A MAC protocol based on Adaptive Beamforming for Ad Hoc Networks

Transcription

1 A MAC protool based on Adaptive Beamforming for Ad Ho Netorks Harkirat Singh and Suresh Singh Department of Computer Siene Portland State University Abstrat This paper presents a novel slotted MAC (Medium Aess Control) protool for nodes equipped ith adaptive antenna array in ad ho netork. The protool relies on the ability of antenna to uses DOA (Diretion-Of-Arrival) information to beamform by plaing nulls in the diretion of interferers thus maximize SINR (Signal to Interferene and noise ratio) at the reeiver. We studied the performane of the protool using joint simulation in OPNET and Matlab. We studied the impat of variable number of antenna elements, DOA algorithm, and nulling. The performane of our ne protool is ompared against one of the reent diretional MAC protools [5]. We observe that despite the simpliity of our protool it ahieves high throughput. I. INTRODUCTION Reently, there has been inreasing interest in developing MAC protools for use in ad ho netorks here nodes are equipped ith diretional antennas. Antenna models used inlude setored fixed beam antennas, idealized adaptive array antennas, and steerable diretional antennas. As previous researhers have shon, using diretional antennas inreases throughput beause of better spatial reuse of the spetrum (see [], [5], [2], [3]). Hoever, e note that these previous orks have not fully exploited the benefits of adaptive array antennas (or smart antennas) suh as the ability to form nulls in the diretion of interferers (resulting in high SINR) and the ability to determine the diretion of transmitters (Diretion of Arrival). We sho that by exploiting these apabilities of smart antennas, a simple protool an yield throughputs that are 2x 4x higher than one of the reent protools [5]. We also note that our simulations use realisti antenna models unlike the idealized models used in many (ith the exeption of [2]) papers and, despite this, our protool out performs most of these existing protools. There are to key apabilities of antenna arrays that e exploit in developing DOA-MAC: the ability to form direted beams and plae nulls in given diretions, and the ability to determine the diretion of arrival of signals from multiple transmitters. In DOA-MAC, a small initial portion of the slot is used for finding the diretion of various transmitters (all of hih transmit diretionally). This is done by requiring eah transmitter to transmit a pure tone (no soure and destination id) toards its intended reeiver for a short interval prior to This ork is funded by the NSF under grant ANIR The tone format is a ombination of the DSSS PLCP Preamble and PLCP header as given in IEEE 82.. transmitting the paket. The reeiver runs a DOA algorithm hih provides information about the reeived signal strength and diretion of the different transmitters. This information is then used at eah reeiver to guide beamforming (beam and nulls) for the remaining duration of the slot. Upon orret paket reeption, a reeiver sends an ACK using the already formed beams. Our ork here differs from all of the above papers in the folloing ays: our adaptive array antenna model is made up of a linear array of antenna elements and e exploit DOA information as ell as the nulling apability of the antenna to maximize SINR at the reeiver. This gives us the ability to develop a simple protool (DOA-MAC) that performs very ell. The remainder of the paper is organized as follos: in the next setion e desribe our system model and provide a brief overvie of adaptive array antennas. Setion III desribes related ork. Setion IV illustrates our protool DOA-MAC in more detail. Setion V presents results of the simulation. II. SYSTEM MODEL We assume that eah node is equipped ith an adaptive array antenna system hih is omposed of a linear array of elements. For simpliity, e assume that the antenna array is perpendiular to the x-y plane in hih the nodes lie. The reason for this assumption is that the beam formed by the antenna is symmetri about the antenna axis and is thus independent of the diretion in hih a node is faing. Figure provides a shemati of an adaptive array antenna system. As illustrated in the figure, the antenna onsists of antenna elements separated from eah other by a knon distane. We an assume that a transmitter is loated far enough aay from the reeiver that all the signals arriving at the different antenna elements are parallel. Hoever, sine the elements are separated by distane, the phase of the different signals is different. Let denote the phase and gain that is added to eah signal. Then, the output sent to the reeiver, an be ritten as, "!$#&%(' *)&+-,./ here is the phase propagation fator, 8 is the avelength, and is an arbitrary gain onstant. The eights

2 used in this paper only shift the phase of the signal and leave the amplitude untouhed. The representation for the eights is,! %(' *) / For a more omprehensive disussion, please see [4]. As e noted in the introdution, a benefiial feature of adaptive array antennas is the ability of these antennas to form nulls in given diretions. In fat, given elements, an antenna an form upto nulls. Hoever, the shape of the desired beam an hange depending on the number of and the diretion of the nulls. Figure 2 illustrates to ases hen using antenna elements ith being the desired diretion. In the first ase, e are forming only to nulls hereas in the seond ase e are forming six nulls. As an be seen, the shape of the beam and side lobes hanges. In this ork e are using MMSE (Minimum Mean Square Error) algorithm to determine eights to form nulls appropriately [4]. We implemented the adaptive array antenna model in MAT- LAB and interfaed it ith the physial layer of OPNET. In our study e use realisti antenna patterns ith the side lobes. 5 Desired = 45 deg, Nulls = 25, 7 deg Desired = 45 deg, Nulls =, 2, 3, 7, 8, 9 deg Signal reeived from transmitter at eah antenna element S Antenna elements Variable gain and phase shifters 8 S 2 S 3 θ d Σ Reeiver 27 S M Fig. 2. Antenna patters ith 8 antenna elements and 2 or 6 nulls. M Fig.. Shemati of a adaptive array antenna. III. RELATED WORK [6] fouses on design the design of Smart antennas for mobile devies ith operating frequeny of 2 Ghz. The authors examine the impat of the antenna design on netork throughput and the impat of mutual oupling 2 on the performane of adaptive algorithms. The paper presents results of detailed OPNET simulations using a TDMA version of the 82. protool as the MAC layer. [7] presents a sheduling-based MAC protools for nodes equipped ith diretional antennas. The diretional antenna onsidered is a multi-beam adaptive array antenna (MBAA) hih is apable of forming multiple beams. The key ontribution of the paper is the development of a neighbor traking sheme that is then used to shedule transmissions by eah node in a distributed ay. [] proposes a MAC protool here nodes are equipped ith diretional antenna elements. Eah of the antenna elements has a onial pattern, spanning an angle of Mutual oupling results in radiation patterns that have shalloer and shifted nulls, and less aurate AOA, thus deteriorating overall netork throughput. radians. The antennas at eah node are fixed ith nonoverlapping beam diretions so as to olletively span the entire plane. The MAC protool is assumed to be apable of sithing any one or all the antennas to ative or passive modes. The paper did not examine the benefits of nulling or the impat of side-lobe interferene. Furthermore, the propagation model as rather simplisti beause of the assumption of omplete attenuation outside the onial pattern. [5] studies MAC protools for Ad ho netork for nodes equipped ith diretional antennas here the main beam is modeled as a one and the sidelobes as a sphere. They develop a novel multi-hop RTS to establish links beteen distant nodes and ) ( diretional gain) is assumed to be higher than (omni diretional gain). The diretion in hih the main lobe is to be oriented is determined by the MAC protool (hih in turn is provided this information by the netork layer hih is assumed to be neighbor-aare). Beamidth of the antenna is assumed to be a onstant. They sho that their protool has a 4-5x throughput as ompared ith 82.. In [2] a node in promisuous mode ahes AOA information based on signals reeived and uses this information for sending RTS. A irular antenna ith 6 elements is assumed, and a node is apable of eletronially steering the boresight toards a speifi diretion. A onstant beamidth of 45 deg assumed.

3 Hoever, it as observed that as the boresight hanges the side lobe pattern hanges drastially. They also observe that using realisti antenna patterns as opposed to a ideal patterns results in a 36% degradation of throughput. There have been several other papers that look at the benefits of using smart antennas in ellular environments see, for instane, [8], [9], [], [], [2]. These papers look at models here the base station is equipped ith one or multiple adaptive antenna arrays. b a o 3 o MUSIC Spatial Spetrum a, b Reeiver 3 o IV. PROTOCOL DESCRIPTION: DOA-MAC In this setion e desribe the behavior of our protool. Hoever, before doing this, e need to make the folloing assumptions: () We assume that nodes are aare of the angular loation of eah of their neighbors (as in [5]) sine this information is needed at transmitters to form direted beams toards reeivers; and (2) For simpliity, e assume that all nodes use the same onstant transmit poer. Consider the ase hen a node needs to transmit a paket to node hih is its one-hop neighbor. Sine knos s angular diretion it an form a direted beam toards. Hoever, in order to maximize SINR at, needs to form a beam toards and form nulls toards all other transmitters. In order to do this, needs to kno to things first, that is attempting to transmit to it, and seond, the angular diretion of all the other transmitters that interfere at. Our protool is based on the slotted ALOHA model ith the addition of a omponent that enables reeivers to form beams and nulls as desribed. Eah slot in DOA-MAC is broken into three minislots. The protool then orks as follos: ) The first minislot in a slot is alled the DOA-minislot and it is here that a node identifies the angular diretion of all transmitters that it an hear. All transmitters transmit a simple tone (i.e., a sine ave) during the DOA-minislot toards their intended reeivers. The signal reeived at some reeiver is thus the omplex sum of all of these tones. The reeiver runs a DOA algorithm to determine the angular diretion of eah of the transmitters and the reeived poer from eah transmitter. There are several different DOA algorithms that an be used and the primary differene beteen them is fidelity versus omputational omplexity. For this ork e hose to use MUSIC (MUltiple SIgnal Classifiation) [4] hih lies somehere in beteen all the other algorithms in terms of omplexity and fidelity. Figure 3 shos an example of running MUSIC at the reeiver hen there are three transmitters (all using the same transmit poer). As e an see, nodes and are lose to one another in angle.r.t. the reeiver hereas is quite distint. The output of the MUSIC algorithm shos that the reeiver is unable to distinguish beteen and beause they are lose in angle. This an result in a higher SINR at the reeiver beause the reeiver s beam ould inlude both transmitters. Poer (in db) Angle (deg) Fig. 3. Example of omputing DOA using MUSIC. 2) One a reeiver determines the DOA of all transmitters it an hear, it forms its direted beam toards the one that has the maximum poer and forms nulls in all the other identified diretions. 3) The seond (and largest) minislot is the paket transmission slot and it is here that the pakets are transmitted. After the reeiver has formed its beam and nulls as desribed above, it reeives the paket from the transmitter. After reeiving the paket, it looks at the header and rejets the paket if it as not the intended destination. An example of this happening is illustrated in Figure 4 here e see that nodes and are transmitting to nodes and respetively. Hoever, node inorretly hooses to reeive s transmission beause that transmission is stronger! Fig. 4. a a has a paket for b d b has a paket for d Node d mistakenly forms a beam toards a beause a s signal is stronger than b s signal at d An example of inorret beamforming leading to a rejeted paket. 4) The last minislot is the ACK slot here the reeiver transmits an ACK using the already formed beam to the sender (if the paket as not rejeted and orretly reeived). In Figure 4 node ill not send an ACK to node or to node beause it did not reeive s paket but rather mistakenly reeived s paket. 5) When a transmitter does not reeive an ACK, it retransmits the paket at a later time (as in slotted ALOHA).

4 TABLE I OPNET SIMULATION PARAMETERS. Simulation Parameters Bakground Noise + ambient Noise -43 db Propagation model Free spae Bandidth, khz Min frequeny 2.4 GHz Data Rate 2 kbps Paket Size 52 bytes Paket Generation CBR Carrier Sensing Threshold +3dB Minimum SINR 9 db Bit Error Based on BPSK Modulation urve Maximum radio range 25 m V. PERFORMANCE STUDY In our simulation study, e examined three questions: Does inreasing the number of antenna elements improve throughput? What is the impat of using a realisti DOA algorithm (MUSIC) as opposed to an optimal algorithm ith an arbitrary resolution (i.e., in Figure 3 it ould orretly disriminate beteen nodes a and b)? Does nulling have any benefits? The simulation parameters e seleted are displayed in table I. We evaluate the performane of DOA-ALOHA using a 5x5 mesh (as used in [5]) ith four pre-defined flos. Figure 5 shos the netork topology and flos used for to of these experiments. For the third experiment, e used a random node plaement on the grid here a node s position is shifted in the x-axis and y-axis by adding a displaement randomly seleted from [-5m, +5m] and the flos are as in Figure 5(b). The traffi is CBR (Constant Bit Rate) hih inreases (per flo) from 75kbps to 2Mbps. The paket size is 52 bytes. In order to examine the impat of the number of antenna elements on throughput, e plot the aggregate throughput as a funtion of data rate of one flo, for the ase shon in Figure 5(a), in Figure 6. We plot the same data for the random topology ase in Figure 7 as ell. As e an see, using 6 antenna elements as opposed to 8 does improve throughput in both ases. This result in not surprising beause larger number of antenna elements results in narroer beams and hene better spatial reuse. Interestingly, the throughput is higher for the random topology ase hen ompared ith Figure 5(a). This is beause, in Figure 5(a), the flos are aligned and need to share bandidth at the seond hop hereas in the random topology ase, there is greater potential for spatial reuse sine flos are not aligned. In order to determine the impat of using MUSIC instead of optimal DOA and to anser the question about the benefits of nulling, e fous on the ase shon in Figure 5(b). Figure 8 plots the aggregate throughput as a funtion of data rate of a single onnetion for the ase hen e have 6 antenna elements (Figure 9 does the same hen e use 8 antenna elements). We observe that using 6 antenna elements as opposed to 8 elements makes a big differene in aggregate throughput. (a) Four flos (some alignment) (b) Randomly seleted flos Fig. 5. 5x5 grid topology used to ompare performane ith [5]. This is beause the beamidth hen using 6 elements is smaller than hen using 8 elements hih results in more simultaneous transmissions/slot. For the flos in Figure 5(a), (hen flos are aligned), e did not notie muh differene in the performane of 6 and 8 antenna elements but for Figure 5(b) and for random topologies e do see a signifiant differene. The reason is that hen flos are not aligned, there is a greater potential for spatial reuse ith 6 antenna elements (due to its smaller beamidth) elements 6 elements Fig. 6. Performane of our protool ith optimal DOA in 5(a). We notie that MUSIC results in poor performane (approx. 8% redution in throughput ith 8 elemenst and 5% redution ith 6 elements) in omparison to optimal DOA, this is for the reason as explained in Figure 3 and 4 that MUSIC fails to distinguish beteen to nodes loated in lose proximity. A high-resolution algorithm ill be able to disriminate beteen these to nodes, hoever, it ill require more training sequenes and its omputation ost ill be high. We are urrently studying the trade-off beteen DOA algorithm omplexity and its omputation ost. Finally, e observe that Nulling has greater impat on 8 elements (improvement of approx. %) than 6 elements (improvement of approx. 5%). Table II summarizes our results and ompares them ith [5]. We provide the maximum throughput hen using the optimal DOA as ell as hen using MUSIC. We observe that our protool is 2x 3x better hen e use 8 elements and is muh better (3x 4x) for 6 elements. We note that the beamidth

5 elements 6 elements 5 Optimal DOA MUSIC No Nulling Fig. 7. Performane of our protool ith optimal DOA in random grid topologies Optimal DOA MUSIC No Nulling Fig. 8. Performane of our protool ith 6 elements in 5(b). used in [5] is. In our ase, the linear array reates to symmetri beams and e define beamidth for our protool as the sum of these to beams. VI. CONCLUSION In this paper e have presented DOA-MAC, a slotted MAC that uses DOA information at the reeiver to beamform in a ay that maximizes SINR. We notie that by exploiting the benefits of smart antenna a simple protool like DOA-MAC an ahieve very high throughput. We studied the impat of using a realisti DOA algorithm as ell as the benefits of nulling. Finally, e ompare the performane of our protool against [5] and sho that our protool has a throughput of 2x 4x higher than the [5]. REFERENCES [] J. Y. A. Nasipuri, S. Ye and R. Hiromoto, A ma protool for mobile ad ho netorks using diretional antennas, in IEEE WCNC, 2. [2] R. B. Mineo Takai, Jay Martin and A. Ren, Diretional virtual arrier sensing for diretional antennas in mobile ad ho netorks, in ACM/SIGMOBILE MobiHo 22, Ot 22. Fig. 9. Performane of our protool ith 8 elements in 5(b). TABLE II Mesh Figure 5(a) ( ) ( ) Our Protool (optimal DOA) 25kbps 22 [5] 8kbps Mesh Figure 5(b) Our Protool (optimal DOA) 3kbps 265 Our Protool (MUSIC) 294kbps 9 [5] kbps Random Mesh Our Protool (optimal DOA) 43kbps 37 [5] kbps [3] V. S. Y.B. Ko and N. Vaidya, Medium aess ontrol protools using diretional antennas in ad ho netorks, in IEEE INFOCOM 2, Marh 2. [4] J. C. Liberti and T. S. Rappaport, Smart Antennas for Wireless Communiations. Prentie Hall, 999. [5] R. R. N. H. V. Romit Roy Choudhury, Xue Yang, Using diretional antennas for medium aess ontrol in ad ho netorks, in ACM/SIGMOBILE MobiCom 22, Sep 22. [6] S. Bellofiore, J. Foutz, R. Govindarajula, I. Bahei, C. A. Balanis, A. S. Spanias, J. Capone, and T. M. Duman, Smart antenna system analysis, integration, and performane for mobile ad-ho netorks (manets), IEEE Transation on Antennas and Propagation, vol. 5, no. 5, pp , May 22. [7] J. G.-L.-A. Lihun Bao, Transmission sheduling in ad ho netorks ith diretional antennas, in ACM/SIGMOBILE MobiCom 22, Sep 22. [8] P. S. A. Chokalingam, Performane analysis of spae diversity s-aloha ith steerable beam smart antenna arrays, in IEEE GLOBECOM, Nov 2. [9] J. Ward and J. R. T. Compton, Improving the performane of a slotted aloha paket radio netork ith an adaptive array, IEEE Transations on Communiations, vol. 4, no. 2, pp , February 992. [] A. S. Aampora, S. Krishnamurthy, and M. Zorzi, Media aess protools for use ith smart array antennas to enable ireless multimedia appliations, in Proeedings 9th Tyrrhenian Workshop on Digital Communiations, September 997. [] C. Sakr and T. Todd, Carrier-sense protools for paket-sithed smart antenna basestation, in ICNP 97, 997. [2] R. R. A. Chokalingam, Ma layer performane ith steerable multibeam antenna arrays, in IEEE PIMRC 98, 998.