Distributed Correlative Power Control Schemes for Mobile \u003cemphasis emphasistype=\"italic\"\u003eAd hoc\u003c/emphasis\u003e Networks Using Directional Antennas

$Page 1: Distributed Correlative Power Control Schemes for Mobile \u003cemphasis emphasistype=\"italic\"\u003eAd hoc\u003c/emphasis\u003e Networks Using Directional Antennas$
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 3, MAY 2008 1733

Distributed Correlative Power Control Schemes forMobile Ad hoc Networks Using Directional Antennas

Basel Alawieh, Student Member, IEEE, Chadi M. Assi, Member, IEEE, and Wessam Ajib, Member, IEEE

Abstract—Medium access control (MAC) protocols simulta-neously integrating transmission power (TP) control with direc-tional antennas have the potential to enhance both energy savingsand capacity throughput in wireless multihop Mobile Ad hoc NET-works (MANETs). In this paper, we present a model to calculatefuture interference in networks with directional antennas, andbased on this model, we derive some relations that should existbetween the required TP of RTS, CTS, DATA, and ACK framesfor successful data packet delivery in MANETs based on thedirectional version of the IEEE 802.11 distributed coordinationfunction. From these relations, we propose a distributed powercontrol scheme. Furthermore, we show, via simulations, that thetrue potentials from the proposed control scheme cannot be showndue to the imperfection of the derived model. Based on theseobservations, we introduce another class of power control algo-rithm that instead deploys a prediction filter (Kalman or extendedKalman) to estimate future interference. Simulation experimentsfor different topologies are used to verify the significant through-put and energy gains that can be obtained by the proposed powercontrol schemes.

Index Terms—Directional antenna, interference, Kalman filter,Mobile Ad hoc NETwork (MANET), power control.

I. INTRODUCTION

THE RAPID evolution of mobile Internet technology hasprovided incentives for building efficient multihop ad hoc

networks. Recent research activities have focused on the designof better physical layers, development of efficient mediumaccess control (MAC) protocols, and use of directional anten-nas. Due to the great potential that directional antennas haveshown in the cellular wireless area, it is expected that usingdirectional antennas in a multihop wireless local area network(WLAN) environment could also lead to better performancein terms of higher data rates, reduced interference, and energyconsumption. However, to take full advantage of these potentialbenefits, efficient MAC protocols that are directional antennafriendly need to be designed.

A. IEEE 802.11 and Power Control

The most popular MAC for WLAN is IEEE 802.11 [1],which is based on Carrier Sense Multiple Access with

Manuscript received March 1, 2007; revised June 24, 2007 and August 18,2007. The review of this paper was coordinated by Dr. E. Hossain.

B. Alawieh and C. M. Assi are with Concordia University, Montréal,QC H3G 1M8, Canada (e-mail: [email protected]; [email protected]).

W. Ajib is with the Department of Computer Sciences, Université du Québecà Montréal, Montréal, QC H3C 3P8, Canada (e-mail: [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TVT.2007.909264

Fig. 1. Power control and directional antenna merits. (a) Power control merits.(b) Directional antenna merits.

Collision Avoidance using omnidirectional antenna. Request-to-send/clear-to-send (RTS/CTS) control messages are addedas extensions to prevent packet collisions and solve the hiddenterminal problem. In particular, the RTS/CTS control messagesare used to reserve a transmission floor for data messagetransmission. Hence, nodes lying in the vicinity of the omni-directional transmission that hear the RTS or the CTS messagedefer their transmissions until the ongoing communication iscomplete. In IEEE 802.11, all packets (RTS/CTS/DATA/ACK)are sent with maximum power. It has been shown that thiskind of handshake communication decreases spatial reuse and,thus, decreases capacity throughput and additionally yieldsunnecessary energy consumption [2]. To illustrate, considerthe scenario in Fig. 1(a), where node A uses its maximumtransmission power (TP) to send packets to node B, and nodes Dand E will try to initiate a communication with nodes C and F,

0018-9545/$25.00 © 2008 IEEE

1734 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 3, MAY 2008

respectively. Nodes C and F will not reply with CTS messagesto the RTS messages sent by nodes D and E since they lie in thetransmission vicinity of nodes A and B, respectively. However,it is easy to show that the three transmissions A → B, D → C,and E → F can overlap in time if nodes are able to appro-priately select their TPs, as shown in Fig. 1(a), consequentlyincreasing the network throughput and possibly reducing theoverall energy consumption. Adjusting the transmitted poweris extremely important in ad hoc networks due to three majorreasons.

1) The transmitted power of the mobile node determinesthe network topology. The network topology, in turn, hasconsiderable impact on the throughput performance of thenetwork [14].

2) Mobile nodes in ad hoc networks are usually energyconstrained; hence, they have to be as energy efficientas possible. Thus, the power control should adjust thetransmitted power to be the least power that is requiredto send the data packet to meet the required signal-to-interference-plus-noise ratio (SINR) threshold.

3) Transmitting at high power can degrade other ongoingtransmissions and, in the meantime, can prevent futuretransmission, as illustrated in Fig. 1(a). Thus, reduc-ing the TP will reduce the interference on other nodecommunications and may enhance the overall networkthroughput as well as energy consumption.

B. Directional Antenna Issues

Directional antennas offer clear advantages for improvingthe network capacity by increasing the potential for spatialreuse [14]. Allowing antennas to direct their transmissions inthe direction of the intended receiver clearly reduces the levelof contention with other nodes, thereby allowing for more si-multaneous transmissions. Moreover, directional antennas canincrease the signaling range without spending extra power (asopposed to omnidirectional antennas), and accordingly, somereceivers outside the omnidirectional range may be reached ina one-hop transmission. This longer range results in a smallernumber of hops on end-to-end paths, yielding an increase inconnection throughput. To elaborate more on the merits ofdirectional antennas, we again consider the scenarios shownin Fig. 1(b). Nodes A and B, nodes C and D, and nodes Eand F will not be able to simultaneously communicate withomnidirectional operation, whereas, with a directional antenna,this can be done; thus, spatial reuse is enhanced. Furthermore,node E cannot reach node A with an omnidirectional antenna,whereas, by using a directional antenna, node E reaches node A.

Since a directional antenna has a higher gain, a transmitterusing a directional antenna requires a lower amount of powerto transmit the same distance as would be needed with anomnidirectional antenna. Hence, transmitting nodes can con-serve power by adequately reducing the power when usingdirectional transmissions. If node A uses an omnidirectionalantenna, the power sent by A toward node B will be uniformlyradiated in all directions at the same time instance, resultingin a circular transmission/reception pattern. This will result in

an unnecessary waste of power. If node A transmits using adirectional antenna, the power will be radiated in a specificdirection toward node B and thus achieves much little energyconsumption. For a given transmission distance, the powerrequired by a transmitter using a directional antenna is pro-portional to its beamwidth. This implies that node A using adirectional antenna with π/6 beam-width will approximatelyrequire one sixth the amount of power that an omnidirectionalantenna needs to transmit to node B.

Operational problems are associated with directional antennaimplementation. Major operational problems [5] are the deaf-ness problem, the hidden terminal problem, and the exposed ter-minal problem. Recently, directional MAC (D-MAC) schemes[6] have been proposed to solve such operational problems. Inthese protocols, deafness is either solved by omnidirectionallysending the RTS message or/and CTS message using an omni-directional or a directional antenna [9], [11]. Another solutionfor eliminating the deafness phenomena can be elaborated byusing a separate channel for control packets or by using a busytone signal or a combination of both [6]. More details aboutthese problems can be found in [14], and a list of solutions forthese problems is presented in [6] and [9].

Moreover, with the increase of the number of concurrentcommunications caused by the use of a directional antenna,the overall interference impact may increase. This problem isclassified as a mutual interference problem [6]. Additionally,a higher gain of directional transmissions leads to more in-terferences at nodes receiving in the omnidirectional mode.Interference, in turn, will increase the probability of packetdelivery error since it decreases the signal-to-interference ratio(SIR), and this may have a strong adverse effect on the overallnetwork throughput and energy gains. This motivates the use ofpower control with a directional antenna [14]. The integrationof a directional antenna with the TP control scheme is shown togive more benefits than anticipated [14]. This is the main topicof this paper.

C. Contributions

We propose distributed correlative power control schemesusing interference estimation techniques with a directionalantenna. We extend an interference model [15] that takes intoaccount the deafness, hidden terminal, and sidelobe effects as-sociated with directional transmission. Using this interferencemodel, we analytically study the power relations that existbetween the directional IEEE 802.11 four-way handshaking forsuccessful packet delivery by achieving a target SIR at both thetransmitter and the receiver. Based on these correlations, a dis-tributed power control (DPC) scheme is proposed. Furthermoreand from the simulation results, we show that the true potentialsfrom the proposed control scheme cannot be shown due to theimperfection of the derived model. From these observations,we introduce an enhanced correlative power control scheme de-ploying prediction filters (Kalman or extended Kalman filter).Prediction filters are shown to achieve a more accurate future-interference estimation.

The organization of the rest of this paper is given as follows:In Section II, we discuss the related work and the motivation

ALAWIEH et al.: DISTRIBUTED CORRELATIVE POWER CONTROL FOR MANETs USING DIRECTIONAL ANTENNA 1735

for the proposed research. The power control schemes usinga directional antenna are presented in Section III. We carryvarious simulations for different topologies in Section IV todemonstrate the significant throughput and energy gains thatcan be obtained under the proposed protocols. Finally, wepresent our conclusions and future work in Section V.

II. RELATED WORK

One of the first modifications of the distributed coordinationfunction that aims at directional antennas was proposed in theD-MAC protocol [10]. It is a rather straightforward extension ofthe IEEE 802.11 protocol. Here, RTS, CTS, DATA, and ACKpackets are directionally sent. Alternatively, RTS packets areomnidirectionally sent if none of the directional antennas of thetransmitter are blocked.

Power control with the use of directional antennas for packetradio network was first proposed in [19], where a slottedALOHA packet radio network was considered. Zander derivedan equation model to calculate the performance improvementthat can be obtained in a slotted ALOHA channel by the use ofdirectional antennas and multiple receivers.

Performance evaluation of a directional antenna with powercontrol was also studied in [14]. The RTS message is sent at apredetermined power, i.e., the maximum power. The receiverwill find the difference between the received power of theRTS message and its threshold power [14]. The value of thedifference is sent within the CTS message. The source nodewill use a power that is equal to the maximum power minus thedifference value.

Fahmy et al. [7], on the other hand, proposed the use ofadaptive antenna arrays. RTS/CTS messages are omnidirection-ally sent with maximum power Pmax, whereas DATA/ACK aredirectionally sent with controlled power that meets the signal-to-noise ratio threshold value. This can be done by using thevalues of the received RTS/CTS power levels to compute howmuch power reduction is required.

Based on the omnidirectional Beginner’s All-purpose Sym-bolic Instruction Code (BASIC) power control protocol, adirectional-antenna-based MAC protocol with power control[13] is proposed. Here, RTS/CTS/DATA/ACK are directionallysent; the RTS and CTS messages are sent with maximumpower, but the data packets are transmitted with power con-trol. Through the RTS–CTS handshake, the power value fortransmitting the data packet is assigned. Moreover, a destinationnode, upon receiving an RTS packet, calculates the differencebetween the values of the SINR of the RTS packet and theSIRmin threshold. This difference value is encapsulated inthe CTS message sent to the source. Based on this value,the source reduces the power value needed for ensuing the datapacket by an amount that is equal to this difference minus amargin of 6 dB, not exceeding the maximum power level of thetransmitter.

A DPC protocol has been introduced for ad hoc nodes withsmart antennas in [8]. In this protocol, the receivers get the localinterference information and send it to the transmitters, whichthe transmitter uses, together with the corresponding minimumSINR, to estimate the power reduction factors for each activated

link. DATA and ACK transmissions are in (beamformed) arraymode since smart antennas are used at both ends of the link.

A directional medium access protocol with power control(DMAP) was presented in [3]. RTS messages are omnidirec-tionally sent; CTS/DATA/ACK messages are directionally sent.The main target of DMAP was to alleviate some of the problemsthat are associated with the use of a directional antenna. More-over, DMAP minimizes the energy consumption by integratingTP control with the use of directional antenna. Separate dataand control (RTS/CTS/ACK) channels were used to rectify thehidden terminal problem due to unheard RTS/CTS messages.In DMAP, a transmitter sends an omnidirectional RTS. Thereceiver, before replying with a directional CTS (D-CTS), willsense the data channel toward the transmitter and measure theinterference. A power control factor is encapsulated within theD-CTS packet for the transmitter to read to assign a power valuefor data packets. Later on, Arora et al. proposed an enhance-ment to DMAP by applying an admission control terminologyfrom cellular networks. They called their new protocol a load-based concurrent access protocol (LCAP) [2]. LCAP aims atincreasing spatial reuse by allowing interference-limited simul-taneous transmissions to take place within the same vicinityby using TP control. The receiver calculates the differencebetween the calculated power value and the minimum powervalue needed to correctly decode the packet and encapsulateit in the CTS packet. This information is used by the nodeshearing the CTS messages to find, in case they have to initiateany communication, the amount of interference that they canput on the receiver.

SIR-based power control schemes with directional antennasuse the current interference measurement. A source node mea-sures the current interference and encapsulates it in either oneof its control or data packets. The receiver node according to apredefined power control scheme will calculate the required TPfor its ensued control or data packet based on this interferencemeasurement. However, the interference may change; thus, thepower assigned to this ensued packet may be insufficient insome cases for successive packet delivery. To the best of ourknowledge, no power control schemes with directional antennasuse the estimation of future interference.

A correlated power control scheme has been proposed fornodes that are equipped with omnidirectional antennas in [20].The interference model, as shown in Fig. 2, was estimated onan assumption that all potential interfering nodes are usingan average TP Pavg and average radius of transmission zoneRavg. Active interfering nodes located outside the transmissionrange of node A (the circle with radius RRTS) are separatedfrom each other by a distance of at least Ravg. Moreover,the density of the simultaneous interfering nodes is upperbounded by a factor of 1/R2

avg; here, the total interference wasderived as

∞∫RRTS,A

2 × π × x × Pavg × dx

R2avg × x4

.

Three scenarios for finding Pavg were presented. In the worstcase scenario, all active nodes generate interference at maximal


Fig. 2. Interference model.

power Pavg = Pmax. In the adaptive scenario, Pavg is esti-mated from the node’s performance, i.e., Pavg increases whenone frame is lost and decreases when N consecutive framesare successfully received. In another adaptive scenario, Pavg

is estimated as a moving average from the network perfor-mance, i.e., Pavg = 0.9 × Pavg + 0.1 × Pt, where Pt is theTP of any captured packet. Zhang and Bensaou reported thatbetter network performance and energy gain can be achievedusing the proposed algorithms. The interference model forthe directional antenna differs since there exist potential andindirect interfering nodes. This is shown in the succeedingsections.

III. CORRELATIVE POWER CONTROL SCHEME

A. MAC and Physical Layer Properties

Our proposed protocol schemes will operate based on sixproperties.

1) The IEEE 802.11 directional physical carrier sensingmechanism is adopted [16]. A transmitter cannot initiateany communication in a specific direction if it receives apower level larger than a given carrier-sensing thresholddenoted by η.

2) The D-MAC protocol is adopted with the so-called direc-tional virtual carrier sensing [16], which is a directionalversion of the IEEE 802.11 network allocation vector(DNAV). Here, a node is not allowed to send any frame ina specific direction if its DNAV in that specific directionis set.

3) Interference power at each time instant can be quicklymeasured but probably with errors at each node. Theinterference power is equal to the difference between thetotal received power and the power of the desired signal.

4) We assume that all nodes in the ad hoc network are ho-mogeneous and that all the radio parameters are the same.

5) The channel loss gain between a pair of nodes can bedetermined. The channel loss gain can be measured asfollows:

Gain =Pr

Pt(1)

Fig. 3. Generic directional antenna interference region.

where Pr is the received power from transmittedpower Pt.

6) A receiver is able to receive and correctly decode a packetif and only if the defined SINR at the receiver side islarger than or equal to a predetermined threshold denotedby ζ; thus, we have the following condition:

Pr ≥ ζ × Pn (2)

where Pn is the sum of the interference power and thethermal noise power. Substituting (1) into (2), we get

Pt ≥ζ × Pn

Gain. (3)

Another necessary condition for a receiver to be able toreceive and correctly decode a packet is that the power ofthe received packet should be equal to or greater than athreshold power level denoted by κ. Thus, the minimumTP is given by

Pmin =κ

Gain. (4)

B. Interference Estimation Using Analytical Models

1) Preliminaries: A generic model of a directional antennafor determining the interference is shown in Fig. 3 [15]. Rdenotes the maximal permission range of node A. R′′ is themaximal range of the sidelobe of node A. R′ is the constraintrange of the sidelobe. θ is the beam width of the main lobe.Here, nodes lying in the area of constrained range and in thearea formed by the intersection of the node’s main beam (thewhite region in Fig. 3) with that of the sidelobe having radius R′

are refrained from transmission in any direction since theirtransmissions may highly affect the ongoing communication.We extend and adapt this model to fit the requirements of ourproposed power control scheme.

Two types of interferers result from the application of direc-tional antennas: 1) the potential interfering nodes and 2) theindirect interfering nodes. All nodes outside the main lobeand outside the sidelobe range (the dotted shaded region inFig. 3) are considered to be potential interferences and mayturn their directional antennas in any direction. All nodes insidethe main beam of A within range R and greater than R′′ or


those inside the sidelobe of A with a range between R′ andR′′ are considered to be indirect interfering nodes (the grayregion in Fig. 3) since they will refrain from transmission inthe direction of node A, and they will not cause any directinterference to node A. These nodes are free to be engaged inany communication toward other directions.2) Directional Interference Model: Let Pt be the TP, Gd

be the gain of the main lobe, and Gs be the gain of the mainsidelobe represented by R′′. The value of Gs is between 0and 1. h is the antenna height. Using the two-way propagationmodel, with the exponential attenuation factor equal to 4, andtransmitted power Pt, the values of R, R′′, and R′ can be foundby [15]

R =(

Pt · G2d · h2

κ

)1/4

(5)

R′′ = R · G1/4s (6)

R′ = R · G1/2s . (7)

Note that potential interferer nodes may turn their antennas toany direction with equal probabilities. As a result, the antennagain of these nodes is a random variable given by GI , i.e.,

GI =(2 · π − θ) · Gs + θ · Gd

2 · π . (8)

The interference power of any interference node is a randomvariable and can be estimated by average value Pavg. We nowproceed to find the total amount of interference as perceivedby node A. Consider the nodes inside the arc-shaped area tobe delimited by the main beam of node A and at distances rand r + dr from node A. Each node in this area is goingto contribute an interfering signal I1(r), whereas each nodeoutside the main beam of node A and at a distance of [r, r + dr]will contribute an interfering signal I2(r). Here, let I1(r) andI2(r) be expressed as follows:

I1(r) =

{Pavg·Gd·GI ·h2

r4 , r > RPavg·Gs·Gd·h2

r4 , R′′ < r < R(9)

I2(r) =

{Pavg·Gs·GI ·h2

r4 , R′′ < rPavg·Gs·Gs·h2

r4 , R′ < r < R′′.(10)

Therefore, the total interference is given by

Itotal = ρ ·

θ ·

R∫

R′′

I1(r) · r dr +

∞∫r

I1(r) · r dr

+

(2 · π − θ) ·

R′′∫R′

I2(r) · r dr +

∞∫R′′

I2(r) · r dr

(11)

where ρ is the uniform active node density determined by thenumber of active nodes in the whole network divided by thearea of distribution of all nodes in the network (in nodes persquare meter) and can be approximated, as will be shown next.

Prior to transmitting a packet, a node can approximate thecommunication activity of its neighbors, i.e., those lying mainlyin the area (π · R2). This is done via the recorded receivedcontrol and data packets in the angle of arrival (AOA) cachetable of every packet not destined to itself from its neighbors.Network nodes that have not transmitted a signal for a whilewill be removed from the AOA cache table; thus, the node willnot consider them in approximating the active node density.Through this approximation method, the node predicts theactivity of its neighbors; if a node checks the AOA table andfinds four recorded entries, the node can foretell that thereare four active nodes that may interfere with its main-lobe orsidelobe transmission and calculates the active node density as4/π · R2. Note that this estimation is not quite accurate sincewe are assuming that the ratio of the total number of activenodes over the total network area should be equal to the activenodes in a node transmission range over the transmission rangearea. Another reason for the inaccuracy of this estimation isthat any neighbor node may become active in the future whilethe ongoing communication is taking place. To account for thissecond reason, we propose safety margin c, i.e., c × ρ. Notethat c is a simulation parameter and may vary, depending on thetopology under study. Therefore, the estimated network densitynaturally accounts for dynamically changing traffic flows andmobility in the network.

Simplifying the total interference given by (11) by using(5)–(10), we obtain

Itotal =ρ · Pavg ·

√κ · (K1 + K2)√Pt

(12)

where

K1 =θ · h ·

(GI +

√Gs − Gs

)2

K2 =(2 · π − θ) · h · Gs ·

(GI√Gs

−√

Gs + 1)

2 · Gd.

Itotal is the total interference estimated; all the variablesare known, except Pavg. Many estimation algorithms proposals[20] can be adopted to find Pavg. The worst-case scenario isto consider Pavg = Pmax. In our model, Pavg is adaptivelydetermined from the node performance. The value of Pavg ateach node starts with an initial value that is equal to Pmax. Here,Pavg is lower bounded by Pmin and upper bounded by Pmax.For every Nframes that a node transmits and consecutivelysuccessfully received at the receiver, Pavg is decreased by afactor of 0.1 × Pavg; otherwise, if one frame is lost, Pavg

is increased by a factor of 0.1 × Pavg. Here, Nframes is asimulation parameter.3) Power Control Using the Interference Model: Using the

derived model, we derive and analyze the relation between RTSand CTS messages and then generalize by induction to four-way handshake. As stated before, a necessary condition forreceiving RTS and CTS messages is{

PrRTS ≥ κPrCTS ≥ κ

(13)


where PrRTS and PrCTS are the received powers of the RTS andCTS messages. Using (1), we get

Gain =PrCTS

PCTS(14)

where PCTS is the power needed to transmit a CTS message.The total noise power estimated at the sender’s side, assumingPthermal = 0, is simply Itotal. By substituting (12) and (14) into(3), we obtain

PCTS ≥ ζ ·(

ρ · Pavg ·√

κ · (K1 + K2)Gain ·

√PRTS

)(15)

where PRTS is the transmitted power of the RTS message.Together, (13) and (15) constitute the sufficient condition forboth the RTS frame and the ensuing CTS frame transmissionto succeed. Similarly, we follow the same procedure to get thetransmitted power of the DATA frame from the CTS frame andthe ACK frame from the DATA frame. Given that the RTSmessage is sent with the maximum power, the TP of the ensuingframes CTS, DATA, and ACK is given by

PCTS = max(

Pmin, ψ · Pavg√PRTS

)(16)

PDATA = max(

Pmin, ψ · Pavg√PCTS

)(17)

PACK = max(

Pmin, ψ · Pavg√PDATA

)(18)

where ψ = ζ · (ρ · √κ · (K1 + K2))/Gain, and Pmin is givenby (4).

C. Interference Power Prediction Using Prediction Filters

1) Preliminaries: The Kalman filter has been recently pro-posed in the literature [12] in different mobile cellular systemsapplications related to power control, such as interference esti-mation and channel gain prediction. A Kalman filter method forpower control is proposed for broadband packet-switched time-division multiple-access (TDMA) wireless networks in [12]. Inthis paper, a terminal starts sending data packets via a TDMAuplink channel to the base station, and upon receiving the firstdata packet in slot n, the base station measures the channelinterference around its area and predicts the future interferenceusing the Kalman filter. Based on the predicted interference,the base station calculates the required optimum power forreceiving the next data packet in slot n + 1. This informationis relayed to the terminal via a downlink channel.

The advantages of the Kalman filter are its simplicity dueto its recursive structure, robustness over a wide range ofparameters and conditions, and the fact that it possibly providesan optimal estimate with minimum square error. These fea-tures and the successful reported application stories in variousresearch fields (such as target tracking and detection, digitalsignal processing, and digital image processing) for the Kalmanfilter have motivated us to apply it to power control in MobileAd hoc NETworks (MANETs).

Fig. 4. Ongoing Kalman filter cycle. The time update projects the current stateestimate ahead in time. The measurement update adjusts the projected estimateby an actual measurement at that time.

The Kalman filter estimates a process by implementing afeedback control form; the filter estimates the process at sometime and then obtains feedback in the form of noisy measure-ment. Thus, Kalman filter prediction equations consist of twotypes: 1) time update equations and 2) measurement updateequations. The time update equations estimate the processa priori value for the next time step by projecting forward intime the current state and error covariance estimates. Moreover,the measurement update equations integrate the new feedbackmeasurement into the a priori estimate to obtain an improveda posteriori estimate. Indeed, the final prediction algorithmshown in Fig. 4 resembles that of a predictor–corrector algo-rithm [18] for solving numerical problems where the time up-date equations are the predictor equations and the measurementequations are the corrector equations.2) Interference Prediction: Let In be the actual

interference-plus-noise power in decibels below 1 mW (dBm)received at time event n. In is to be considered the processstate to be predicted by the Kalman filter. The thermal noisepower, which depends on the channel bandwidth, is givenand fixed. The total interference is simply the thermal noiseplus the measured interference. The system dynamics of theinterference-plus-noise power can be modeled in state-spaceform as

In = In−1 + Nn (19)

where Nn is the variation of the interference-plus-noise poweras new interfering nodes may start to initiate transmissionsand/or adjust their TPs at time event n. According to theKalman filter state-space mode, Nn is the process noise. LetXn be the measured interference-plus-noise power at timeevent n. Then

Xn = In + En (20)

where En is the measurement noise. Equations (19) and (20)are commonly referred to as the state space generation model.The time equations of the Kalman filter in this case are

In+1 = In (21)

Pn+1 = Pn + Qn (22)

where In+1 is the a priori predicted interference at the next timeevent. In is the a posteriori estimate of In. Pn+1 and Pn are


the a priori and a posteriori estimates of the interference-plus-noise error covariance at time events n + 1 and n, respectively.Qn is the covariance of process noise Nn. The measurementupdate equations are

Kn =Pn

Pn + Rn

(23)

In = In + Kn × (Xn − In) (24)

Pn = (1 − Kn) × Pn (25)

where In and In are the a priori and a posteriori estimatesof In, respectively; Pn is the a priori estimate of the errorvariance at time event n; Kn is the Kalman gain; and Rn isthe covariance for measurement noise En.

In the actual tuning operation of the filter, the measurementnoise covariance Rn and Qn can be determined as follows:

Qn =1

M − 1×

M∑n=1

(Xn − Xn)2 (26)

Rn =C × Qn (27)

where Xn is the mean of the last M measured values at timeevent n. Event n is when a node successfully receives a controlor data frame. Xn is the last obtained measured value. C is aconstant between 0 and 1. Qn is an estimate of the varianceof the sum of the process and measurement noise, becausemeasurements Xn include the fluctuation of both interferenceand measurement errors.

A necessary condition for the Kalman filter to operate is thatprocess noise Nn should have a normal distribution [18]. Theradio channel model considered in this paper includes two-raypath loss, antenna gain, and shadowing. Shadowing is a lognor-mal distributed random variable caused by terrain features. Thereceived signal power at any node can be formulated as

Pr = Pt × r−4 × G2 × h2 × 10�/10 (28)

where r is the distance between the two nodes, and h is theheight of the antenna. G is the antenna gain of the nodes and isconsidered to be identical for all nodes. Pr is the received powerfrom transmitting power Pt. Note that � is the shadowing com-ponent, which is characterized by a Gaussian random variablewith zero mean and a standard deviation of σ dB. This makesNn in dBm normally distributed, which verifies the use of theKalman filter in the prediction of interference in the MANETenvironment.

The convergence properties of the Kalman filter is dependenton the values of the variance denoted by P [18]. We will showby simulation that P is within limits and has the convergenceshape at the end of the performance evaluation section. TheKalman filter algorithm functions as follows: For event n, theinterference measurements are input to (26) and (27) to estimateQn and Rn. Using these values and the current measurementXn in (23)–(25), we get Kalman gain Kn and the posterioriestimates for In and Pn, respectively. The a priori estimatesfor the next time event are given by (20) and (22). Specifically,In+1 in (22) is used as the predicted interference-plus-noisepower for power control, as will be discussed in the next

Fig. 5. Power control using the Kalman filter.

section; note that Pn+1 is used as an initial value to get thenext predicted value.

The extended Kalman filter (EKF) [18] attempts to correctthe error induced between the process and measured values,and it is mainly used for nonlinear systems. The process inevaluation has been modeled as a linear system; thus, a smallenhancement can be added using the EKF scheme. This en-hancement is incorporated within time equations of the EKF.Thus, the EKF time update equation is given by

In+1 = In + (In − In−1). (29)

3) Power Control Using Prediction Filters: Before initiat-ing a transmission, node A measures the interference at timeinstance t, and the interference-plus-noise power is used asinput to the Kalman filter or EKF to predict the estimatedinterference-plus-noise power I at a future time, as discussed inthe previous section. Without loss of generality, the I notationwill be used as the predicted interference-plus-noise power (inmilliwatts) in the coming formulas and for all nodes. The RTSmessage, which is sent at maximum power, carries interferenceinformation I to node B, as shown in Fig. 5. Upon receivingthe RTS message, node B uses this value I to calculate therequired power of CTS as shown here and according to the basicrule stated in (4). Equations (16)–(18) can be reformulated asfollows: The TP of the CTS message is given by

PCTS = max(

Pmin, ζ × I

Gain

). (30)

Before sending the CTS message, node B measures theinterference around its transmission zone and then predicts theinterference-plus-noise power I+ in the future for node A to beable to successfully assign a power value for its data packet.This predicted value is sent to node A in the CTS message.When node A receives the CTS message, it repeats the sameprocedure taken by node B to assign a suitable power value forthe DATA frame. The TP of the DATA message is given by

PDATA = max(

Pmin, ζ × I+

Gain

). (31)

Node B receiving the DATA frame will assign a power valueto the ACK frame as follows:

PACK = max(

Pmin, ζ × I++

Gain

)(32)

where I++ is the predicted interference-plus-noise power atnode A upon receiving the CTS message and is sent to node Bin the DATA message.


TABLE ISIMULATION PARAMETER SETTINGS

IV. PERFORMANCE EVALUATION

A. Simulation Setup

We use Qualnet [17] to evaluate by simulation the per-formance of our proposed power control schemes. We com-pare the proposed power control schemes with IEEE 802.11b,D-MAC schemes, and directional BASIC (D-BASIC) [13]. Ourcomparison with D-BASIC would establish the virtues of theproposed schemes. The power control scheme using the inter-ference model for prediction is termed as adaptive. Moreover,we denote the power control scheme using the Kalman filteras Adaptive-K and that using the EKF as Adaptive-EK. Thechannel rate is 11 Mb/s, and the constant bit rate (CBR) andfile transfer protocol traffic are presented in our study. Twonetwork topologies are adopted in our simulations; one is a10 × 10 grid network with ten end-to-end CBR or transportcontrol protocol (TCP) flows. The flows start from the leftmostnode to the rightmost node along the same row. The route to thedestination node is determined via a routing algorithm, and inour simulations, we used ad hoc on-demand vector routing. Thedistance between each node pair is 100 m. The other scenariois a 50-node uniform random network with ten CBR or TCPflows, and all nodes are inside a square measuring 1000 m ×1000 m. The packet size is 512 B, and the packet-sending ratefor CBR is 400 packet/s.

We are considering the following six scenarios:1) grid network with CBR flows;2) grid network with TCP flows;3) static random network with CBR flows;4) static random network with TCP flows;5) dynamic random network with CBR flows;6) dynamic random network with TCP flows.Moreover, the c for each simulation topology is determined

to be 1.2, 1.2, 1.3, 1.3, 1.2, and 1.2, respectively. Other simu-lation parameters are shown in Table I. In each scenario, onenode may directly communicate with another node or use amultihop route, depending on the transmit power. We use threemetrics to evaluate IEEE 802.11, D-MAC, D-BASIC, Adaptive,Adaptive-K, and Adaptive-EK: 1) aggregate throughput, whichis the sum of the data frames correctly received by the receiversper time unit; 2) effective data delivered per joule, which isthe received effective data frames divided by the entire energyconsumption; and 3) data frame corruption ratio, which isthe portion of the MAC layer frames corrupted by interferingnodes.

Fig. 6. Network end-to-end throughput.

B. Results and Analysis

1) Average Throughput: Fig. 6 shows the total networkend-to-end throughput for different MAC protocols, i.e.,IEEE 802.11, D-MAC, D-BASIC, Adaptive, Adaptive-K, andAdaptive-EK. Clearly, the throughput of adaptive schemes ishigher than that of others. As can be seen in scenarios 1 and2, IEEE 802.11 suffers from channel access problems, whichmake it inefficient in terms of throughput. This is because nodeswithin the transmission zone of the sender or the receiver arerefrained from initiating any transmission for the duration of theongoing communication between the sender and the receivernodes. Thus, as traffic load increases, the duration to win thechannel decreases. As a result, packets will be dropped, becausetheir transmission retry limit threshold is reached. The trafficload of 400 packet/s is considered to be high; thus, the IEEE802.11 starts to show its limitations in sharing the channel inthe time domain.

D-MAC and D-BASIC outperform IEEE 802.11 due totheir directivity gains and the lack of deafness events in thegrid topology. D-MAC reduces the number of blocked nodes;thus, spatial reuse increases, and hence, throughput increases.D-BASIC effectively avoids interference, reduces the con-tention between nodes, and reduces the number of blockednodes. This is the reason D-BASIC outperforms D-MAC interms of throughput in most of the scenarios. Nevertheless,D-BASIC suffers from the hidden node problem, which clearlyshows its effect in scenario 5 (Fig. 6).

The proposed adaptive scheme can detect the active nodedensity and, based on this estimation, assigns consecutivepower to frames, which differentiate it from others in terms ofachieving better throughput gains. In the case of a lack of de-tected activity within the sender’s region, packets are transmit-ted with a sufficient minimal power Pmin for correct reception.This, in turn, decreases interference and accordingly enhancesspatial reuse, which results in a better throughput gain. Theadaptive method uses an interference model to estimate theinterference, whereas tuned prediction filters predict this inter-ference. Fig. 7 shows an intuitive deviation comparison for thefirst 40 s between these two techniques for random node 28in scenario 3 and node 42 in scenario 6. As shown, prediction


Fig. 7. Interference error percentage.

Fig. 8. Average interference error percentage.

filters are able to follow the measured value with less deviation.It is to be noted here that deviation is defined as the measuredvalue minus the estimated or predicted values. Fig. 8 showsthe average error, considering all the scenarios for interferenceestimation using the Kalman filter, EKF, and the model-basedtechniques. An interference estimation value that is higher thanthe measured actual interference with an error that is more than10% may highly affect the overall performance of the network.This is because nodes with a higher interference estimationvalue may increase their TPs to overcome this high interfer-ence; thus, the higher the TP, the more the interference effecton other nodes. As a result, this may highly affect the overallnetwork performance, which is shown in Fig. 6. Moreover,nodes with a lower interference estimate value with an errorthat is greater than or equal to 10% will decrease their packetTP. In such a case, there is a likelihood that these packetsmay be corrupted along the path due to actual interference thatnodes were not aware of when they assigned power values totheir consecutive packets. In conclusion, tuned prediction filtersoutperform the model based on an average error enhancementof 12.4%, which was the reason behind the prediction filtersachieving higher throughput gain.

Fig. 9. Interference prediction under variable traffic load.

A slight improvement in terms of CBR end-to-end through-put for the IEEE 802.11 is reported in scenario 3. The random-ness of nodes enhances the performance of IEEE 802.11 due tothe fact that there exist cases where fewer nodes may lie in thevicinity of the transmission range of the sender. On the otherhand, D-BASIC transmits data packets at low TP; there is highprobability for data reception failure, and accordingly, this as-pect decreases network throughput, as shown in scenarios 3–6.When the load is high, the interference is large. Thus, D-MACand D-BASIC protocols are not able to use all their antennabeams due to the mutual interference problem. The adaptiveschemes are shown to better perform in scenarios 3–6 since theyresolve mutual interference problem.2) Prediction Filters: To test the effectiveness of predic-

tion filters, we consider scenario 5 and try to vary the inter-ference fluctuation by adding additional flows between twosimulation instances. Initially, ten CBR flows were running.We injected ten additional CBR flows between 5 and 200 sto increase the interference. A snapshot for the same ran-dom node 28 is reported in Fig. 9. The predicted value inFig. 9 is simply estimated at the previous event for the nextevent. Thus, this figure shows the estimated value predictedat event t for the next event Vs, i.e., the value measured attime instance t + 1. As can be seen, the measured interferenceincreases between these two time instances. The prediction


Fig. 10. Ratio of throughput by average energy consumption per node.

filters were able to accurately follow the actual measuredinterference.3) Throughput/Energy: Fig. 10 depicts the ratio of the

throughput to the average energy consumption per node inkilobytes per second per joule. Energy consumption includesthe energy of a successful transmission of packets, the lostenergy in retransmitting a packet in case of collisions, andthe energy of the node while receiving a packet and when itis in idle state. The power savings are attributed to the gainof the directional antennas and to the correct assignment ofpower values for the adaptive schemes. Reduction in the mutualinterference makes it feasible for nodes to efficiently deliverpackets with less energy consumption.4) Mobility: We study the impact of mobility on the average

throughput for scenario 5. By increasing the mobility, thesource and receiver nodes may not be able to communicatewith each other due to the reason that either one of them willbe out of range of the other. This may trigger link failuresthat may frequently occur due to the disconnection of adjacentnodes in a route. Route table entries thus may get stale due tonode mobility and may require updating. This will add morecongestion on the network. This is why, for all case stud-ies (IEEE 802.11, D-MAC, D-BASIC, Adaptive, Adaptive-K,and Adaptive-EK), the throughput starts to decrease as themobility increases. As shown in Fig. 11, the percentage de-crease in throughput as mobility increases in the case wherenodes are equipped with directional antennas is much less thanthat in the omnidirectional transmission. This is due to thedirectional transmission properties (directivity gains—higherrange of communication), which may keep a link between twonodes stronger if one of the nodes is moving in the same direc-tion as the directional antenna. This is, however, not the casewith omnidirectional transmission, where the range is limitedand is usually much shorter. However, directional transmissionmay suffer from more frequent link breakages resulting fromnodes that are moving outside the main lobe beamwidth area.This explains the fact that, with directional transmission, thethroughput decrease starts at a lower mobility compared withthat in the case of omnidirectional transmission, as shownin Fig. 11.

Fig. 11. Scenario 5 CBR end-to-end throughput (in kilobytes per second).

Fig. 12. Data frame corruption ratio.

5) Data Frame Corruption Factor: Fig. 12 shows the dataframe corruption ratio in all scenarios. Power control usingadaptive schemes causes less corruption than the others. Thereason behind this aspect is that all packets with the IEEE802.11 are transmitted with maximum power; thus, the inter-ference increases, and this results in more packet corruption.D-MAC suffers from deafness and mutual interference. Thiscauses higher data corruption. D-BASIC and Adaptive schemesalso suffer from deafness since both are based on D-MAC.Nevertheless, power control integrated with their operationdecreases the mutual interference, thus achieving higher packetdelivery rate. The effectiveness of the power assignment in thepower control scheme adopting prediction filters is intuitivelyshown to decrease the number of packets dropped.6) Deafness: In all the schemes considered, deafness

played a major role in decreasing the anticipated throughput.D-MAC and D-BASIC suffer from this problem, and this couldbe viewed mainly in the random topology (i.e., cases 3–6).Deafness has not been completely rectified with the proposedpower control scheme since, as mentioned before, the schemewas built on D-MAC. Nevertheless, we verify that the efficientassignment of the D-CTS has decreased, to a high extent,the deafness consequences. To verify our proposition, let us


Fig. 13. Probability of RTS retransmission due to timeout.

Fig. 14. Convergence of the Kalman filter.

take a look at Fig. 13, which shows the probability of RTSretransmission due to timeout in scenario 5. The probabilityof RTS retransmission is simply the ratio of the RTS packetsretransmitted due to timeout to the total number of RTS packetssent throughout the simulation time. Deafness has a majoreffect on the RTS transmission parameter. IEEE 802.11 hasthe least value since it does not suffer from this problem.D-MAC and D-BASIC have nearly the same values since bothof them send the CTS packet at a fixed power value. Theadaptive scheme sends the RTS packet at a fixed power and thenadapts the value of the TP of the CTS packet to fit the node’sactivity around the sender. If the node activity density is high,the power assigned to the CTS packet transmission will give itthe potential to reach most of the active nodes. Consequently,this results in a decrease in, but not the elimination of, thedeafness phenomena. A suggested solution for deafness is toadd a busy tone channel, which is out of the scope of this paper.7) Kalman Filter Convergence: Fig. 14 shows the conver-

gence of the error of interference-plus-noise power covarianceP estimates for a random node 28 in scenario 5 under normaltraffic conditions for a simulation time of 500 s; P is within avalue between 0 and 1 and has a convergence shape at the endof the simulation time. This result shows that the Kalman filterhas successfully operated and has maintained the convergenceshape. We verify the Kalman convergence conditions [4], [18]

with other nodes from other topologies, and almost the sameconvergence shape has been depicted.

V. CONCLUSION

In this paper, we have proposed a power control schemefor the D-MAC protocol, which requires the use of a singlechannel for the transmission and for the reception of bothcontrol and data packets. We have derived the temporal TPcorrelations that exist between the D-MAC protocol four-wayhandshake packets (RTS/CTS/DATA/ACK) for successful com-munication, taking into account directional operational accessproblems, such as hidden terminal problems, deafness, andsidelobe interference. Based on the node activity density, aninterference model was estimated, and together with the corre-lations derived, we were able to induce efficient constraints thatensure the correct delivery of each individual frame in this four-way handshake. Moreover, we introduced the Kalman filter andEKF as another solution to estimate the future interference. Itis shown that the prediction filters outperform the model-basedtechniques by an average error of 14.3%.

It has been shown, by simulation, that the proposed powercontrol schemes are efficient in terms of throughput and en-ergy consumption. We have compared the performances of ourproposed schemes with the performances of MANETs usingdifferent MAC protocols, such as the standard IEEE 802.11bMAC protocol, the D-MAC protocol with no power control,and the D-MAC protocol with D-BASIC power control scheme.Our simulation results showed that the correlated power controlschemes are improving the throughput compared to D-MAC by48.6%, on average. The proposed schemes outperform IEEE802.11 by a factor of 78.1%. At the same time, a 74.2%reduction in energy consumed over IEEE 802.11 is achievedby the correlated schemes.

REFERENCES

[1] IEEE 802.11 Wireless LAN Media Access Control (MAC) and PhysicalLayer (PHY) Specifications, 1999. IEEE 802.11 b.

[2] A. Arora and M. Krunz, “Interference-limited MAC protocol forMANETs with directional antennas,” in Proc. IEEE Int. Symp. WoWMoM,Taormina, Italy, Jun. 13–16, 2005, pp. 2–10.

[3] A. Arora, M. Krunz, and A. Muqattash, “Directional mediumaccess protocol DMAP with power control for wireless Ad Hoc net-works,” in Proc. IEEE GLOBECOM, Dallas, TX, Nov. 29–Dec. 3, 2004,pp. 2797–2801.

[4] R. Brown and P. Hwang, Introduction to Random Signals and AppliedKalman Filtering. New York: Wiley, 1997.

[5] R. Choudhury and N. H. Vaidya, “Impact of directional antennas onAd Hoc routing,” in Proc. Conf. Pers. Wireless Commun., Venice, Italy,Sep. 23–25, 2003, pp. 590–600.

[6] H. Dai, K.-W. Ng, and M.-Y. Wu, “An overview of MAC protocolswith directional antennas in wireless Ad Hoc networks,” in Proc. Int.Conf. Wireless Mobile Commun., Bucharest, Romania, Jul. 29–31, 2006,pp. 84–90.

[7] N. S. Fahmy, T. D. Todd, and V. Kezys, “Ad Hoc networks withsmart antennas using IEEE 802.11-based protocols,” in Proc. IEEE ICC,New York, Apr. 28–May 2 2002, pp. 3144–3148.

[8] N. S. Fahmy, T. D. Todd, and V. Kezys, “Distributed power control forAd Hoc networks with smart antennas,” in Proc. IEEE VTC, Vancouver,BC, Canada, Sep. 24–28, 2002, pp. 2141–2144.

[9] H. Gossain, C. Cordeiro, and D. P. Agrawal, “Minimizing the effect ofdeafness and hidden terminal problem in wireless Ad Hoc networks usingdirectional antennas,” Wireless Commun. Mobile Comput.—Special Issueon Wireless Ad hoc Networks: Technologies and Challenges, vol. 6, no. 7,pp. 917–931, Sep. 2006.


[10] Y. B. Ko, V. Shankarkumar, and N. H. Vaidya, “Medium access con-trol protocols using directional antennas in Ad Hoc networks,” in Proc.INFOCOM, Tel-Aviv, Israel, Mar. 26–30, 2000, pp. 620–626.

[11] T. Korakis, G. Jakllari, and L. Tassiulas, “A MAC protocol for full ex-ploitation of directional antennas in Ad Hoc wireless networks,” in Proc.ACM Int. Symp. MobiHoc, Annapolis, MD, Jun. 1–3, 2003, pp. 98–107.

[12] K. K. Leung, “Power control by interference prediction for broadbandwireless packet networks,” IEEE Trans. Wireless Commun., vol. 1, no. 2,pp. 256–265, Mar. 2002.

[13] A. Nasipuri, K. Li, and U. Sappidi, “Power consumption and throughputin mobile Ad Hoc networks using directional antennas,” in Proc. Int. Conf.Comput. Commun. Netw., Miami, FL, Oct. 14–16, 2002, pp. 620–626.

[14] R. Ramanathan, “On the performance of Ad Hoc networks with beam-forming antennas,” in Proc. IEEE Int. Symp. MobiHoc, Long Beach, CA,Oct. 4/5, 2001, pp. 95–105.

[15] A. Spyropoulos and S. Raghavendra, “Capacity bounds for Ad Hocnetworks using directional antennas,” in Proc. IEEE ICC, Seattle, WA,May 28–30, 2003, pp. 348–352.

[16] M. Takai, J. Martin, R. Bagrodia, and A. Ren, “Directional virtual car-rier sensing for directional antennas in mobile Ad Hoc networks,” inProc. ACM Int. Symp. MobiHoc, Lausanne, Switzerland, Jun. 9–11, 2002,pp. 183–193.

[17] Qualnet Simulator. [Online]. Available: www.scalablenetworks.com[18] G. Welch and G. Bishop, “An introduction to the Kalman filter,” Dept.

Comput. Sci., Univ. North Carolina Chapel Hill, Chapel Hill, NC, Tech.Rep. TR95-041, Aug. 2006.

[19] J. Zander, “Slotted ALOHA multihop packet radio networks with di-rectional antennas,” Electron. Lett., vol. 26, no. 25, pp. 2098–2100,Sep. 1990.

[20] J. Zhang and B. Bensaou, “Core-PC: A class of correlative power controlalgorithms for single channel mobile Ad Hoc networks,” IEEE Trans.Wireless Commun., vol. 6, no. 9, pp. 3410–3417, Sep. 2007.

Basel Alawieh (S’00) received the B.E and M.Edegrees from the American University of Beirut,Beirut, Lebanon, in 1997 and 2000, respectively.He is currently working toward the Ph.D. degreewith the Department of Electrical and ComputerEngineering, Concordia University, Montreal, QC,Canada.

From July 2004 to August 2005, he was a Re-search Associate with Ottawa University, SITE,Ottawa, ON, Canada, where he worked on next-generation test beds and MPLS. His research inter-

ests are mobile ad hoc networks.

Chadi M. Assi (M’03) received the B.Eng. degreefrom Lebanese University, Beirut, Lebanon, in 1997and the Ph.D. degree from the Graduate Center, CityUniversity of New York, NY, in April 2003.

He was a Visiting Scientist for one year withNokia Research Center, Boston, MA, working onquality of service in optical access networks. InAugust 2003, he joined the Concordia Institute forInformation Systems Engineering, Concordia Uni-versity, Montreal, QC, Canada, as an Assistant Pro-fessor, where he is currently an Associate Professor.

His current research interests are optical networks, wireless and ad hoc net-works, and security.

Dr. Assi was the recipient of the prestigious Mina Rees Dissertation Awardfrom the City University of New York in August 2002 for his research onwavelength-division-multiplexing optical networks.

Wessam Ajib (M’05) received the EngineerDiploma in physical instruments from the InstitutNational Polytechnique de Grenoble, ÉcoleNationale Supérieure de Physique de Grenoble,Grenoble, France, in 1996 and the Diplôme d’ÉtudesApprofondies degree in digital communicationsystems and the Ph.D. degree in computer sciencesand computer networks from the École NationaleSupérieure des Télécommunications, Paris, France,in 1997 and 2000, respectively.

From October 2000 to June 2004, he was withNortel Networks, Ottawa, ON, Canada, as an Architect and Radio NetworkDesigner. He had conducted many projects and introduced different innovativesolutions for the UMTS system. From June 2004 to June 2005, he was aPostdoctoral Fellow with the Department of Electrical Engineering, ÉcolePolytechnique de Montréal, Montréal, QC, Canada. Since June 2005, he hadbeen with the Department of Computer Sciences, Université du Québec àMontréal, where he is currently an Assistant Professor of computer networks.

Documents

Distributed Correlative Power Control Schemes for Mobile \u003cemphasis emphasistype=\"italic\"\u003eAd hoc\u003c/emphasis\u003e Networks Using Directional Antennas