules: (a) if tione of thc directional antennas at node X arc blocked, then node X will send an ainnidirectional RTS (ORTS). (b) otherwise, notlc X will send a dircclional RTS (DRTS) provided that the desired dircctional antenna is not hlocked. If thc desired antenna is blockcd, node X will defer until that antenna becomes tinblocked For cxamplc, i n Figure 5, assume that whcn node B wants to send a packet to node C, n m c of the antennas ;It B are blocked. In this case, node B will broadcast an ORTS packet (as pcr rule (a) abovc). Since this packct will be receivcd by node A, its directional antenna pointing towards B will be blocked for the duration of the transfer from 5 to C. Now considcr two cases: I If node A wants to send data to node B, it will wait for the duration uf transfer from B to C (until thc corresponding directional antenna hccomes unblocked, as per rule (b) above). I If node A wants to send data lo node F, node A will send a DRTS to nodc I?, provided that thc directional antenna pointing towards nodc F is not blocked (as per rule (b) above). Thc combination of DRTS and ORTS packcts in schcme 2 can reduce the C ~ S C S of collisions bctween control packcts (although i i ilucs not clirninatc thc possit>iIityj. Apart from the [WO r d c s mcntioncd above which dclerniine if a iiodc will send ORTS or
'Recall that wc niakc the pessimistic assulnption that signals rccclvcd on different diicctiunal antciinas nt n givzn node can intcrfcrc. If this assumption is nor true, rhen scheme 2 i s not necdcd, and performancc of echernr 1 would bc better than reporlcd hcm. However, when rhc rcccivcr hrrdwni-e is coiistmined, uiir pessiinistic nssulnprion mliy also be mnic.

lr. PERFORMANCE EVALUATION

To evaluate our protocols, we pcrrormed simulatiofls using an cxtended vcrsion of tho UCBLRNL network simulator, tis-2 [3], 161.The tis-2 simulator is a discrete evcnt network simulator ihat WAS dcvcloped as part ol'VINTproject at the 1,nwrencc Derketey National Laboratory. Thc cxtensions iniplerncnted by the CMU Monarch piojecl - which enable it to accurately simulate mobilc nodes conncctcd by wircless netwnrk interfaces and multi-hop wireless ad hoc networks - were used. Wc modified the ns-2 10 implement 1he directional antennas which could transmit in a particular direction (90 dcgrccs). In our morlificahi, location information was incorporated into the wirelcss transmissions.

A. Simulation Model

We consider the 5x5 mesh tupology illustratcd in Figurc 6. Thc nodes form 5 rows and 5 columns, with two adjacent rows and two adjacent columns being separated by 200 mcters. We briefly evaluate the caw of 3x3 and 6x6 mcsli topologies at the end of the next subsection. Transmission range of cach nodc is 250 metcrs and the wireless link bandwidth is 2 M b p TCPRcnu i s used for lhc transport layer ovcr the IEEE 802.1 1 MAC layer. The traffic modcl used in our simulation is FTP with infinite backlog at each sourcc node. Thc TCP packet size is 1460 bytes and the innximum advertiscd window is 8 packcts. Each siniulntion is performed for a duration of 900 Tecoricls. Each performance tneasurcment rcported bclow is averaged o w 20 excwtions.
13. Sinzuiariori Rcsult~

The performance metric uscd to evnlualc the protocols is TCP throughput. Thc unit for all throughput measurements rclm-tcd here is KiIobils/second (Kbps).
IEEE INFOCOM 2000

0-7803-5880-5/00/$10.U0 (c) 2000 IEEE

16

Notc that thc aggregate throughput o f the two connections 802.11 is comparable to that of a single TCP conncction using IEEE K02.11 - essentially, in this caw, when 2 connections are opcnetl, thcy share the handwidth [hat would have hccn otherwise available to il singtc connectiun.
shovc using IIlEE
TABLE I1

- . . .
The second scenario considcrcd here also consivls of two single-hop TCP connections, conncction numhcrcd 3 horn node 6 to nodc 1 nnd connection numbered 4 from tiodc 11 to nodc 16 (similar to Figure 3). The simulation rcsults from the second scenario are reported i n Tdilc 11. The total throughput using the 802.11 protocol in sccnario 2 is camparoblc with sccnario 1. Howcver, the pcrformancc of the D-MAC scheme 1 in 'Lhblc I1 is bettcr than in scenario 1 . This is becausc, in sccnario 2, therc i s a sinallcr probability of control packet collisions whcn using the D-MAC schcmc 1, as compared to in scenariw I . For instancc in scenario 1, imaginc that nodc 6 transniits a UKTS packet to node I 1 while nodc 16 has alrcady started a packct transfer to node 2 I . Because node 11 would not h w c received node I O ' S DRTS packet, it may scntl an OCTS packet in reply to the DRTS from node 6. This OCTS packet can intcrkre with tho reception or ACK from 21 to 16, causing degradation or the pcrformancc. Despite this possibility of collision of OCr'S anti ACK control packets, Table I also shows that aggregate throughput with D-MAC scheinc 1 is better than the 802.1I - the rcason is that the performance hcncfit of being ablc to perform inultiplc transfers in vicinity of each other (which may be disallowed in 802.1 l), outweighs the putcntial perforriame loss duc to collision d control packcts. In sccnario 2, such collisions cannot occiir (sincc the direction of datn transfer is diffcrcnt from scenario I) - thereforc, observe that scheme I yiclds throughput twice that ol802. I1 , Scenario 2 rcpresents the hcst case Tor the me of directional nntcnnas.
'TABLE 111

In our simulalions, we experimented with TCP conncctions that lrawrsc diffcrcnt numbcr of bops. Notc that throughput of a TCP connection decreases quite rapidly when the numbcr of wirelcss hops is increased from 1 to 4. For future reference, the throughput of a single TCP conncction ufiing the 802.11 protocol, as a function of the number of hops, is as follows: one hop 1383.4 Kbps, two hops 687 Kbps, three hops 412.5 Kbps and four hops 274.8 Khps. 'I'he first sccnario considercd in our evaluation consists of two single-hop TCP connections, coiinection nutnbercd L from node 6 lo nodc I I and connectioii nuinbcrcd 2 from node 16 to node 21 (this scenario is similar to that in Figure 2). Table 1presents the results for thc first sccnario.
TABLE I

No.1(6+11)
No.2(16 3 21) Total Throughyur (Kbps)

1 130.42 214.57 1344.99

771.27
1040.21 I 8 11.48

In Table I (and other [ablcs i n this paper), B row labeled N o . i provides throughpui measurements for connection numbered 1. The row labeled Total Throughput lists the sum of throughput of all TCP connections considered in the sccnario. Dii'ferent columns of thc table correspond to diffcrcnt MAC schcmes. 111 Table I, the total throughput of I)-MAC Scheme 1 is higher than the IEEE 802.1 1 MAC scheme. This is because, with DMAC Schemc I , simultaneous transmissions on the two TCP connections arc allowcd by using directional RTS packct. On the nther hand, whcn using thc 802.1 1 schcmc, the two connections cannot transmit packets at the s i l m timc. Another intcresting issue in the ubove sccnario isfuirriess. As can be sccn i t ) Tahle I, thc behavior or IEEE 802.11 protocol is unfair in that throughput o r conncction nutnhcr 2 is much lower than that achicvcd withconnection number I.The fairness using schcme 1 is much better than 802.1I .
0-7803-5S80-~/001~l0,00 (c) 2000 l1XE

1 conncctions

11

IEEE 802.1 I

Schemc 1

Schrinc 2

Now we consider scenario 3. in which 1%P conncction number 5 i s cstablishcd from iiode 1 to ncidc 21, and conncction number 6 from nodc 1 to nodc 5 in Figurc 6. Thus, coriiicction 5 traverses a row in the 5x5 mesh, and connection 6 tra1 1 verses a column. Buth paths consist of four hops. Table 1 presents tbc throughput mcasurcments. Observe that, in this case, all three MAC scheincs are quitc Lair. Ikom thc pcrfortrtancc point o C view, 1)-MAC schemes 1 and 2 both achicve significant improvement ovcr 802.11, with schcme 2 nchicving
[BEE INFOCOM 2000

17

the largest throughput. Recall that, with D-MAC scheme 2, the probability o f control packet collisions decreasos. Therefore, scheme 2 sotnetimcs (not always, as seen later) achieves higher ttiroughpur than scbernc 1
I

In this case, although both D-MAC schemes achieve sigtiificantly better throughput than 802.1 1, D-MAC schcme 1 pcrforins bctlcr than scheme 2.
As discussed earlier, D-MAC scheme 2 cun reduce the probability of collision of control packets, This factor usually contributes to an increase in aggregate throughput. However, ORTS packets in schemc 2 also reduce thc possibilities for simultaneous transmission by neighboring nodcs (below we prcsent an example to illustrate this). Thus, the nctwork performa~~ce improvement .by scheme 2 (compared to scheme 1) depends on whether the benefit of reducing control packet collision outweighs the decrease in throughput resulting from reduccd possibilities for simultaneous packet transinissions. Thus, here exists a trade-off between probability of collisions of control packets and djsallowed simultaneoiis transmissions, when D-MAC schemes 1 and 2 are compared,

TABLE 1V

Connections

lEeE 802.1 1

Scheme 1

Scheme 2

No.7 (1 + 21)
~~~

No.8 12 + 221 No.9 (3 -+23) No. 10 (4 -+ 24) No.1 I (5 + 25) Total Throughput

157.50 89.90
22.00 89.29 157.94 5 16.63

146.73 85.31
91.39

165.89 81.30
105.03 82.83

82.30 153.30
559.03

163.37
598.42

Scenario 4 consists of 5 TCP connections, each connection traverses one row of the 5x5 mesh in Figure 6. Thus, the 5 TCP connections are from node 1 to 21, node 2 to 22, node 3 to 23, nodc 4 CO 24 and node 5 to 25. These connections wjll be referred to as connections numbered 7 through 11, respectively. lable IV prcscnts the throughput measureinent~for scenario 4 . For all three schemcs, the "bordcr" connections numbered 7 and I 1 show much highcr throughput than other interior connections (connections 8, 9 and 10). This is because the border connections share wireless medium with only onc other conuection, whereas the interior connections share the medium with two other connections, Similar to the case of two connections ill l'able 111, both directional MAC schemes have better total throughput than 1EEE 802.1 1, with D-MAC scheme 2 achieviiig the highesl total throughput. Howevcr, the percentage improvernenl in scenario 4 is not as large as scenario'3, because even with directional antennas transfers on two ndjaccnt TCP connections may interfcrc. However, observe that the D-MAC schemes arc somewhat hirer (particularly. to connection 9) than 802.1 1 i n this scenario.
TABLE V

&-(-$
DRTS

DRTS
(a) Simullaneous Ticasmission

with D-MAC scheme 1

(b) No Simultaneous Transmission with D-MACscheme 2

Fig. 7. D#ercnce behveeir D-MACschemds I on112

Con nccri ons

[EEE802.11

1 Scheme 1 I Schenic2

To illustrate the above issue, consider thc network consisting of 6 nodes in Figure 7. Assume [hat 2 TCP connections are established - one from node E to C nnd another from A to B. In Figure 7(a), when D-MAC scheme I is used, node E can transmit to nodc C, whilc node A is transmitting to node B. Howevcr, with D-MAC scheme 2, this may not always be possi blc - whcn nodc E sends an DRTS to nodc C, nodc C will not send OCTS to E, if it has heard an ORTS for an (in-going transfer from node A (See Figurc 7(b)). Due to scenarios similar to thc above, we believe Ihat, D-MAC sctiemc I allows more sitnultaneous transmissions when the i~umbetof Connections are incrcased both horizontally and vertically, compared to D-MAC schcmc 2. This results io Inrgcr aggregate throughput for D-MAC schcmc I. To vcrify this intuition, we now consider similar siinulations with 3x3 and 6x6 topologies. We tnensure thc aggregatc throughput for 6 TCP connections in 3x3 mesh topology (one connection along each row and column) and for 12 TCP coliticctions in 6x6 topology (again, one connection along cach row and column), to compare with the results for the IO connections in 5x5 topology. Table VI presents the aggregate throughput achicved by thc TCPconnections using thc three MAC schcines. Obscrvc that D-MAC schcme 2 has the largest throughput in the case o f 3x3 topology, whcrcas D-MAC schclne I has the largest throughput with 6x6 topology. In summary, we conclude that thc crfects of concurrent transmission is inversely proportional to the complexity of TCP conncciion "topology."

Tolal Throughput

)I

537.96

657.61

593.20

Ncxt, in scenario 5 , we increase the number of TCP conncc[ions to 10, with 5 connections travcrsing the 5 rows, and 5 conncctions traversing the 5 columns i n Figure 6. Bach of thcse connections traverses four hops. Table V prcsents the thruughput measuremonts lor scenario 5.
0-7803-5880-5/OO/$10.00

(C)

2000 IEEE

18

IEEE INFOUOM 2000

packet. Figurc 9 illustrates this mechanism. In the figure, a DRTS 11 BFE 802 1 1 IP Scheme 1 Scheme 2 I I Touoloav . packet from nodes E to D follows an omnidirectional CTS --901.22 1165.32 653.42 3x3 ( 6 conn.) (OCTS) packet from nodc C. Upoti recciving the DRTS, node D 657.61 593.20 537.96 5x5 (10 conn.) returns a DWTS p x k c t back to nodc E hccause D cannot reply 495.86 635.60 576.85 6x6 (12 conn,) with an OCTS packet for nodc E at this time, Using directional WTS packets cm avoid useless rctransmission of DRTS packet by node E until the timo specified in thc duration field of DWTS packet has elapsed. Whcn E sends the iiext DRTS {after wailiny; VI. ADD~TIONAI, DISCUSSIONS appropriatc duration), node D replies with an OCTS (SCC FigA. Optinrization: Dirmtiotzak Wait-W S e n d (DWTS)pucket urc 9). The main idca of using DWTS packet is to Ict node E We showcd above that using dircctional RTS packets (DRTS) know about how much to wait before retrying thc DRTS pxkct. can potentially improve pcrformancc of wireless ad hoc nctworks. I c t us consider another scenario for using DRTS packets. In Figurc 8, nodes 13 and C coriimunicate wilh each othcr for sumc duration of time, similar tu Figure 2. However, unlike Figurc 2, wtierc node D lias data packds for notlc II during that pcriod of time, now nodc E wishcs to transmit to node D. Wlicn using thc first dircctional MAC mechanism, node P, scnds a UKTS in the direction of node D and expects an OCTS packcl to be rcturned from D. Nodc D inay know the fact that node C is rccciving data packcts from nudc B so its OCTS reply for node 1 : can disturb node C's data recepticm f'rorii nodc B. Tliereforc, 13 will be silent despite n TIRTS from node E until the proposcd lransinission hctween R and C is done. This can cause unncccssary retransmission of directional RTS from E to D (SCCFigure 8). This situation would happen in the current IEEE 802.1 1 prntocol as well.
-I

TABLE VI

B. Conj%ct-Free ACK
In thc current IEUE 802.11 MAC protocol stantlard, iniriiediate !ink layer acknowledgements" are anployed to deicrmine if the data packet was succcssfully rwcivcd. Thus, RTS-CTSIM'A'TA-ACK excliangc mcchanisrn is wed to enbance reliability of data transmission. Note thal, in our proposed dircctionnl MAC schemes, rcturning thc ACK packet immcrliately aftcr thc DATA is also assumed. 111 802.11. ACK pnckct collisiotis are minimized since thc transmiusion range of both a transmiltcr and n rcccivcr is reserved. However, in our D-MAC scheincs, iherc is nu guarantee of collision-free ACK packet reception cvcn ihougli schcmc 2 rcduccs the prohabiljty of packet collisions by using ORTS packcis. To rcrnedy thisprohlcm of no ACK collision guarantee, we present somc approaches below, which would be evaluatcd in futurc work a) Use Wo Channels '1'0 guarantee no conflicts of ACK packets, the single c o n " channel inay he split into tw; scpnrate channcls: me for D N A

I
Y

/ I

One srilution to prcvciit this utuation is to introduce a short control packet, directiond Wuit-fii-Send (DWTS). DWTS inessages can be iiscd for prevcnling useless retransmission of KTS packcts by lclling how much timc to wait hct'ore retrying the K1S packcls. Ttius. a DWTS packet contains a duration held that indicates thc period a nodc must wait fur transmission. Whcn B nodc rcceives a directional R'TS (DR'I'S) packet frwin its neighbor while it is aware {it' another on-going transmission, it replies with a DWTS packet to thc neighhor that scnt the DKTS
0-7803-5880-5/00/$I0.00 (c) 2000 IEIT

'.

'Hcm, an ACK is ticared ns il coiitrol packet sent by U MAClayer. Thcrcfola, no RTS is wilt for the ACK. 6Morc disciissIons itbout guaranteeing nn conflict of ACK Imckcts can t ~ c

found in [e].

29

IEEE INFOCOM 20D0

and ACK packet transmission, and the other for RTS and CT5 packct transmission. MAC-lcvcl ucknowlcdgemen~ requires the receiving node or dah packets to respond with an ACK immcdiatcly, without exchanging K W C T S control packets. This itnplies that ACK packets are gcncrated by thc MAC laycr and thcy are scnt on the data channel which has b e c ~ used for the corresponding DATA rcception. Since ACK packets arc transmitted on a different chaiiiiel than other control packcts (RTSKTS), conilict-l"rce transinissioti for thc ACK can he gunrantccd. i J ) Exchangc Another RTS/C'I'S for ACK packets Another possiblc solution is to perform RTS/L'TS exchunge for !he ACK h e w . A singlc cotnmon channel is assumed hcrc. The basic idea is lhat an ACIC. picket i s considered as another data packet requiring a succcssful RTS/C'I'S zxchange. Unlike immediate MAC-leve.1 ackimwlcdgcnient mcchariisms described ahrive, ACK packets arc gcneerated hy an uppcr layer such as logical link control (1,LC). To send thc ACK successfully, another succcssful exchange of RTS nnrl CTS packcts is requircd. Or course, this additional RTS/CTS cxchange niectinnism w o d d decreasc bandwidth efficiency due to overhead. Thus, thcrc cxists a trade-ob bctween rcliahility of data transmission and the control packct overhead.

the CTS message, node X has accurate lucation information. No& X can use this infurmation to choose thc appropriate direct.imal antennas. There is always a (smell) probability that nodc Y moves out or scope of the chosen directiona1 antenna durirrg the data transfer. This may rcsult in the loss of the dara packet, nnd inay be hantllcd similar to a loss due to transmission crrnrs.

VI[. CONCLUSION
The current MAC protocols using oninidircctional RequestLo-Send (RTS) and Clear-to-Send (CTS) can waste wireless bandwidth hy rescrving thc wireless medium over a large m a . 'Io improve bandwidth cfficiency of the previous MAC protocols, we proposo c ncw approach, namcd "Directional MAC" (D-MAC), utiljzing the directional transmission capabilily of fl dircctional antenna. We considered scveral possiblc cases and proposed two diffcrcnt schemes: D-MAC scheme 1 for Using only directionr1 RTS (DRTS)packets, atid D-MAC scheme 2 for using bnlh directional RTS and omnidirectional RTS (ORTS) packets. We also discusscd an optimialion u&g dircctional Wail-lo-Send (DWTS) packets to prevent uiiiiecessary retransmissions of RTS ptickctr. By siinulation studies, wc cornparcd our Jjrectional MAC mcchanistns to the IEEE 802.1 I protocol. In suinmary, our dircctional MAC protocols can improve performance by allowing sirnultaneous transmissions that arc not allowed in (he ciirrcnt MAC protocols.

C. Location Information
Thc assumption in the above discussion is thar a node knows its own location and neighbors' location accuratcly - this information is neccssary to dclertnine which direclionel antcnna to use cithcr to send DR'I'S or DA'lA. Whcti the nodes are mobile, i t is hard to know rhe precise location of fl node at all times. A mobilc iiode may inform irs location lo its neighbor periodically using beacons. Also, the location infomiation could be included i n othcr messagcs (such as KI'S and CTS). However, diic to node mobility, the location inPorimtion can become stale. Since wc suggcsl using directional antennas [or DRTS and/or D N A , i t is uscful to consider how the prolocnl should he ~nodifiedwhen Iocatioii information is not known accurately. When a nodc X wishes to sciid data L o nnde Y, it may sctid DRTS or OKTS, using our protocols. Of coursc, for sending ORTS. nude X nced not know Y's ](>cation. However, to send DRTS, X needs to know thc location. If X does not have any location information for Y , then the DRTS m a y be replaced hy ORTS, without loss of correctness. On thc other hand, if nodc X does know, potentially out-dated, location of node Y, then X can transmit the DR1'S in thc appropriatc direction, A reply l o the hrst DR'I'S may nut bc rcccived, duc to various reasons, such ns transniissiotl errors or because the oot-dated location information resulted in the iisc of a directional antenna that does not covcr the cui-rent location of nnde Y. To deal with causes such as errors, nodc X inay t.ctrannnit thc PRTS after a suitable back-(iff interval. Howcvcr, to recovcr from out-dated locatiaii information, s n OKIS must he transmitted. Thus, in gctieral, node X may retransmit the DRYS upto a specificd threshold, and then det'ault to using an ORTS. It i s important to note that using an OKTS instcad of a DRTS does not affect coirectness o f Ihc MAC protocol. When sciiding the data as well, nodc X uses a directional antcnna. Since an RTS/CTS exchange precedcs data transmission, and since location information of nodc Y can bc included in
0-7803-5880-5100/$10.00 (c) 2000 1EEE

ACKNOWLEDGMENT
We thank the rcfcrees for their helpful comments.

APPENDLX
Thc pscudo-ctide for directional MAC scheme 1 is presented helow. This prcscntatioii nssumcs familiarity with IEEE 802. I I MAC and thc associated jargon and abbraviatioiis. Thcre we two limcrs in 802.1 1: defer timer and backoff timer. A defer timer is uscd to wait for thc appropriate interframe spacc (IFS) when the transmission mcdiurn is idle. Note that threc IPS intervals are specified in the standard: short IPS (SIFS), point coordination function IFS (PIPSIIand distributed coordination futiction IFS (DIFS). A hackoff tinier is used to wait for thc duration o f a contention window (CW) as per 802.1 1. NAV stands for network allocation vcctor and i t indicates the atnount or time ttint rniist elapsc until the current transmission session is COIRpletc. The tcrm "selccted antcniia" in thc pseudo-corle helow refers to the mtcnna to he usctl for the desired dircctiotial transmission.
Procedure M A l N O Benin S e t each o the antenna MAC states as IDLE
Initialize NAV foi the directional antennas

i f (Packet p detected1 chan C a l l procedurn REc'V(pI else i f ( a timaout occurredl t h e n


End
C a l l tho appropriate irandler

Procedure RECV ( P I Beqin

lfcz!?l proccdure 5ENDIpI :


BEgh

is an o u t g o i n g p a c k e t ) than

else i f (MAC recaption s t a t e iu XIXEl then


s e t the MAC s t a t u a s MACJECV;

End
Erid

c a l l ReceivPPacket(pl

20

IEEE INFOCOM 2000

Procedure SEND ( p ) Begin i f ISelertad Ant.Bnnd is i d l o l then

J . Uroch, D. A. Maltz, D. E , Johnson. flu, and I. Jetchcvn. +Aperfoimnnce coriipnnson of ~nulti-hop wirclcss nd hoc network miling prulocals," in ACMAEEE h r . Col$ m n Mobilc Cnriiprrfing nnd Nelrvorkiilji (MfJbiC(u!r9H), OCl. 1998. !I. Chhaya, P e r h i i a n c c cvalunlion OF tjic E E E 802.1 I MAC protocol for wirelcss LANs, Mastcrs thesis, tllinois Institute of Technology,

olse

End

S ~ F C t h e backoft: timer. / * Aft-er the timer expires, X M I T l )

called * I

Chicago, 1996. S . C o r m and J. Mnckcr, Mobile ad hoc networking (MANIX): Routing protocol pcrformimcc issucs and evaluntion coiisidcrationa ([ntci-netDmlt), Mubiie Ad-hnc Mefwd {MANKY) IVorkiiig Group, IElF, Oct.
1948.

Procedure XMICtPktl Begin i f ( P k t i s o f control type


Begin
C a s e , (Subtype o f

except o r RTS)

chen

K . Pall a i d K. Vnmdhan, ns Notes and Dncunicntntiun, August 1998.


Weh sitc at htt~i://www-t~~asI~.cs.bc:rk~lcy.cduli~s/. M . Holnolfcr nnd 13. Plnwimnn, Directed anicnrias in the Inohilc brondband sysreii~,io Pruu. (ifiEIX INFUCOM 96, p p 704-712, lW6. P. Knrn, MACA - A new channcl access inethod for Iiackct radio. iii Proc. (fARRUUHIjL Anirriear Rnrlin Yrlr C i q w e r Neer\uJ,rkir!gSmfere rice, Scpteniber 1940. Y,-U. Ko nnd N.H , Vaidyn, Medirltn access control protocols using tlirectiunel nntcnnas in nd hoc nctworks,Tech.Rep. 99-010, CS Ilept.. Texas A&M University, May 1999. Y.-U. Ku ntid N. I. Voirlyn, Lncation-aided ruiiliiig (LAR) i n muhilc ad hoc Iictworks, in ACM/llXl< the 4111Anrruul iiirl. Confererrcr IJII Mobile Compirrihji nrid Neiivorkirrg (MobiCom 98).October 1998, C. LIIIr!nd C, Lcung, A slolrcd ALOHA packet rarlin nctworkswilh r n d tiple antcniias and 1-eceivcrs: 1rm s[i ctiiiiis on vehicu lo r /euhn doji y , vol. 39, nu. 3. pp. 218-226, I G . T, Okamuto, Smrrr-!A!hwrn Sys/en$s * r i d Wirekss /AN%. Klurver Acadernjc Publishers. 1998. N Pronios, Perfurmancc considenitions Cor slotted spread-spectrun~ ran. doni access nelworkr with direcliunnl antcniins,i n PTUL.. q I E E E GLORECCJM 89,Nov. 1989 C.Sakr and 1 .lbtld, Cnnicr-scnse protucwls Tor packct-switched sinart nntrnna bnsestntiwn, in PfUC. oJlCNIP7, 1997, F. Shad, T. Torld, V. Kezys. and I. Litva, lnilour SDMA capacity using a s[nm antcnnn basestetion, it1 Proc. trT/EfiE ICUPC57, pp. 868-872,
1997,

nr!qln

the P k t )

CTS :
if

(All clie ant-cnnas are noc i d l e ) chen discerdtpkt); return ;

set FIRC c r a n s m i s s i o i i Atat-e 4s MAKCTS. s @ t all antennas as busy. s e t the cimeout according t n che 802.11 Spec. break; ACK: set t h e MAC transmission state as MA-ACK; sell the selcceed a n t c M a a5 Busy; set tho timeout according CO the 8 0 2 . 1 1 Spec.
Rncl I case * / THANSMIT(pkC); / + End

hyak:

transmit. a packet to a channel * /

[I I ]

r h s e i f I P k l is o f RIS t . w e ) then . nayin i c a n send n T S on1.y when t h e snncified * / ;* a n t n n n a is i d l r i * / G ~ E iiic sele,ctc!ci Antenna; I - the a n t e n n a that. HTS

(121
1131

will be t ransm it t e d * / i f (r;oiected Antenna j s no^ i d l e ) then clack o f f the cont-ention window; Set t h e nilckoff timer return ;

[ 141

% ! h e i f lPkt is of DArAl thon Begin i f (Selscced Ant-rxma is not. i d l e i then bnck o f f che contcntion window; s e t khr? hackoff L i n i e r . rearm i
set the ~ r a n : ; m i s s i o ~ s t a l e a s mnC-SEND: sec t h o selected anLenna as busv: s e t Lhe timeout. agpropriacely. TKliNSMIT (pkc 1
CIld

T W S M I T ( p k t ): End

sot the MAC VransmisGion s t a t e a8 MACCRTS. set the tinieaut a s sgeciEiecl in b O 2 . 1 1 Spec.

I151

End
Promilure Receive-Packet l 1

n e p
If

(the P a c k e t i s an R T S ) then

Begin 11 ( t h e ills is for c h i s node) then


send CTS nacket.

1 % omnidjraation - 1

[16j V,!nhargavnn, A,Dctiicrs, SShcnkcr, nlbtl L.Zhang. MACAW - A mctlia acccss protocol for wiiclcss LANs. in Plnc. fflACM SII;COMM94, Scptemhcr 14911. 117) J. Word and R. Coniptm, Improviiig the pcrrnrInsnccofsluited ALOHA packet radio network with nn adaptive may, JEEE Tfirrrraciioiis 011 (Iiinr. /rrinricaliurts,vol. 40, pp. 292-300. Feb. 1992. [IS] I. Ward and ti,Coinpton, High thruiiyhp~ sloticd R W 1 [ A pnckct radio networks with nrliiptivc arrays: J.WE X~rnsuciir~ns on ~ ~ r r ~ m u a i i . r r / i o n s , voI.41,pp.4~0-47a,Mnr.1393. [191 T.-S Yum nnd K,-W.Hung. Design algnrirhms lor m i l isiworks with multiplc dircctionnl antelinns stations: f or! cor~triiuniciitiuns,vol, 40. no. 1 I , pp. 1716-1724, 1892. [ZOl J . Znndcr, Slotted ALOHA ~uultil~op packet nirlio nctworks wilh dirccx tiunnl nniciilias: ~ie~lmlrii..~Iailerr, vol 26, no. 2.5, I99U.

hegir!

iL (tta

else End

,swa

CE f o r t . h i s riade & & MAC statu i s MAC-RTS) then

Data packeL.

I = :;olecced d i r e c t i o n * I

updace t h e NnV f o r t h i s antenna and s ~ t it a s WJSY

elr;o i f

Begin
~f
0 1

(thi. Packet is Daca) than

(the packet is Ear this nnde) then send ACK p a c k e t into snlcctecl direction; * I pass the packet up to t h e link-layer.
C D

l _ l l

disc3rd t.tie Packot.

reset. t h e c o n t o n t i o n ?window; s L x L t h v b a c k c f f timer: l i r r t MAP s r a r e s 2 s I D L E ,


End

r e w m ~transmission of ttle next p a c k e t .

REFERENCES
[I] "lows Statr lhivcrsity GPS page. Web site at lutp ://wwwciide.iastnte.~da/gpr.litinl. 121 Wiwlcss LAN incdiwn access control (MAC) find physical layer (PlIY) spccilicatiuns. 1917. h f t Slantlard IAEE HO2.11, PX02.I I f l l l : The editors of 1Et:F. 802.1 I.

0-7803-~880-jl00iSl0.00( c ) 2000 IEEE

21

IEEE INFOCOM 2000

You might also like