1

Maximizing Transmission Opportunities inWireless Multihop Networks

Jeong-Yoon Lee, Student Member, IEEE, Chansu Yu, Senior Member, IEEE,Kang G. Shin, Fellow, IEEE, and Young-Joo Suh, Member, IEEE

Abstract—Being readily-available in most of 802.11 radios, multirate capability appears to be useful as WiFi networks are getting moreprevalent and crowded. More specifically, it would be helpful in high-density scenarios because inter-node distance is short enoughto employ high data rates. However, communication at high data rates mandates a large number of hops for a given node pair in amultihop network and thus, can easily be depreciated as per-hop overhead at several layers of network protocol is aggregated overthe increased number of hops. This paper presents a novel multihop, multirate adaptation mechanism, called Multihop TransmissionOPportunity (MTOP), that allows a frame to be forwarded a number of hops consecutively to minimize the MAC-layer overhead betweenhops. This seemingly collision-prone nonstop forwarding is proved to be safe via analysis and USRP/GNU Radio-based experiment inthis paper. The idea of MTOP is in clear contrast to the conventional opportunistic transmission mechanism, known as TXOP, where anode transmits multiple frames back-to-back when it gets an opportunity in a single-hop WLAN. We conducted an extensive simulationstudy via OPNET, demonstrating the performance advantage of MTOP under a wide range of network scenarios.

Index Terms—Opportunistic communication, wireless multihop networks, medium access control, data rate adaptation, multiraterouting.

F

1 INTRODUCTION

MULTI-HOP wireless networks pose more impor-tance as we have seen various types of such

networks on the horizon such as wireless sensor networks(WSNs), vehicular ad hoc networks (VANETs), and wirelessmesh networks (WMNs), and more recently, network ofunmanned aerial vehicles (UAVs) [2], [3] and mobile so-cial networks (MSNs) [4], [5]. These emerging multihopnetworks exhibit characteristics that deviate significantlyfrom the traditional ad hoc networks in terms of scale,traffic intensity, node density, and/or speed. For exam-ple, MSN scenarios typically envisaged around crowdspots, where the number of nodes within range couldbe hundreds or thousands [6]. A similar high densityscenario can also been observed in 802.11 deploymentsand WMNs in US cities.

Being readily available in most of 802.11 radios, mul-tirate capability seems to be promising and can effec-tively exploits the short inter-node distance in high-

• This work was presented in part at IEEE INFOCOM 2010 [1].

• Jeong-Yoon Lee is with the Department of Computer Science andEngineering, POSTECH, Pohang, Korea. Email: jylee9@postech.ac.kr.

• Chansu Yu is with the Department of Electrical and ComputerEngineering, Cleveland State University, Cleveland, OH and the Divisionof ITCE, POSTECH, Pohang, Korea. Email: c.yu91@csuohio.edu.

• Kang G. Shin is with the Department of Electrical Engineering andComputer Science, University of Michigan, Ann Arbor, MI. Email:kgshin@umich.edu.

• Young-Joo Suh is with the Department of Computer Science and Engineer-ing, POSTECH, Pohang, Korea and the Division of ITCE, POSTECH,Pohang, Korea. Email: yjsuh@postech.ac.kr.

density networks. However, it is important to observethat performance does not improve linearly as data rateincreases. This is due to the rate-independent overheadat the PHY and MAC layers, which are imposed byindustry standards such as 802.11 [7]. Moreover, thisoverhead becomes more dominant as rate increases be-cause the transmission time of the payload decreasesproportionally (see Section 2.2).

Opportunistic transmission protocols (TXOP) [8]–[11]have been proposed to alleviate the MAC-layer overheadby allowing a node to transmit multiple frames back-to-back when it transmits at high rate. A node is granteda dedicated time duration, which is called TXOPlimit(3264 or 6016µs) in 802.11e [8], promoting time-basedfairness. Figs. 1(a) and 1(b) show the benefit of TXOPin comparison to 802.11. In TXOP, node 0 is allowedto transmit multiple frames consecutively with just ashort gap between frames (SIFS, 10 µs), reducing theMAC overhead. However, TXOP is only applicable toWLANs and may defeat the usual expectation in multi-hop networks because (i) a node may not have multipleframes to transmit back-to-back although it is givenan opportunity, (ii) it can easily overload intermediatenodes in a multihop chain when a predecessor grabsmore transmission opportunities than its successors, and(iii) TXOP’s time-based fairness is not appropriate inmultihop networks as described in Section 2.4.

This paper proposes a novel frame forwarding mecha-nism, called Multihop Transmission OPportunity (MTOP),that extends the idea of TXOP in multirate, multihopnetworks. In MTOP, a frame is forwarded over multiplehops (say, 0 → 1, 1 → 2, 2 → 3, etc.) with a singlechannel contention as shown in Fig. 1(c). This reduces

2

SIFSDIFS & Backoff SIFS

DIFS & Backoff

DATA DATA DATAACK ACK ACK

SIFS

0 1 other node pair other node pair

DIFS & Backoff

(a) 802.11

SIFS SIFS

DATA DATA DATAACK ACK ACK

SIFS

0 1 0 1 0 1

SIFS SIFSDIFS & Backoff

(b) TXOP

SIFS SIFS

DATA DATA DATAACK ACK ACK

SIFS

0 1 1 2 2 3

SIFS SIFSDIFS & Backoff

(c) MTOP

Fig. 1: Communication sequence in 802.11, TXOP and MTOP.(In (a), the MAC overhead such as DIFS and backoff is as highas 95% of total communication time, which is minimized in (b)and (c). In (b), node 0 sends three back-to-back frames. In (c),nodes 0, 1, and 2 relay a frame back-to-back.)

the MAC overhead and at the same time, resolves theabove-mentioned problems of TXOP.

This paper extends our earlier work on MTOP [1],particularly in the following three ways:• First, this paper presents a more accurate analysis on

the maximum number of nonstop forwarding (hi)(Section 4.2).

• Second, several enhancements have been made tothe original MTOP algorithm. For example, in Fig.1(c), node 1’s nonstop forwarding to node 2 servesas an ack to node 0 (implicit acknowledgment), whichis similarly approached in 802.11 PCF (point co-ordination function) [7] (Section 4.4). Another en-hancement is to (nonstop) forward frames in theorder of their arrivals. In Fig. 1(c), nodes 0, 1 and2 transmit/forward a frame consecutively but itis important to note that it does not have to bethe same frame. This improves (space-based) fairnessbut also brings in complications too (Section 4.3).Another major enhancement is to use Srcr [12] asan underlying multirate routing algorithm (Section4.5).

• Third, performance study has been expanded sig-nificantly. OPNET [13] is used instead of ns-2. Cu-mulative interference model has been incorporated,Ricean channel model is additionally considered, traf-fic load is varied in two different ways (the numberof traffic sessions and the packet generation rate),and also a variety of performance measures areused: packet delivery ratio, delay, the number oftransmissions, frame drops, duplicate frames, andso on (Section 5).

A small-scale experiment based on Universal SoftwareRadio Peripheral (USRP) [14] and GNU Radio [15] has beenconducted with low-rate DBPSK (300kbps) and high-rate DQPSK (600kbps) modulation scheme. This showsthat the additional transmission opportunity exists inmultirate communication environment using the new

concept called multirate margin and thus, that nonstopforwarding does not cause additional collisions due tothis margin (Section 5.1). Also, an extensive simulationstudy based on OPNET [13] has been presented (Section5.2 and 5.3). Our evaluation study has indicated thatMTOP outperforms fixed-rate cases (DSR1 and DSR11)as well as Srcr in most of the simulation scenarios tested.Compared to Srcr, MTOP improves the packet deliveryratio by as much as 13.8% and the average end-to-endpacket delay is reduced by 12∼43%. The performancegain is contributed most by a significant reduction inpacket drops (43.7∼150.5% less drops than Srcr) andMAC overhead.

The rest of the paper is organized as follows. Section2 overviews the characteristics of multirate radio anddiscusses performance anomaly in multirate networks.Section 3 analyzes the defer threshold at different datarates, and quantitatively provides the multirate marginvia analysis. Section 4 describes the proposed protocol,MTOP, which is followed by GNU Radio/USRP-basedexperiment and OPNET-based evaluation in Section 5.The paper concludes with Section 6.

2 BACKGROUND AND RELATED WORK

2.1 Multirate support in IEEE 802.11 standards andrate adaptation algorithmsAccording to IEEE 802.11 PHY-layer specifications [7], itsupports 2.4 GHz Direct Sequence Spread Spectrum (DSSS)at the data rate of 1 and 2 Mbps while a later standard,IEEE 802.11b, supports the additional data rate of 5.5and 11 Mbps that trade off interference tolerance forperformance. 802.11a/g supports 6, 9, 12, 18, 24, 36, 48and 54 Mbps.

There have been a number of proposals on multiratealgorithms for 802.11-based WLANs in the literature.Auto-Rate Fallback (ARF) [16] is the first multirate algo-rithm, the basic idea of which is to use a higher rate uponconsecutive successful transmissions and to fall back toa lower rate after a number of consecutive transmissionfailures. Variations of the ARF includes adaptive ARF(AARF) [17], adaptive multi rate retry (AMRR) [17], and es-timated rate fallback (ERF) [18]. In Receiver-Based Auto Rate(RBAR) [19], the receiver estimates the channel qualitybased on the SINR of the received RTS frame, determinesthe best data rate that the transmitter must use and then,informs it by piggybacking in the CTS packet. Oppor-tunistic Auto Rate (OAR) protocol [9] exploits durationsof high-quality channel conditions and sending multipleback-to-back data packets without gaps. This is similarlyapproached in Medium Access Diversity (MAD) [20].

Since this paper concentrates on multihop networks,it is important to discuss rate-aware multihop routingalgorithms in the literature. Typically, they have con-centrated on developing a rate-aware link cost whichis then integrated with a multihop routing algorithmto find the path that minimizes the total cost. Linkcosts used include delay [21], bandwidth distance product

3

(BDiP) [22], medium time metric (MTM) [23], estimatedtransmission time (ETT) [12], weighted cumulative ETT(WCETT) [24], and bandwidth delay product (BDP) [25].Most of previous studies employ proactive routing al-gorithms such as Destination-Sequenced Distance Vector(DSDV) [26]. On other hand, Srcr [12] and MR-LQSR [24]rely on on-demand routing principles borrowed fromDynamic Source Routing (DSR) [27].

2.2 PHY- and MAC-layer overheadsTo understand the PHY and MAC-layer overhead, letti be the transmission time of a PHY frame for a 512-byte payload at rate i and Ti be the time duration ofthe frame sequence at MAC. Note that a PHY frame iscomposed of PLCP preamble/header and the payload,where the former shall be transmitted using the lowestdata rate (1 Mbps) while the latter is transmitted at ahigher rate. Since the PLCP preamble/header are 192bits(192µs), the overall frame size is 4288µs at 1Mbps (t1).Since the payload can be transmitted at higher rates, itbecomes 2240µs (t2), 937µs (t5.5), and 564µs (t11) for 2,5.5 and 11Mbps, respectively.

To estimate Ti, we assume no RTS/CTS exchange andassume a high traffic condition in which every frametransmission contend for medium access by waiting for arandom time chosen within the contention window (CW ).CW is 31∼1023 and slot time is 20µs. The time forcontention on the average is 31×20

3 or 207µs when CW is31 with two contending nodes. Here is the explanationon the denominator, 3. With two contending nodes,they choose random slots within the contention window(620µs). Since the losing station (that chooses the largerslot) uses the remaining backoff time, the sum of thecontention time of the two stations is same as that ofthe losing station, which is two thirds of the contentionwindow. Therefore, the average contention time for boththe winning and losing station is calculated as 1

3 of thecontention window.

Now, the time duration for the frame sequence atdata rate i, Ti, consists of DIFS and contention (tc or50+207µs), Data (ti), SIFS (tSIFS or 10µs) and ACK (tACKor 304µs), i.e.,

Ti = tc + ti + tSIFS + tACK . (1)

It totals 4859µs (T1), 2811µs (T2), 1508µs (T5.5), and1135µs (T11) for 1, 2, 5.5 and 11Mbps, respectively.

Now, per-frame PHY overhead due to PLCP pream-ble/header is 4.5%, 8.6%, 20.5% and 34.0% at 1, 2,5.5 and 11Mbps. The MAC-layer overhead amounts to11.8%, 20.3%, 37.9% and 50.3% for 1, 2, 5.5 and 11Mbps,respectively. Efforts have been made to reduce the PHYand MAC overheads. For example, a later standard802.11b introduces a shorter PLCP preamble (72bits) andallows the PLCP header (48bits) to transmit at 2Mbps forhigh-rate transmission (5.5 and 11Mbps). Also, 802.11a/greduce the MAC overhead by adopting a smaller CW(15∼1023) as well as a smaller slot size (9µs).

On the other hand, the MAC overhead can be reducedby directly reducing the overhead including the backofftime or indirectly reducing it based on collision avoid-ance or collision masking. One of the latter is packetsalvaging at the MAC. Biswas and Morris proposedExtremely Opportunistic Routing (ExOR) [28], in which acollided packet can be saved by intermediate nodes thatcan be effective in wireless environment with abundanttemporary link errors. This is similarly approached in[29]. Direct approaches include Sift [30], which uses acarefully-chosen, non-uniform probability distribution oftransmitting in each slot within the contention. The mostrelevant to our approach is Aggregation with FragmentRetransmission (AFR) [31] as it addresses the problem ofrelatively large overhead at high data rates. That is, itmitigates the overhead by supporting transmissions ofvery large frames and partial retransmissions in the caseof errors.

2.3 Transmission opportunity and performanceanomalyAnother important development in reducing the MACoverhead is Transmission Opportunity (TXOP), which al-lows a node to transmit multiple frames with a singlechannel access. This was originally proposed in 802.11eto improve fairness by granting a node with a lowerchannel access priority a dedicated time duration, whichis called TXOPlimit (3264 or 6016µs) [8].

In fact, the fairness problem and the associated perfor-mance anomaly have been observed by many researchersin the context of multirate WLANs [10], [11], [32]. While802.11 MAC guarantees that each node gets an equalchance of transmitting its frames, it does not necessarilymean that each node gets an equal share of the channel(time) in a multi-rate environment. With TXOP, a low-rate node pair is not impacted significantly in termsof throughput but a high-rate node pair is benefitedsignificantly.

To demonstrate this time-based fairness, consider anexample scenario in Table 1(a) for a mixture of low(A ← B) and high-rate (C → D) communication. Btransmits one 512-byte frame during T1 or 4859µs andC does one during T11 or 1135µs. Assuming that thetwo transmitters get equal chance of medium access, theaggregate throughput is

Two 512B frames

T1 + T11= 1.37Mbps, (2)

which is barely larger than the lower data rate. Moreover,C-D node pair is not fairly treated because it uses only19% of medium time ( T11

T1+T11).

TXOP improves the situation. Let us first compute themaximum allowable number of frames (ki) to transmitconsecutively at data rate i during TXOPlimit. SinceSIFS (tSIFS) replaces DIFS and contention (tc) betweenki consecutive frames, ki can be obtained as follows:

maxki

(ki·Ti−(ki−1)·tc+(ki−1)·tSIFS ≤ TXOPlimit). (3)

4

TABLE 1: Performance anomaly. (In (a), A-B: 1Mbps, 272m, C-D: 11Mbps, 118m. In (b), A-B: 1Mbps, 272m, C-to-D: 11Mbps,118m each hop.)

(a) Single-hop scenario

t t�A B t t-C D

A-B C-D

Throughput (Mbps, 802.11) 0.683 0.683Medium time (%, 802.11) 81 19

Throughput (Mbps, TXOP) 0.527 1.581Medium time (%, TXOP) 63 37

(b) Multihop scenario

t t�A B t t t t- - -C E F D

A-B C-D

Throughput (Mbps, 802.11) 0.595 0.198Medium time (%, 802.11) 69 31

Throughput (Mbps, TXOP) 0.393 0.393Medium time (%, TXOP) 46 54

Throughput (Mbps, MTOP) 0.527 0.527Medium time (%, MTOP) 63 37

When TXOPlimit is 3264µs [8], ki is 1 for 1 and 2Mbps,2 for 5.5Mbps and 3 for 11Mbps. Now, while node Btransmits one frame (k1) during T1, node C transmitsthree frames (k11) consecutively during T ′11 = 3 · T11 −2 · tc + 2 · tSIFS or 2911µs. Therefore, the aggregatethroughput is improved to

Four 512B frames

T1 + T ′11= 2.11Mbps. (4)

More importantly, C-D node pair uses 37% of mediumtime, which is a significant improvement in terms offairness. Please refer to Table 1(a) for summary.

2.4 Multihop anomaly

In a multihop network, the problem becomes more com-plicated due partly to inter-hop interference and rate-hop count tradeoff. Consider an example in Table 1(b),where B wants to talk to A at 1Mbps and C wants to talkto D at 11Mbps with two intermediate nodes, E and F .Note that the communication range at 1 and 11Mbps is272m and 118m, respectively, and the carrier sense rangeis 589m as detailed later in this paper (see Table 2).

Two transmitters (B and C) and two intermediatenodes (E and F ) can sense each other and thus, theywill get an equal chance for medium access as long asthey have packets to transmit. When only B and C areready, they transmit one packet each. When B, C andE are ready, they transmit one each, which is similarlythe case when B, C, E and F are ready. Assuming thatthose cases occur with the same probability and that twosource nodes, B and C, always have packets to transmit,one destination A receives three packets while antherdestination D receives one.

With 802.11, the aggregate end-to-end throughput willbe

Four 512B frames

3 · T1 + 6 · T11= 0.79Mbps, (5)

and C-D node pair occupies 6·T11

3·T1+6·T11or 31% of medium

time, which indicates a serious fairness problem. Here,tc becomes 174µs because there exist four contendingnodes instead of two. Correspondingly, T1 and T11 are4776 and 1052µs, respectively.

Interestingly, TXOP improves fairness in multihop net-works too but does not increase the throughput unlikein single-hop networks. Since C, E and F will transmitthree frames at once during T ′11 each, the aggregate end-to-end throughput will be

Six 512B frames

3 · T1 + 6 · T ′11= 0.79Mbps, (6)

and C-D node pair occupies 6·T ′11

3·T1+6·T ′11

or 54% of mediumtime. In other words, TXOP trades throughput in favorof fairness. More seriously, A-B pair achieves only twothirds of the throughput in comparison to 802.11 inTable 1(b). Unlike in single-hop networks, TXOP greatlyimpacts the low-rate communication.

Therefore, it is necessary to apply the idea of oppor-tunistic transmission in a different manner in multihopenvironment. The multihop opportunistic transmissionalgorithm proposed in this paper allows nodes E andF to forward a frame with no additional contention. Inother words, C, E and F will transmit/forward a frameback-to-back during T ′11. Therefore, the aggregate end-to-end throughput becomes

Two 512B frames

T1 + T ′11= 1.05Mbps, (7)

which is 34% higher than TXOP and at the same timeequally shared by two flows (A-to-B and C-to-D), im-proving the fairness compared to 802.11.

3 MULTIRATE MARGIN

MTOP encourages nonstop forwarding of a frame overmultiple hops as discussed in Introduction. It seems ap-parent that collisions are abundant because intermediaterelay nodes do not appropriately compete for a chanceto use the shared medium. However, when the first nodein the multi-hop chain is given an exclusive right basedon the underlying MAC protocol, it in fact inhibits alarger set of nodes than necessary, which is enough tokeep the next hop communications from interferenceif it is transmitted at “high rates”. The correspondingquantitative measure is called multirate margin in thispaper.

3.1 An Illustrative Example

Before describing the multirate margin in detail, thissubsection presents an illustrative example that explains

5

Charlie DaveAdam Bob

(a) Charlie does not talk because it would bother Bob

Charlie DaveAdam Bob

(b) Charlie talks because it does not bother Bob

CharlieAdam

??Dave

(c) Considering the worst case, Charlie does not talk

Charlie DaveAdam Bob Ed

(d) Knowing Charlie does not talk, Bob exploits the opportunity

Fig. 2: The defer-if-hear-anything principle. (The same, lowdefer threshold is employed, resulting in the multirate margin.Multihop forwarding is drawn in (d).)

the multirate margin and multihop forwarding mecha-nism in the proposed MTOP protocol. Consider the voicecommunication scenario among four persons as in Fig.2: Adam, Bob, Charlie and Dave. Adam wants to talkto Bob and Charlie wants to talk to Dave. They use thesame defer-if-hear-anything principle (like CSMA) and non-negligible inter-message pause (like DIFS and backoff) toavoid collisions.

In Fig. 2(a), when Adam talks to Bob, Charlie wouldnot begin his conversation to Dave because he knows itwould interfere with Adam-Bob’s communication (anal-ogous to low defer threshold at 1Mbps). In Fig. 2(b),Charlie would begin his conversation because he knowsit would not interfere with Adam-Bob’s communication(high defer threshold at 11Mbps). In reality, however,Charlie does not know whom Adam talks to but justoverhears Adam’s voice as shown in Fig. 2(c). Consid-ering the worst-case scenario, Charlie would not beginhis conversation until Adam completes (a lower oneis specified as the defer threshold). Now, here is theinteresting part. In Fig. 2(d), knowing that Charlie wouldnot talk, Bob exploits this opportunity to immediatelyforward the message to Ed. Time is saved because Bobdoes not “pause” between the messages.

3.2 Receive sensitivity and SINR requirementSteps to analyze the multirate margin are as follow:(i) Estimate the communication range (ri) based on thereceive sensitivity at different rates. (ii) Estimate the SIRrequirement using analysis. (iii) Receive sensitivity issubtracted from the SIR requirement for target BER of10−5 to estimate the maximum tolerable interference,which translates to the minimum RI (receiver to inter-ference) distance. (iv) This is added to the communica-tion range to estimate the minimum TI (transmitter tointerferer) distance, which is translated to the required

TABLE 2: Characteristics of an 802.11b multi-rate radio. (Trans-mit power: 15 dBm, indoor radio propagation model with pathloss exponent of 3.3 [36]. Values in the last six rows are fortarget BER of 10−5. )

Data rate (Mbps) 1 2 5.5 11

Receive sensitivity (dBm) -94 -91 -87 -82Range or ri (m) 272 221 167 118

SIR requirement (dB) 2.2 5.2 4.4 7.6Max. interference (dBm) -96.2 -96.2 -91.4 -89.6

Min. RI distance (m) 317 317 227 200Min. TI distance (m) 589 538 394 318

Required defer -105.1 -103.8 -99.3 -96.2threshold (dBm)

Defer threshold (dBm) -105.1

defer threshold at different rates based on the transmitpower and path loss model. (v) Finally, multirate marginis the difference between the defer threshold at 1Mbpsand the required defer threshold at high rates. Table 2summarizes the results. Note that this is not our owndevelopment for the sake of MTOP but is the case inpractice [33].

Step (i): For a successful communication, the receivedsignal power must be higher than the receive sensitivityin the presence of path loss over distance. Table 2 showsthem at four data rates of 2.4 GHz 802.11b radio [34].Indoor path loss model by Marquesse [35] has been usedto derive the “communication range”, i.e. path loss =40.2 + 20 · log10(d) if d ≤ 8m, and 58.5 + 33 · log10(d/8),otherwise.

Step (ii): Moreover, the received signal power must bestrong enough to overcome the influence of noise andinterference from all other simultaneous transmissions,i.e., SINR must be higher than a certain threshold [37]. Ahigher-rate communication requires a higher threshold,which means that it is more subjective to interference.Based on the study in [36], BER calculation for 802.11b1Mbps is as follows:

BER1 = Q(√

11 · SIR), (8)

where Q function is defined as

Q(x) =1√2π

∫ ∞x

e−(t2

2 )dt. (9)

BER calculation for 802.11b 2, 5.5 and 11Mbps are givenas follows:

BER2 = Q(√

5.5 · SIR), (10)

BER5.5 ≤24−1

24 − 1(14 ·Q(

√8 · SIR) +Q(

√16 · SIR)),

(11)

and

BER11 ≤28−1

28 − 1(24 ·Q(

√4 · SIR) + 16 ·Q(

√6 · SIR)+

174 ·Q(√

8 · SIR) + 16 ·Q(√

10 · SIR)+

24 ·Q(√

12 · SIR) +Q(√

16 · SIR)).(12)

6

SIR (dB)-2 0 2 4 6 8 10 12 14

BE

R

1e-91e-81e-71e-61e-51e-41e-31e-21e-11e+0

1Mbps2Mbps5.5Mbps11Mbps

(a) BER versus SIR

Interferer strength at receiver (dBm)-100 -90 -80 -70 -60

BE

R

1e-91e-81e-71e-61e-51e-41e-31e-21e-11e+0

1Mbps2Mbps5.5Mbps11Mbps

(b) BER versus interferer strengthat the receiver

Fig. 3: Multirate margin. (Fig. 3(b) shows the highest tolerableinterference when the signal strength at the receiver is equalto the receive sensitivity.)

Fig. 3(a) shows the BER curve for four different datarates. The “SIR requirement” for target BER of 10−5 isshown in Table 2. Note that we use SIR instead of SINRas in [36] because interference is generally much strongerthan noise [38] and the capacity of multihop networksis determined by the communication robustness in thepresence of co-channel interference.

3.3 Defer threshold and multirate marginStep (iii): Assume that the signal strength at the receiveris equal to the receive sensitivity in Table 2, Fig. 3(b)shows the maximum tolerable interference to meet theSIR requirement of Fig. 3(a). Those for the target BERof 10−5 are shown in Table 2 along with the equivalent“RI distance”. For instance, nodes within 317m froman 1 Mbps “receiver” must not transmit concurrentlyas drawn in Fig. 4(a); otherwise, the T-R communicationwill fail due to the lower SIR than required.

Step (iv): In order to refrain a potential interferer fromtransmitting, defer threshold is employed. In other words,an 802.11 PHY performs Clear Channel Assessment (CCA),which involves declaring the channel busy if it detectsany signal energy above the pre-specified defer threshold[39].

However, since the receiver does not transmit signals,the minimum TI distance from the previous step (or thecorresponding signal threshold) cannot be implemented.Instead, this can only be mandated in practice by sensingthe signal from the transmitter. Therefore, nodes within589m from an 1 Mbps “transmitter” are forbidden fromtransmitting concurrently as in Fig. 4(a), which is ob-tained by adding the communication range (r1) to the RIdistance [40]. This “minimum TI distance” is translatedto the defer threshold by using the indoor path lossmodel and the transmit power of 15dBm.

Step (v): Repeating the steps (i)-(iv) above at differ-ent rates provides different communication range, SIRrequirement, min RI distance, and min TI distance (andcorrespondingly, the required defer threshold) as sum-marized in Table 2. Fig. 4(b) shows the case for 11Mbps.Now, notice that potential interferers are oblivious ofthe data rate that the transmitter use and thus, it isunavoidable to employ the same, lowest threshold [22]

Min. TI(589m)

T R I

Min. RI(317m)

IFR

IF2

2 AB

DF

(a) A routing path consisting of a few, low-rate(1Mbps) links

T R I

Multiratemargin

R ITT RR

DF

Min. TI(318m)

IFR

IF2

IF3

IF42

3 4A

B

(b) A routing path consisting of many, high-rate(11Mbps) links

Fig. 4: Multirate margin and the MTOP mechanism. (Multiratemargin is the difference in the required minimum TI distancebetween low and high rate transmissions. Nonstop forwardingR → 2 will not be successful in (a) because A ∈ IF2 but willbe okay in (b) because A /∈ IF2.)

(like Charlie in Fig. 2(c)). This is -105.1 dBm as in Table2 and the corresponding range is denoted as DF in Fig.4.

Since the required defer threshold at 11Mbps is -96.2dBm, there exists an 8.9 dB margin, which we callmultirate margin in this paper. In other words, “multiratemargin” is defined as the difference between the pre-specified defer threshold (for the lowest-rate transmis-sion) and the required defer threshold for high-ratetransmission.

4 MULTIHOP TRANSMISSION OPPORTUNITY(MTOP)4.1 Multihop nonstop forwardingThe MTOP protocol exploits the above-mentioned mul-tirate margin by allowing a frame to travel a few morehops with a single medium access. In Fig. 4(b), whilenode T transmits a frame to node R, node A is allowedto transmit its own frame to node B because A is outsideof DF , i.e., A /∈ DF . When node R forwards the frameto node 2 based on the MTOP mechanism, it would notbe interfered with because A /∈ IF2. (Here, IFi denotesthe interference range of node i, which is determinedbased on the min RI distance.) This holds true for the

7

next node (A /∈ IF3) but not for the following hop node(A ∈ IF4).

In other words, when node T transmits data frameat 11 Mbps, most of potential interferers for the currentcommunication (T → R) as well as the next two hopcommunications (R→ 2 and 2→ 3) would be inhibited.This is due to the additional 8.9dB margin discussedin the previous section. On the other hand, this is notthe case for communication at low-rate. As shown inFig. 4(a), node R’s (nonstop) forwarding to node 2 willbe interfered with A’s transmission because A ∈ IF2.

We define MTOPlimit as the remaining margin that anode can exploit for successive transmission to the nexthop without an additional contention for medium access.While TXOPlimit is measured in time and is associatedwith a node as discussed in Section 2.3, MTOPlimit ismeasured in dB and is associated with a frame. However,similar to TXOPlimit, MTOPlimit can be translated tothe number of hops or distance, for convenience. Thatis, each node decides that it can make additional MTOPforwarding if the remaining margin is sufficient for thenext hop transmission at the given data rate i, i.e.,

MTOPlimit − ri > Min. RI distancei. (13)

This is based on the conservative assumption that eachtransmission makes the farthest possible progress (ri) atthe specified data rate. The remaining multirate marginis initialized to Min. TI distance1 at the first node ofthe forwarding chain and is updated as MTOPlimit =MTOPlimit − ri.

In MTOP, the updated MTOPlimit is included inthe packet header just like the duration field. Also,the duration field is computed normally hop by hop.Although each node estimates the duration just for itstransmission instead of the whole process comprisingmultiple nonstop forwarding, this does not cause anytrouble like the transmission of fragments in the original802.11 MAC.

To avoid collision from hidden terminal, MTOP op-tionally uses RTS/CTS as in 802.11. However, it is usedonly for the first transmission in MTOP but not for thefollowing nonstop forwarding. In other words, node Tand R exchanges RTS and CTS before transmitting adata packet in Fig. 4(b) but node R nonstop-forwardsthe packet without the virtual carrier sensing. Notethat RTS/CTS helps reduce the collisions but it has notbeen widely used in practice due to the correspondingoverhead. This overhead is not as significant in MTOPbecause more than a half of all communications arenonstop-forwarded in MTOP as observed in Section 5.3.

4.2 Number of multihop forwarding (hi)

When every hop communication is at rate i, the numberof hops (hi) to nonstop-forward is

hi = bMin. TI distance1 −Min. RI distanceiri

c. (14)

Direction

toward

the destinationƟ

0

1r

x

A2

A1

Fig. 5: hi analysis. (hi depends on x, which in turn dependson node density, λ, and transmission range, r.)

According to parameters in Table 2, hi is one for 1Mbpsand 2Mbps, two for 5.5 Mbps, and three for 11 Mbps.

However, the hi calculation is considered pessimisticand could be larger because intermediate nodes (T , R,2, and 3 in Fig. 4(b)) neither lie at a straight line nor atthe edge of the communication range. To better estimatehi, consider Fig. 5, where node 0 transmits to node 1.Node 0 desires to make the farthest progress toward thedestination within its transmission range r. However, itis unable to achieve that far due to the sparsity of nodesin the neighborhood. Here, we analyze the expectedvalue of the progress as a function of node density.

Let node 1 be the next hop node, which makes theprogress of x toward the destination as shown in thefigure. Expected value of x or E[X] can be computed as

E[x] =

∫ r

0

xfX(x)dx, (15)

where fX(x) denotes the probability density function ofx. Since node 1 is chosen because no other node is foundin A1,

fX(x)∆x= p(x ≤ X ≤ x+ ∆x)= Pr{no node in A1} · Pr{at least one node in A2}= e−λA1(1− e−λA2),

(16)assuming that node locations follow Poisson distributionwith node density (λ). Here, A1 = θ · r2 − r · sin θ · x andA2 = 2r · sin θ · ∆x taking into account both the upperand the lower half of the circle in the figure.E[X] is used in place of ri to calculate hi in equation

(14). Our calculation shows that hi becomes 3 for 5.5Mbps and 8 for 11Mbps when node density is 30 nodesin 300×1500 m2 (λ = 6.67× 10−5) and 2 and 4 when thenumber of nodes is 110 (λ = 2.44× 10−3). In practice, λcan be estimated, for example, based on the number ofneighboring nodes.

4.3 Space-based fairnessAnother important issue with MTOP is fairness. Asdiscussed in Sections 2.3 and 2.4, time-based fairnessin 802.11 and TXOP is not appropriate in multihopenvironment. For example, a node that transmits at highrate contributes to the network by reducing the channeloccupancy time but is disadvantaged by making less

8

progress toward its respective destination due to therate-distance tradeoff, creating a possibility of unfair-ness.

This paper introduces space-based fairness, which weclaim is more appropriate in multi-rate, multihiop net-works. Here, the fairness is measured by the quantity of“work” that moves an object (packet) over the distancetoward the destination. Applying it to end-to-end flows,it is considered perfectly fair when, for example, a pairof nodes separated by 100m end-to-end achieves 100packets while another pair separated by 400m achieves25 packets. In other words, γi · disti is the “total work”done on behalf of flow i and is desired to be balancedamong the flows, where γi and disti are the numberof received packets and the end-to-end distance forflow i, respectively. The MTOP mechanism proposed inthis paper facilitates the space-based fairness becausea packet transmitted at high rate is given additionalopportunity to travel further and thus, to achieve thesame work.

One concern regarding fairness is that a node could(nonstop) forward a packet earlier than others in thepacket queue. In MTOP, the node does not forwardthe newly arriving packet when it is given an oppor-tunity. Instead, it puts the packet in the packet queueand attempts to transmit the head of the queue. Someopportunities can be lost when the head of the queue isnot a high-rate packet but fairness at a node is improved.It is also important to note that control frames such asprobe packets always get a higher priority than (normaland nonstop-forwarded) data packets.

4.4 Implicit ACK

As an optimization technique, MTOP allows intermedi-ate nodes to skip an explicit ACK and to use the immedi-ate (non-stop) forwarding of a frame as an implicit ACK.Since the immediate forwarding occurs an SIFS after theprevious transmission, it coincides with an explicit ACKin terms of frame schedule. However, it is clear thatthe last node in the chain of MTOP forwarding musttransmit an explicit ACK. (It is interesting to comparethis to Block Ack mechanism in TXOP [8].)

Two important questions in the implementation of theimplicit ACK are: (i) What if a predecessor does notreceive an implicit ACK although the next node forwardsthe frame? (ii) What if the next node forwards the dataframe at a higher rate than the predecessor can receive?

The latter problem can easily be resolved by using anexplicit ACK in such a case. As for the former, the prede-cessor retransmits the same frame, which is a duplicateto the next node. Such duplicate frames can be filteredout within the intermediate’s MAC based on the originalfunctionality of the 802.11 MAC, called duplicate framefiltering [7]. This algorithm matches the sender addressand the sender-generated sequence control number of anew frame against those of previously received ones. Ifthere is a match, the receiver transmits ACK but ignores

the duplicate frame. According to our simulation studydetailed in the next section, the duplicate frames are lesswith MTOP.

4.5 Multirate routingMTOP uses Srcr, which is the default multirate routingprotocol for MIT Roofnet [12]. It is based on DSR withlink cache and tries to find the shortest route usingDijkstra’s algorithm on its link cache. The quality ofa route is calculated as sum of estimated transmissiontime (ETT) of each link. More specifically, each nodebroadcasts probe packets at every data rate every onesecond (with jitter) [12]. Then, its neighbors count thenumber of probe packets received within the probewindow (e.g., 10 seconds) to estimate the quality of thecorresponding wireless links in terms of ETT. As in DSR,intermediate nodes forward an RREQ to discover a routeto the destination. In Srcr, they forward an RREQ if theroute quality (i.e., sum of link ETTs) is lower than thepreviously identified value.

In our implementation of Srcr and MTOP, the size ofprobe packet is set to the same as normal data packet(e.g., 512B) as in [12] to estimate the ETT for a datapacket correctly. Since the ETT metric favors higher datarates than the traditional hop count metric, both Srcr andMTOP would result in routing paths consisting of morenumber of high-rate links.

5 PERFORMANCE EVALUATION

5.1 Multirate margin via USRP/GNU Radio-based ex-perimentationSince radio propagation and its channel dynamics cannoteasily be captured using analytical or simulation models,we conducted an experimental study to demonstrate themultirate margin based on a small-scale testbed usingUSRP [14] and GNU Radio [15].

The following are the details of the experiment (see[41] for a similar setup). (i) The testbed includes 3USRP systems (version 5b), 3 RFX2400 transceivers (2.3-2.9 GHz) and GNU Radio software (version 3.1.3). (ii)Modulation schemes used are DBPSK (low-rate) andDQPSK (high-rate). (iii) Carrier frequency and band-width we have tested are 2.4835GHz and 300KHz, re-spectively. Therefore, the maximum data rate is 300Kbpsand 600Kbps for DBPSK and DQPSK, respectively. Asmaller bandwidth and data rates are used partly dueto bandwidth constraints imposed by the USRP [42]. (iv)Transmitter amplitude is set to 8,000, which is smallerthan the default value (12,000). This is to make the com-munication range no farther than 300 feet1. (v) Packetsize is 1,500 bytes and 3,300 packets were transmittedfor each experiment.

Our goal is to observe a similar trend as in Table 2,particularly the multirate margin with two data rates

1. This experiment was conducted in the Edgewater Park near Lake Erie inCleveland, OH.

9

Distance (m)0 50 100 150 200

RS

SI

0.05.0e+31.0e+41.5e+42.0e+42.5e+43.0e+43.5e+44.0e+44.5e+45.0e+4

(a) RSSI versus distance

RSSI

0.0 6.0e+3 1.2e+4 1.8e+4

PD

R (

%)

0102030405060708090

100

DBPSKDQPSK

(b) PDR versus RSSI

SIR-5 0 5 10 15

BE

R

1e-6

1e-5

1e-4

1e-3DBPSKDQPSK

(c) BER versus SIR

RSSI1e+2 1e+3 1e+4

BE

R

1e-6

1e-5

1e-4

1e-3DBPSKDQPSK

(d) BER versus interferencestrength (RSSI) at the receiver

Fig. 6: USRP/GNU Radio-based experimental results.

supported by DBPSK and DQPSK modulation schemes.The experiment has been conducted in two phases. First,in order to obtain communication range (ri) with DBPSKand DQPSK, we set up two USRP systems and measuredreceived signal strength indicator (RSSI) versus distanceand packet delivery ratio (PDR) versus RSSI2. Accordingto our experimental results in Figs. 6(a) and 6(b), rifor DBPSK and DQPSK is estimated as 215ft and 150ft,respectively. Note that 90% PDR is used to estimate thecommunication range, which is equivalent to BER of10−5.

Second, in order to obtain the minimum RI distance,we set up three USRPs, a transmitter (T ), a receiver (R)and an interferer (I) on a straight line (T -R-I). The TRdistance is fixed to the communication range, i.e., 215ftand 150ft for DBPSK and DQPSK, respectively. Figs. 6(c)and 6(d) show BER versus SIR and BER versus RSSI(from I to R), which must be compared to Figs. 3(a)and 3(b), respectively. Note that SIR at the receiver iscalculated as RSSI from the sender minus RSSI from theinterferer [44]. Note also that the observed SIR gap in Fig.6(c) is larger than the theoretical gap of 3dB. We believethis is due to the small number of measurements.

According to the experiment results, we observed thatthe low-rate communication (DBPSK) is more robustto interference than high-rate (DQPSK) as similarly ob-served in [42]. Minimum RI distance is estimated as 220ftand 235ft for DBPSK and DQPSK, respectively, and theminimum TI distance for 90% PDR is about 435ft and385ft. Comparing to Table 2, we can conclude that thesame trend and the multirate margin at a high rate (435versus 385 ft) has been observed.

2. Note also that RSSI obtainable from USRP/GNU Radio is “digital RSSI”value, meaning that it is based on the output of the analog-to-digital converter(ADC), which is not the true RF power at the antenna [43].

5.2 Simulation environmentIt is generally understood that the implementation ofCSMA is hard for the current USRP/GNU Radio plat-form due to hardware limitations [45]–[47]. For thisand other practical reasons, the detailed analysis ofMTOP performance is conducted via OPNET [13], whichsimulates node mobility, a realistic physical layer, radionetwork interfaces, and the 802.11 MAC protocol.

We compare 4 different schemes: fixed data rate of 1and 11Mbps with DSR (denoted as DSR1 and DSR11,respectively), Srcr, and MTOP. In fixed rate cases, everydata packet is transmitted at the specified data rate. In asparse network (e.g., 30 nodes in the network), we expectDSR11 suffers most because of the connectivity problem.But it will become advantageous as N increases. Perfor-mance metrics are (i) packet delivery ratio (PDR) and(ii) average packet delay. Since MTOP potentially causesadditional collisions, we also report (iii) total numberof transmissions, (iv) total frame drops, (v) duplicateframes and (vi) mixture of data rates used, all measuredat the MAC layer.

Our evaluation is based on the simulation of 30∼110mobile nodes located in an area of 1500 × 300 m2. Thedata traffic simulated is constant bit rate (CBR) traffic.At default, 30 CBR sessions are simulated at the rate offive 512B packets/second. However, a larger number ofCBR sessions and a higher packet rate are also simulatedto see the impact of traffic intensity on performance.To better understand the adaptive behavior of MTOPunder different channel condition, Ricean channel modelis also simulated. No mobility is assumed to clearlysee the performance improvement due to the nonstopforwarding of MTOP. (Results with mobility are reportedin our earlier work [1].) Simulation time is 900 secondsfor each run and 10 simulation runs are repeated for eachsimulation scenario to obtain more accurate results.

The aforementioned simulation parameters are typicalin many previous studies on mobile ad hoc networksincluding [48] except that the traffic intensity and thenumber of nodes (N ) are higher than usual. The trafficintensity of 30 sessions with 5 packets/second eachcould be overwhelming at 1Mbps but it can be reason-ably handled at 5 or 11Mbps. N is as many as 110 in oursimulation study because it allows more chances to usehigh data rates.

Note that the size of a probe message is set to the sameas normal data packet (e.g., 512B) as described earlier.Since each node transmits one every second at every rate,the corresponding control overhead could be significantwhen N is large. Note that DSR1 and DSR11 do not havesuch overheads.

5.3 Simulation resultsPDR and delay

Fig. 7 compares PDR and average packet delay ofDSR1, DSR11, Srcr, and MTOP. Fig. 7(a) shows the PDRversus N . DSR11 does not function well as shown in

10

0

20

40

60

80

100

Number of Nodes(N)30 50 70 90 110

DSR1DSR11SrcrMTOP

(a) PDR(%)

0123456

Number of Nodes(N)30 50 70 90 110

DSR1DSR11SrcrMTOP

(b) Average packet delay(sec)

Fig. 7: Performance of MTOP in comparison to fixed rate cases(DSR1 and DSR11) and Srcr

the figure, particularly with a small N . This is duemainly to the lack of end-to-end connectivity. However,its performance increases rapidly as N increases. InDSR1, PDR stays almost constant regardless of N .

Srcr in general achieves a better performance thanDSR1 and DSR11 because it uses a combination of allavailable data rates to maximize the network perfor-mance. However, as shown in Fig. 7(a), when N islarger than 90, DSR11 performs better than Srcr. It is notsurprising because of the extra overhead due to probemessages in Srcr (and MTOP). On the other hand, MTOPoutperforms DSR1, DSR11, and Srcr in the entire rangeof N simulated as shown in Fig. 7(a). MTOP carries thesame extra overhead as in Srcr but reduces the MACoverhead (tc and tACK in Section 2). MTOP achieves asmuch as 13.8% higher PDR than Srcr.

Fig. 7(b) shows the average packet delay versus N .DSR1 experiences the largest packet delay because ofits slow packet transmission speed. DSR11 shows thelowest packet delay in the entire range of N . However,it does not represent its true performance because itsPDR is low too, particularly at low node density, andthe computation of the average packet delay does nottake the lost packets into account. MTOP exhibits thelowest packet delay among the rest.

Mixture of data rateIn order to understand how MTOP improves the

network performance, we collect statistics about the datarate used when applying the nonstop forwarding. Fig. 8shows the mixture of data rates for N=30 and N=90.As expected, in Srcr, high rates are used more in high-density network due to the availability of nodes in eachnode’s vicinity.

In comparison, MTOP apparently uses more low-ratetransmission as evident in Fig. 8. However, the combi-nation of the two statistics (“Normal” and “Non-stop”)results in a similar data. Note that nonstop forwardingin MTOP does not use 1Mbps. Comparing Figs. 8(a) and8(b), the mixture of data rate is desirable as more low-rate transmissions are used when network is sparse andvice verse in both Srcr and MTOP.

Frame dropsSince the fixed rate cases (DSR1 and DSR11) do not

possess an adaptive capability and the corresponding

MTOP(Non-stop)

MTOP(Normal)

Srcr

0 10 20 30 40 50 60 70 80 90 100

1Mbps 2Mbps 5.5Mbps 11Mbps

(a) N=30

MTOP(Non-stop)

MTOP(Normal)

Srcr

0 10 20 30 40 50 60 70 80 90 100

1Mbps 2Mbps 5.5Mbps 11Mbps

(b) N=90

Fig. 8: A mixture of data rate used

0

1

2

3×105

Number of Nodes(N)30 50 70 90 110

MTOP(Non-stop)MTOP(Normal)

Srcr

(a) Total transmissions

0

10000

20000

30000

40000

Number of Nodes(N)30 50 70 90 110

MTOP(Non-stop)MTOP(Normal)

Srcr

(b) Total drops

0

5

10

15

20

Number of Nodes(N)30 50 70 90 110

MTOP(Normal)MTOP(Non-stop)

Srcr

(c) Frame drop ratio(%)

0

2000

4000

6000

8000

Number of Nodes(N)30 50 70 90 110

SrcrMTOP

(d) Duplicate packets received

Fig. 9: Frame transmissions and drops in Srcr and MTOP

performance is not competitive, a more detailed perfor-mance measures have been analyzed only for Srcr andMTOP. Fig. 9(a), 9(b), 9(c) and 9(d) compare the totalnumber of transmissions, total frame drops, frame dropratio, and duplicate frames, respectively. They have beenmeasured at the MAC layer to include all the forwardingand retransmissions and thus, can be regarded as theactual traffic load in the network. We made the followingobservations:• First, as in Fig. 9(a), the total number of transmis-

sions is almost constant for both Srcr and MTOPregardless of N . This is because the traffic intensityis the same. When N is small, low data rates suchas 1Mbps and 2Mbps are used more. This results inless hop count for a given source-destination pairand thus, leads to less number of transmissions.

• Second, in MTOP, more than a half of transmissionsare based on nonstop forwarding as shown in Fig.9(a). This observation allows us to estimate the ben-efit of the implicit ACK (immediate ACK) becauseall the non-stop forwarding is used as an implicit

11

ACK to the predecessor node. The same numberof ACK frames has been saved and the networkbandwidth is better utilized for delivering usefuldata. Assuming that a half of data transmissionsare nonstop-forwarded, the benefit of implicit ACKis the reduction of bandwidth usage as much astSIFS+tACK

T1orT11× 1

2 or 3.2∼13.8%.• Third, while the total number of transmissions

is very close between Srcr and MTOP, the latterachieves a higher PDR. This is due to a higher framedrops in Srcr as drawn in Fig. 9(b). Compared toMTOP, it drops 43.7∼150.5% more frames at theMAC layer.

• Fourth, the gap in frame drop can be better ex-plained by investigating the two different framedrops in MTOP. Those with contention (denotedas “Normal” in Fig. 9(b)) increase rapidly with N ,which is similarly observed in Srcr. On the otherhand, those without contention (denoted as “Non-stop” in Fig. 9(b)) are held almost unchanged. Non-stop forwarding is less vulnerable to collisions be-cause it effectively keeps potential interferers silentby taking advantage of the short inter-frame gap asshown in Fig. 1(c).

• Fifth, to see the difference in frame drop moreclearly, Fig. 9(c) shows the frame drop ratio withN . In fact, normal transmissions experience thesimilar drop ratio in Srcr (4.9∼16.8%) and MTOP(4.0∼12.8%), but nonstop forwarding achieves a flatdrop ratio (2.9∼3.9%). This again verifies that MTOPdoes not cause additional collisions.

• Sixth, the frame drops reported in Fig. 9(b) aremeasured at the transmitter. When it does not re-ceive an ACK (either explicit or implicit), it countsit as a drop. As discussed earlier, it is caused byeither a lost frame or a lost ACK. The latter can bemeasured by counting the duplicate frames, whichis shown in Fig. 9(d). Comparing it with Fig. 9(b), itaccounts for 25.1∼28.6% of all frame drops in Srcr.It is 14.8∼24.2% in MTOP. Less duplicate frames inMTOP explains that implicit ACK in MTOP worksreasonably well.

FairnessAs discussed earlier, another important measure in

evaluating the MTOP protocol is fairness. Since nodesmay forward without appropriately competing for themedium, it might cause unfairness among the nodes.However, MTOP helps packets to travel similar distancetoward prospective destinations regardless of the datarate used, which we believe is more important in im-proving fairness in multihop environment. Based on thediscussion in Section 4.3, we define space-based fairnessindex, F , to measure the balance of total work done onflows in the network, i.e.,

F =(ΣLi=1(γi · disti))2

LΣLi=1(γi · disti)2, (17)

TABLE 3: Fairness index of Srcr and MTOPN=30 N=50 N=70 N=90 N=110

DSR1 0.7936 0.7203 0.6977 0.7290 0.6987DSR11 0.8402 0.7627 0.7413 0.6954 0.7080

Srcr 0.7353 0.7148 0.7471 0.7398 0.7512MTOP 0.7320 0.7087 0.7245 0.7203 0.7215

Thro

ughp

ut(b

ps)

10000

20000

End-to-end Distance(m)0 500 1000 1500

(a) Srcr

Thro

ughp

ut(b

ps)

10000

20000

End-to-end Distance(m)0 500 1000 1500

(b) MTOP

Fig. 10: Distance (disti) versus throughput (γi) (The figureshows a total of 300 flows comprising 30 flows per each of10 simulation runs when N=50.)

where L is the number of traffic flows. As discussed in[49], the index value ranges from 0 (completely unfair)to 1 (perfectly fair).

According to our simulation result shown in Table 3,MTOP seems not improve the fairness in comparisonto others including Srcr. This is due to the fact thatmany flows achieve almost 100% PDR regardless of theend-to-end distance of the flow when network is lightlyloaded. This becomes clearer in Fig. 10, which showsthe scatter plot of distance (disti) and throughput (γi).Similar values of F in MTOP and Srcr is contributed bythe flows represented by dots on the top in the figure.However, for the rest of the flows, MTOP offers morebalanced service to flows depending on their end-to-enddistances.

Virtual carrier sensing using RTS/CTSVirtual carrier sensing helps reduce collisions for

nonstop-forwarded packets in MTOP as shown inFig. 11(a). Both MTOP and Srcr drops more packets thanin Fig. 9(b) but Srcr consistently drops more packetsthan MTOP. However, it is more important to note thatnonstop-forwarded packets in MTOP drops rarely andis almost negligible. Nodes become better aware of themultihop forwarding due to the virtual carrier sense andthus, can avoid collisions and reduce packet drops.

One more note is that the number of total drops ismuch bigger than that in Fig. 9(b). This is not surprisingbecause the RTS/CTS exchange itself is an additionaloverhead and at the same time reduces the spatial resua-bility of the channel due to the more conservative settingof NAV (network allocation vector) for nodes near thereceiver.

Effect of defer thresholdAs discussed in Section 3.3, the defer threshold is an

essential parameter in carrier sensing mechanism as itdetermines neighboring nodes that need to refrain fromtransmitting. In multirate environment, it is normally set

12

MTOP(Non-stop)MTOP(Normal)

Srcr

0

50

100×103

Number of Nodes(N)30 50 70 90 110

(a) Effect of virtual carrier sensingusing RTS/CTS

Non-stopNormal

0

10000

20000

30000

Number of Nodes(N)

1st Bar: -96.2dBm2nd Bar: -99.3dBm3rd Bar: -103.8dBm

4th Bar:-105.1dBm

30 50 70 90 110

(b) Effect of defer threshold

Fig. 11: Total drops (comparing with Fig. 9(b))

to the value corresponding to the minimum TI distance,which is −105.1dBm in Table 2. In order to see theeffect of different setting of the defer threshold, we used−103.8, −99.3, and −96.2dBm that correspond to 2, 5.5,and 11Mbps. As shown in Fig. 11(b), a higher thresholdgenerates more collisions. At the defer threshold of−96.2dBm, the number of packet drops is 1.6∼5.4 timesmore than that of −105.1dBm. However, the differenceis reduced in high density scenario (large N ). This is be-cause more communications are made at high rates suchas 11Mbps and the communication at 11Mbps would notbe a problem with the defer threshold corresponding to11Mbps.

Effect of unreliable linksIn order to see how MTOP performs in a more realistic

environment, a set of experiments has been conductedwith Ricean model instead of the conventional two-rayground propagation model used above [50]. Accordingto [51], the Ricean channel is described by the K factor(K), which is defined as the ratio of mean power indominant component (line-of-sight) over the power inthe other scattered paths. We varied K from 0 (harshchannel condition) to 30 (better channel condition) butshow the results with K=5 only for brevity. The max-imum velocity (vmax), which represents the movementspeed of surrounding objects, is fixed to 0.5m/s in oursimulation [51].

Fig. 12 shows PDR, total number of transmissions,total frame drops, frame drop ratio, and mixtures of datarate (N=30 and N=90).• As shown in Fig. 12(a), the harsh channel condition

impacts all four. PDR reduces by 14.6∼21.1% and10.0∼31.0% in Srcr and MTOP, respectively. MTOPis affected more although the actual PDR is stillas much as 13.5% higher than Srcr. Note that itimpacts DSR11 most severely. DSR11 does not per-form well even in high-density scenarios (N=110)because high-rate links (11Mbps) suffer most inharsh conditions.

• We observed that there are more “Normal” trans-missions than “Non-stop” forwarding over Riceanchannel in Fig. 12(b) while this is the opposite inFig. 9(a). While nonstop forwarding in MTOP issupposed to happen over high-rate links, the harshchannel condition makes less use of high-rate links,

and so is nonstop forwarding.• Total number of frame drops increases with Ricean

channel, which is particularly prominent at low-density scenario in both Srcr and MTOP as shownin Fig. 12(c). In fact, frame drops in Srcr and normaldrops in MTOP increase sharply in Fig. 9(b) but itis almost constant in Fig. 12(c). The same can besaid based on Figs. 12(d) and 9(c). Frame drops withtwo-ray model are mainly caused by collisions butthose with Ricean channel are caused by channelvariations [52]. More frame drops render both Srcrand MTOP to use low-rate links, which is evidentin Figs. 12(e) and 12(f).

It can be concluded that the performance gain ofMTOP over Srcr is reduced under harsh channel con-dition because nonstop forwarding over high-rate linksgets stopped more often than in better channel condition.This leaves a room for improvement in our future work.

Performance with various traffic conditionsFigs. 13 and 14 show the performance in differently-

loaded networks by increasing the number of flows(Fig. 13) and the packet generation rate of each flow(Fig. 14). More traffic means more transmissions (Figs.13(b) and 14(b)), more collisions, and more framedrops (Figs. 13(c) and 14(c)). However, MTOP consis-tently exhibits a better PDR (0.5∼16.7% in Fig. 13(a)and 2.0∼15.6% in Fig. 14(a)) and less frame drops(31.5∼57.5% in Fig. 13(c) and 33.7∼56.3% in Fig. 14(c)).In terms of frame drop ratio, “Normal” and “Non-stop”frames are dropped 0.7∼4.8% and 4.4∼12.7% less thanSrcr, respectively. Therefore, it is concluded that MTOPoutperforms Srcr in a wide range of traffic conditions.

6 CONCLUSION AND FUTURE WORK

Multirate adaptation is a promising tool in wirelessmultihop networks as the corresponding hardware hasbeen available off-the-shelf. This paper proposes Multi-hop Transmission OPportunity (MTOP), which implementsnonstop frame forwarding mechanism and achieves low-latency, high-throughput communication. Feasibility ofMTOP has been proven via analysis and a small-scaletestbed based on USRP/ GNU Radio platform. Ourperformance study based on OPNET network simula-tor shows that MTOP performs better than fixed-ratescenarios (DSR1 and DSR11) and Srcr in terms of PDRand packet delay in the entire operating conditionssimulated. This is due to the adaptive behavior of MTOPdepending on data rate used and the aggressive frameforwarding mechanism along with the reduced MACoverhead.

MTOP opens up several interesting directions of re-search to pursue. First, the two new concepts, multiratemargin and space-based fairness, need a closer inves-tigation as they have a potential to further improvethe operation of multihop networks. Second, there areseveral ways to improve the MTOP. For example, opti-mal value of hi and the corresponding network density

13

0

20

40

60

80

100

Number of Nodes(N)30 50 70 90 110

DSR1DSR11

SrcrMTOP

(a) PDR(%)

0

1

2

3×105

Number of Nodes(N)30 50 70 90 110

MTOP(Non-stop)MTOP(Normal)

Srcr

(b) Total transmissions

0

20000

40000

60000

Number of Nodes(N)30 50 70 90 110

MTOP(Non-stop)MTOP(Normal)

Srcr

(c) Total drops

0

10

20

30

40

Number of Nodes(N)30 50 70 90 110

MTOP(Normal)MTOP(Non-stop)

Srcr

(d) Frame drop ratio(%)

MTOP(Non-stop)

MTOP(Normal)

Srcr

0 10 20 30 40 50 60 70 80 90 100

1Mbps 2Mbps 5.5Mbps 11Mbps

(e) A mixture of data rate used (N=30)

MTOP(Non-stop)

MTOP(Normal)

Srcr

0 10 20 30 40 50 60 70 80 90 100

1Mbps 2Mbps 5.5Mbps 11Mbps

(f) A mixture of data rate used (N=90)

Fig. 12: Performance comparison with Ricean fading (K=5)

0

20

40

60

80

100

Number of Flows20 25 30 35 40

Srcr MTOP

(a) PDR(%)

0

1

2

3×105

Number of Flows20 25 30 35 40

MTOP(Non-stop)MTOP(Normal)

Srcr

(b) Total transmissions

0

10000

20000

30000

40000

Number of Flows20 25 30 35 40

MTOP(Non-stop)MTOP(Normal)

Srcr

(c) Total drops

0

5

10

15

20

Number of Flows20 25 30 35 40

MTOP(Normal)MTOP(Non-stop)

Srcr

(d) Frame drop ratio(%)

Fig. 13: Impact of traffic load (number of flows)

0

20

40

60

80

100

Packet Generation Rate (pkts/sec)4 5 6 7

Srcr MTOP

(a) PDR(%)

0

1

2

3×105

Packet Generation Rate (pkts/sec)4 5 6 7

MTOP(Non-stop)MTOP(Normal)

Srcr

(b) Total transmission

0

10000

20000

30000

40000

Packet Generation Rate (pkts/sec)4 5 6 7

MTOP(Non-stop)MTOP(Normal)

Srcr

(c) Total drops

0

5

10

15

20

Packet Generation Rate (pkts/sec)4 5 6 7

MTOP(Normal)MTOP(Non-stop)

Srcr

(d) Frame drop ratio(%)

Fig. 14: Impact of traffic load (packet generation rate)

estimation constitute another future work. Third, MTOPcan be usefully employed in multi-radio/multi-channelnetworks, typically found in the backhaul of wirelessmesh networks. Fourth, it is noted that TXOP and MTOPare not mutually exclusive and can be combined todiversify and maximize the transmission opportunitiesin multihop networks. When a node transmits a frame, itmakes a prudent decision whether to seek an additionaltransmission opportunity according to TXOP or MTOP.

ACKNOWLEDGMENTS

This research was supported in part by the NSF underGrant CNS-0821319, Basic Science Research Programthrough the NRF (Korea) funded by the Ministry ofEducation, Science, and Technology (2010-0029034), andWCU program through the NRF (Korea) funded by the

Ministry of Education, Science, and Technology (R31-10100).

REFERENCES

[1] C. Yu, T. Shen, K. G. Shin, J.-Y. Lee, and Y.-J. Suh, “Multi-hoptransmission opportunity in wireless multihop networks,” in Proc.of IEEE INFOCOM, 2010.

[2] Civilian applications of unmanned aircraft systems (CAUAS,2007), http://www.cauas.colorado.edu/.

[3] T. Zajkowski, S. Dunagan, and J. Eilers, “Small UAS communi-cations mission,” in Proc. of the 11th Biennial USDA Forest ServiceRemote Sens. Applicat. Conf., 2006.

[4] P. Hui, J. Crowcroft, and E. Yoneki, “Bubble rap: social-basedforwarding in delay tolerant networks,” in Proc. of ACM MobiHoc,May 2008, pp. 241–250.

[5] H. Rheingold, Smart mobs: the next social revolution. Cambridge,Perseus Book Group, 2002.

[6] K. Seada and C. Perkins, “Social networks: the killer app forwireless ad hoc networks?” Nokia, Tech. Rep. NRC-TR-2006-010,Aug. 2006.

14

[7] IEEE 802.11-1999, Part 11: Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) Specifications, IEEE Std., 1999.

[8] IEEE 802.11e-2005, Part 11: Wireless Medium Access Control (MAC)and Physical Layer (PHY) Specifications: Medium Access Control(MAC) Enhancements for Quality of Service (QoS), IEEE Std., 2005.

[9] B. Sadeghi, V. Kanodia, A. Sabharwal, and E. Knightly, “Oppor-tunistic media access for multirate ad hoc networks,” in Proc. ofACM MobiCom, Sept. 2002, pp. 24–35.

[10] G. Tan and J. Guttag, “Time-based fairness improves performancein multi-rate WLANs,” in Proc. of The USENIX Annual TechnicalConference, 2004.

[11] I. Tinnirello and S. Choi, “Temporal fairness provisioning inmulti-rate contention-based 802.11e WLANs,” in Proc. of IEEEWoWMoM, June 2005, pp. 220–230.

[12] J. Bicket, D. Aguayo, S. Biswas, and R. Morris, “Architecture andevaluation of an unplanned 802.11b mesh network,” in Proc. ofACM MobiCom, Aug.–Sept. 2005, pp. 31–42.

[13] OPNET Modeler, http://www.opnet.com.[14] Universal Software Radio Platform (USRP), http://www.ettus.

com/.[15] GNU Radio Project, http://www.gnuradio.org/trac/.[16] A. Kamerman and L. Monteban, “WaveLAN-II: a high-

performance wireless LAN for the unlicensed band,” Bell LabsTechnical Journal, vol. 2, no. 3, pp. 118–133, Summer 1997.

[17] M. Lacage, M. H. Manshaei, and T. Turletti, “IEEE 802.11 rateadaptation: A practical approach,” in Proc. of ACM MSWiM, 2004.

[18] A. Akella, G. Judd, S. Seshan, and P. Steenkiste, “Self-managementin chaotic wireless deployments,” in Proc. of ACM MobiCom, Aug.–Sept. 2005, pp. 185–199.

[19] G. Holland, N. Vaidya, and P. Bahl, “A rate-adaptive MAC proto-col for multi-hop wireless networks,” in Proc. of ACM MobiCom,July 2001, pp. 236–251.

[20] Z. Ji, Y. Yang, J. Zhou, M. Takai, and R. Bagrodia, “Exploitingmedium access diversity in rate adaptive wireless LANs,” in Proc.of ACM MobiCom, 2004.

[21] S.-T. Sheu, Y. Tsai, and J. Chen, “MR2RP: The multi-rate and multi-range routing protocol for ieee 802.11 ad hoc wireless networks,”Wireless Networks, vol. 9, no. 2, pp. 165–177, 2003.

[22] H. Zhai and Y. Fang, “Physical carrier sensing and spatial reusein multirate and multihop wireless ad hoc networks,” in Proc. ofIEEE INFOCOM, Apr. 2006, pp. 1–12.

[23] B. Awerbuch, D. Holmer, and H. Rubens, “The medium timemetric: High throughput route selection in multi-rate ad hocwireless networks,” Mobile Networks and Applications, vol. 11, pp.253–266, 2006.

[24] R. Draves, J. Padhye, and B. Zill, “Routing in multi-radio, multi-hop wireless mesh networks,” in Proc. of ACM MobiCom, Sept.–Oct. 2004, pp. 114–128.

[25] Strix Systems, http://www.strixsystems.com/products/datasheets/StrixWhitepaper\ Multihop.pdf (accessed March12, 2008).

[26] C. E. Perkins and P. Bhagwat, “Highly dynamic destination-sequenced distance-vector routing (DSDV) for mobile comput-ers,” ACM SIGCOMM Computer Communication Review, vol. 24,no. 4, pp. 234–244, 1994.

[27] D. Johnson and D. Maltz, Dynamic Source Routing in Ad-HocWireless Networks. Ch. 5, Mobile Computing, edited by T. Imielinski,H. Korth. Kluwer Academic Publishers, 1996.

[28] S. Biswas and R. Morris, “Opportunistic routing in multi-hopwireless networks,” in Proc. Second Workshop Hot Topics in Networks(HotNets-II), 2003.

[29] M. Zorzi and R. Rao, “Geographic random forwarding (GeRaF)for ad hoc and sensor networks: Energy and latency perfor-mance,” IEEE Trans. Mobile Computing, vol. 2(4), 2003.

[30] K. Jamieson, H. Balakrishnan, and Y. C. Tay, “Sift: a MAC protocolfor event-driven wireless sensor networks,” in Proc. of the ThirdEuropean Workshop on Wireless Sensor Networks (EWSN), 2006, pp.260–275.

[31] T. Li, Q. Ni, D. Malone, D. Leith, Y. Xiao, and T. Turletti,“Aggregation with fragment retransmission for very high-speedWLANs,” IEEE/ACM Trans. Netw., vol. 17(2), pp. 591–604, 2009.

[32] M. Heusse, F. Rousseau, G. Berger-Sabbatel, and A. Duda, “Per-formance anomaly of 802.11b,” in Proc. of IEEE INFOCOM, vol. 2,Mar.–Apr. 2003, pp. 836–843.

[33] A. Kamerman and L. Monteban, “Wavelan-ii: A high-performancewireless lan for the unlicensed band,” Bell Labs Technical Journal,pp. 118–133, 1997.

[34] Orinoco 11b Client PC Card Specification, Available from http://www.proxim.com/learn/library/datasheets/11bpccard.pdf,2003.

[35] K. Marquess, “Physical model sub-group discussion and ques-tions,” IEEE 802.15/138R0, 1999.

[36] IEEE 802.15.2-2003, Part 15.2: Coexistence of Wireless Personal AreaNetworks with Other Wireless Devices Operating in Unlicensed Fre-quency Bands, IEEE Std., 2003.

[37] C. Yu, K. G. Shin, and L. Song, “Link-layer salvaging for makingrouting progress in mobile ad hoc networks,” in Proc. of ACMMobiHoc, May 2005, pp. 242–254.

[38] M. Bertocco, G. Gamba, and A. Sona, “Experimental optimizationof CCA thresholds in wireless sensor networks in the presence ofinterference,” in Proc. of IEEE EMC Europe, 2007.

[39] R. Gummadi, D. Wetherall, B. Greenstein, and S. Seshan, “Under-standing and mitigating the impact of RF interference on 802.11networks,” in Proc. of ACM SIGCOMM, Aug. 2007, pp. 385–396.

[40] S. Narayanaswamy, V. Kawadia, R. S. Sreenivas, and P. R. Kumar,“Power control in ad-hoc networks: theory, architecture, algo-rithm and implementation of the COMPOW protocol,” in Proc.of European Wireless Conference, Feb. 2002.

[41] A. Jow, C. Schurgers, and D. Palmer, “CalRadio: a portable, flexi-ble 802.11 wireless research platform,” in Proc. of ACM MobiEval,June 2007, pp. 49–54.

[42] K. A. Jamieson, “The SoftPHY abstraction: from packets to sym-bols in wireless network design,” Ph.D. dissertation, MIT, June2008.

[43] GNU Radio Source, http://gnuradio.org/trac/browser/gnuradio/trunk/usrp/fpga/sdr lib/adc interface.v.

[44] J. Lee, W. Kim, S.-J. Lee, D. Jo, J. Ryu, T. Kwon, and Y. Choi, “Anexperimental study on the capture effect in 802.11a networks,” inProc. of ACM WinTECH, Sept. 2007, pp. 19–26.

[45] BBN ADROIT Project, http://acert.ir.bbn.com/projects/adroit/.[46] R. Dhar, G. George, A. Malani, and P. Steenkiste, “Supporting

integrated MAC and PHY software development for the USRPSDR,” in Proc. of IEEE Workshop on Networking Technologies forSoftware Defined Radio (SDR) Networks, 2006.

[47] T. Schmid, O. Sekkat, and M. B. Srivastava, “An experimentalstudy of network performance impact of increased latency insoftware defined radios,” in Proc. of ACM WinTECH, Sept. 2007,pp. 59–66.

[48] J. Broch, D. A. Maltz, D. B. Johnson, Y.-C. Hu, and J. Jetcheva, “Aperformance comparison of multi-hop wireless ad hoc networkrouting protocols,” in Proc. of ACM MobiCom, Oct. 1998, pp. 85–97.

[49] R. Jain, D. Chiu, and W. Hawe, “A quantitative measure offairness and discrimination for resource allocation in sharedsystems,” DEC, Tech. Rep. Research Report TR-301, Aug. 1984.

[50] R. J. Punnoose, P. V. Nikitin, and D. D. Stancil, “Efficient simula-tion of ricean fading within a packet simulator,” in Proc. of IEEEVTC-Fall, 2000.

[51] Z. Chen, X. Yang, and N. H. Vaidya, “Dynamic spatial backoff infading environments,” in Proc. of IEEE MASS, 2008.

[52] J. Kim, S. Kim, S. Choi, and D. Qiao, “CARA: Collision-aware rateadaptation for IEEE 802.11 WLANs,” in Proc. of IEEE INFOCOM,Apr. 2006.