1

Improving the Performance of OverlappedTransmission in Wireless Ad Hoc Networks

Surendra Boppana, Student Member, IEEE, and John M. Shea, Member, IEEE

Abstract—Multi-packet reception (MPR) has recently receivedattention for use in military wireless ad hoc networks becauseof the potential to greatly improve spatial reuse. However,MPR typically requires complicated hardware that makes itprohibitive to implement in many current military platforms.An alternative, cross-layer approach is to use an overlappedtransmission technique, in which multiple transmissions can occurin the same geographical area because the communicators canuse their knowledge of some of the interfering packets to recoverthe desired information. Examples of overlapped transmissiontechniques include overlapped carrier-sense multiple-access (OC-SMA), physical-layer network coding, analog network coding,and Katti et al.’s COPE protocol. These techniques have muchlower signal processing demands than MPR techniques and havealso stimulated a lot of interest for use in military networks. Muchof the previous research on overlapped transmission focuseson the physical-layer: how transmissions should be allowed tooverlap, how the signals should be designed and detected, andthe error probabilities under different conditions. These workstend to make optimistic assumptions about the availability ofpackets to perform overlapped transmission: often performanceis evaluated for a simple three-node network under the assump-tion that traffic is always available at each of the edge nodes.In this paper, we explain why these protocols may suffer underreal-world traffic conditions because of directional traffic flowsand congestion control at the transport layer. We focus on howtransmission control affects the performance of the OCSMAprotocol. We propose modifications to the TCP parameters andthe OCSMA protocol to improve the performance. We also showthat OCSMA offers better fairness over the IEEE 802.11 MACprotocol in several scenarios.

Index Terms—overlapped transmission, analog network cod-ing, cross-layer interaction, multi-packet reception

I. INTRODUCTION

The performance of wireless ad hoc networks is typi-cally limited by the need to use MAC protocols to preventtransmissions from nearby radios from interfering with eachother. Thus, any technique that can improve spatial reuseand reduce contention among the radios has the potential tosignificantly improve the performance. Although multi-userdetection (MUD) schemes have been investigated for manyyears (cf. [1]), multi-packet reception (MPR) techniques [2]have been drawing increasing interest from a cross-layer andmilitary perspective over the last decade (cf. [3–9]). Mostof these works on MPR would require MUD schemes withhigh computational complexity, which may prevent MPR

Surendra Boppana was, and John M. Shea is, with the Department ofElectrical and Computer Engineering, University of Florida 32611-6130.Surendra Boppana is now with Qualcomm, Inc., San Diego, CA 2121. Email:sboppana@qualcomm.com,jshea@ece.ufl.edu

The material in this paper was presented in part at the IEEE Military Comm.Conf., Oct. 2008, San Diego CA.

This work was supported by the National Science Foundation under grantnumber CNS-0626863 and by the Air Force Office of Scientific Researchunder grant number FA9550-07-10456.

capabilities from being implemented in most military radiosfor many years to come. Thus, alternative approaches that canimprove spatial reuse at lower complexity are desirable.

One way to increase spatial reuse without the complexityof MUD is to have radios use their knowledge of interferingpackets to do a very simple form of interference cancellation.We call this approach overlapped transmission, as packets areallowed to overlap in the communication medium, and can stillbe recovered without full MUD. Several different overlappedtransmission schemes have been proposed. To the best ofour knowledge, we published the first work on such schemesin [10]. In [10] and its extensions [11, 12], the overlappedcarrier-sense multiple access (OCSMA) protocol is proposedand evaluated. OCSMA coordinates packet transmissions sothat packets are allowed to interfere if the nodes know thecontents of the interfering packets and can remove the in-terference. Working independently, Zhang et al. proposed aclosely related scheme called physical-layer network coding(PNC) [13] at approximately the same time. PNC relies onhard-decision demodulation and remodulation at a relay (in theterminology of cooperative communications, it is essentially adecode-and-forward scheme). In [14], analog network coding(ANC) is proposed. ANC is similar to PNC, except thatsoft demodulation and retransmission is used (similar to theamplify-and-forward scheme in cooperative communications).In fact, one of the forms of ANC described in [14] isessentially the same as our overlapped transmission schemesdescribed in [10–12]. Note that PNC and ANC are also derivedfrom earlier work on wireless network coding [15–18].

It is convenient to divide the previous papers on overlappedtransmission and wireless network coding into two groups.One group [13–16, 19] focuses on problems that traditionallylie in the physical layer, such as how packets should becoded together, how packets can be recovered if they areallowed to combine in the air, the link performance of suchschemes, and the performance improvement for some idealnetwork scenarios (typically consisting of 3-5 nodes). Theother group considers the performance of such schemes inthe context of a larger network [10–12, 17, 18]. This secondgroup of papers have all identified a significant problemfor overlapped transmission/wireless network-coding schemes:when the traffic is dominated by a small number of TCP flows,the performance gain will be severely limited. This is becauseTCP flow control results in nodes not having the appropriatepackets to combine or overlap. We call this packet starvation.

In this paper, we investigate packet starvation for OCSMAwith unidirectional TCP flows. In Section II, we provide anoverview of the OCSMA protocol. In Section III, we explainwhy TCP causes packet starvation for the OCSMA protocol.We investigate the factors that affect packet starvation andpropose some modifications to the MAC and TCP parameters

2

to improve the end-to-end throughput. In Section IV, wepropose a new variant of the OCSMA protocol to address thepacket starvation problem with TCP traffic. In Section V, weinvestigate the effects of the OCSMA protocols on fairness fornetwork topologies with interacting TCP flows. We concludethe paper in Section VI.

II. OCSMA PROTOCOL OVERVIEW

We briefly review the OCSMA protocol [12]. The OCSMAprotocol is based on the distributed coordinated function(DCF) mode of the IEEE 802.11 MAC protocol [20, Section9.2]. Unless stated explicitly, the terminology used in thefollowing sections corresponds with that in the IEEE 802.11standard [20]. The OCSMA protocol can be summarized infour phases using the example network of Fig. 1(a):

1. Primary Handshaking Consider the network inFig. 1(a), where at some point of time, node C intends toforward a packet to D that it has received from B in an earliertransmission. C transmits a Request To Send (RTS) frame toD, and if D senses the medium to be free, it responds with aClear To Send (CTS) frame, as shown in Fig. 1(b). Nodes Cand D are called the primary transmitter and primary receiver,respectively. This is similar to the RTS/CTS exchange of theIEEE 802.11 MAC protocol [20].

2. Secondary Handshaking Upon receipt of the CTS,the primary transmitter sends a Prepare To Send (PTS) frameto the node from which it received the present data frame inan earlier transmission. Upon the transmission of PTS, theprimary transmitter defers the transmission of the data frameuntil the completion of the secondary handshaking.

After the completion of the RTS/CTS between C and D, Csends a PTS to B. The node receiving the PTS frame is calledthe secondary receiver. Upon receipt of the PTS, the secondaryreceiver ensures that there are no other transmissions occurringin its sensing range except for the primary transmission. If true,it identifies a suitable partner for secondary transmission. Thesecondary receiver sends a Request to Transmit (RTT) frame tothe selected secondary transmitter. If the secondary transmitterfinds the medium to be free and has a packet to transmit, itresponds with a Clear to Transmit (CTT) frame. Transmissionof the CTT implies that the secondary transmitter is capableof transmitting overlapped data without causing interferenceto any of the transmissions in its communication range.

In the example network of Fig. 1(a), when B receives thePTS from C, it ensures that its Transmit Allocation matrix(TAX) is not set. TAX [12] consists of an array of TransmitAllocation Vectors (TAV), which are responsible for virtualcarrier sensing. Similar to the NAV in IEEE 802.11, a TAV isset for each valid frame the node receives that is not addressedto it. The medium is considered busy if any of the TAVs areset. For more details on TAX implementation, refer to [12].

Based on the selection criteria for choosing a partner,assume node B chooses node A to send the RTT. If A sensesthe medium to be free, it responds with a CTT.

3. Primary and secondary transmissions Upon com-pletion of the secondary handshaking, C starts the datatransmission to D, as shown in Fig. 1(d). The secondary

transmitter starts its overlapped transmission ∆0 seconds afterthe commencement of the primary transmission [12]. Fig. 1(d)also depicts B receiving an ODATA frame while using itsknowledge of the packet being transmitted by C to cancel outthe interference caused by that transmission.

4. Data Acknowledgments Upon the successful receptionof DATA and ODATA frames, the primary and secondarytransmitters sequentially transmit ACKs, as shown in Fig. 1(e).

III. PACKET STARVATION IN TCP FLOWS

In this section, we investigate the interaction between TCPand OCSMA, and the effect that TCP has on the opportu-nities to perform overlapped transmission. The focus is onidentifying factors that cause packet starvation and adjustingthe parameters of the MAC and transport layers to alleviatethis issue. We present simulation results from ns2 [21] andstudy the interaction between the two layers. We compare andcontrast the OCSMA protocol with the IEEE 802.11 protocol.In all reported results, the effects of the additional overheadof the OCSMA protocol, including delay to acquire the DATApacket before the ODATA packet is transmitted, have beentaken into account. In the following, we refer to a MAC servicedata unit (MSDU) as a frame, and a transport layer service unit(TSDU) as a packet. MAC layer acknowledgments are denotedby ACK, while those at the transport layer are denoted by ack.

101 2 3 4 5 6 7 8 9

Fig. 2. Ten node linear network.

To focus on the factors that limit the performance ofoverlapped transmission techniques with TCP flows, in thispaper we only consider linear topologies and one topologythat is the superposition of two linear topologies. Results onrandom and mobile networks are included in our previouspaper on OCSMA [12]. We also focus on a protocol model,in which a packet is received correctly if the intended receiveris within a radius of 250 m of the transmitter and there are noother transmissions within a radius of 550 m of the receiver.

We first consider a ten-node linear network as depicted inFig. 2. Node 1 is the source, and node 10 is the destination.The nodes are placed at regular intervals of 200 m. In Fig. 2,the transmission ranges of nodes 3 and 9 are denoted by solidcircles, and their respective sensing ranges are denoted bydashed circles. The default values of the simulation parametersare summarized in Table I.

A. Impact of TCP Congestion Window Size

In this section, we evaluate the impact of the TCP con-gestion window (CW) size on the end-to-end throughput of

3

CTT DATA ACK1 ACK2

O!DATA

G

A

B

E

F

RTT

C

RTS CTS PTS

D

(a) Ad hoc network

1b−CTS

G

A

B

E

F

D

C

1a−RTS

(b) Primary Handshaking

2c−CTT

C

D

G

B

E

F

A

2a−PTS

2b−RTT

(c) Secondary Handshaking

3b−ODATA

C

D

G

A

B

E

F

3a−DATA

(d) Primary and Secondary Transmissions

4b−ACK2

C

D

G

A

B

E

F

4a−ACK1

(e) Data Acknowledgements

Fig. 1. Typical frame exchanges in the OCSMA protocol.

TABLE ISIMULATION SETUP.

Parameter ValueData rate 1 Mbps

Traffic Model FTP “simulated application”Congestion Control Mechanism TCP Reno

Simulation duration 2000 sWarm-up time 200 s

Routing protocol AODVChannel model Two ray propagationRTS Threshold 150 Bytes

Transmission radius 250 mCarrier-sensing radius (Interference range) 550 m

IFQ length 100Overlapped Delay ∆0 240 µs

STA Retry Limits (Short, Long) (7,4)TCP packet size 1400 Bytes

the network. In ns2, the CW parameter represents the receiveradvertised window size, and defines the maximum number ofpackets to be sent at every round-trip-time. TCP is designedto adjust the flow based on the CW size and the congestion inthe network. Henceforth, unless otherwise stated, CW refersto the receiver’s advertised CW.

Fig. 3 compares the end-to-end throughput of the ten-nodelinear network of Fig. 2 under the OCSMA and IEEE 802.11MAC protocols as a function of the TCP CW size. In the caseof IEEE 802.11, the throughput of the network increases withan increase in CW size until the CW size is 4, beyond whichit decreases. This behavior of TCP under IEEE 802.11 has

5 10 15 20 25 307

8

9

10

11

12

13

TCP Congestion Window size (packets)

End!

to!e

nd T

hrou

ghpu

t (pa

cket

s/s)

OCSMAIEEE 802.11

Fig. 3. End-to-end throughput comparison in a ten-node linear network withTCP traffic.

been reported in [22, 23], where it was noted that the bestperformance is achieved when the TCP CW size is a fraction(usually 1/4) of the number of nodes in the linear network.

The TCP throughput under OCSMA shows different behav-ior than IEEE 802.11, increasing with an increase in CW sizeand saturating for CW sizes greater than 14. This behaviorof OCSMA can be better understood by analyzing the linkthroughput. The link throughput under OCSMA and IEEE802.11 are plotted in Fig. 4. The left ordinate scale is for DATAframes, and the right ordinate is for ODATA. Note that underOCSMA, both DATA and ODATA frames contribute to the link

4

2 4 6 8 10 12 14 16 18 2060

80

100

120Li

nk T

hrou

ghpu

t (fra

mes

/s)

2 4 6 8 10 12 14 16 18 200

10

20

30

TCP Congestion Window size (packets)

Link

Thr

ough

put (

fram

es/s

)

OCSMA, DATAIEEE 802.11, DATA

OCSMA, ODATA

Fig. 4. Link layer throughput comparison of OCSMA and IEEE 802.11 ina ten-node linear network.

throughput. The link throughput under IEEE 802.11 mimicsthe end-to-end throughput curve of Fig. 3. For IEEE 802.11,the rate at which data frames are dropped due to collisionsincreases as the CW size increases. In the case of OCSMA,both the DATA and ODATA reception rates increase with anincrease in the CW size and saturate for CW sizes greaterthan 14. An increase in the CW size increases the number offrames available for overlapped transmission in the network.The DATA reception rate under OCSMA is less than the DATAreception rate under IEEE 802.11. However, the combinationof DATA and ODATA frames in OCSMA provides a greaterend-to-end throughput than IEEE 802.11.

TABLE IIEVENTS AT THE MAC LAYER IN A TEN-NODE LINEAR NETWORK UNDER

THE OCSMA PROTOCOL.

Frame CW = 1 CW = 4 CW =8 CW =16type (Events/s) (Events/s) (Events/s) (Events/s)RTS 63.8 155.4 182.4 190.0CTS 63.8 103.2 106.6 107.1PTS 49.6 78.9 80.1 79.6RTT 49.6 58.6 57.8 56.6CTT 0 11.5 18.3 21.5NPT 49.6 38.5 28.7 23.4

DATA 63.8 94.1 96.0 95.8ODATA 0 11.5 18.3 21.5COLL 0 7.8 8.7 8.8

NPT/(NPT+CPT) 100% 77.0% 61.1% 52.1%ODATA/DATA 0 % 12.2% 19.1 % 22.4%

We further investigate the behavior of OCSMA by analyzingthe MAC-layer events1 across the network. The MAC-layerevents under OCSMA are tabulated in Table II for severalvalues of CW size. We begin by noting the effects of increasingthe CW on the collision rate (COLL). The collision rate underOCSMA increases to over 8% for CW sizes of 8 and 16.In contrast, the collision rate of IEEE 802.11 is less thanapproximately 1% for CW sizes up to 20. The average rateof RTS frames received increases as the CW size increases,as does the reception rate of RTT frames (indicating that theopportunity to perform overlapped transmissions increases).However, the ratio of the reception of CTT to that of RTT is

1In this scenario, events correspond to either reception of a frame or acollision.

significantly lower than one, which indicates that the actualnumber of ODATA transmissions is significantly less than thepotential number of overlapped transmissions.

The No Packet to Transmit (NPT) frame [11] was intro-duced during the simulations to investigate the reason for thelow rate of overlapped transmissions. A secondary transmitterresponds to an RTT with an NPT if the secondary transmitterdoes not have a packet in its queue that is intended forthe secondary receiver, and therefore an overlapped transmis-sion is not possible. This is the packet starvation problem.When a secondary transmitter receives an RTT packet, itwill always transmit a CTT or NPT packet. Thus, the ratioNPT/(NPT+CTT) gives an indication of how often packetstarvation is occurring. Referring to Table II, we note thatthe ratio NPT/(NPT+CTT) is very high for small CW sizes,which indicates that many opportunities for overlapped trans-missions are missed because of a lack of suitable frames atthe secondary transmitters. Even for large CW sizes, the ratioof NPT to (NPT+CTT) is over 50%. The result is the lowproportion of ODATA packets to DATA packets, which is atmost 22.4%. Thus there is considerable room for improvementin OCSMA if the packet starvation problem can be addressed.

To further understand packet starvation and its relation tothe TCP CW, consider again the network of Fig. 2 and supposethat TCP is operating in the congestion-avoidance phase withan empty link-layer queue at node 1. When node 1 receives anack packet for a TCP packet that it sent to node 10, it pushes apacket (most of the time) to the link-layer queue at the MAClayer. The MAC of node 1 transmits this “DATA frame” tonode 2, which it forwards to node 3, and so on. When node3 forwards this frame to node 4, an overlapped transmissionbetween nodes 1 and 2 is possible. However, it is possibleonly if node 1 receives a new ack for a TCP packet that ittransmitted earlier. When the receiver CW size is small, theprobability of this happening is very small, and the opportunityfor overlapped transmissions are wasted every time a TCPpacket is forwarded along the linear network during thecongestion avoidance phase. Thus, one method through whichthe CW affects the ability to perform overlapped transmissionis through the effect on the release of packets at the source.However, this is not the only way in which the CW affectsthe ability to perform overlapped transmissions. Note fromTable II that when the CW size increases, the collisions inthe network increase. Thus, there is a high probability thatthe MAC protocol of a node is in the backoff state and thusits queue is non-empty. This also increases the probabilityof overlapped transmission. We further investigate this effectin Section III-B and propose a modification to the TCPparameters to improve the performance of OCSMA.

The results in Fig. 5 show the evolution of the sourcecontention window for several values of the maximum TCPCW2. Note that when the CW size is low, TCP is in thecongestion-avoidance phase, which implies that the probabilityof transmitting exactly one packet upon reception of an ackis very high. As the CW size increases, TCP experiences

2The actual number of packets transmitted by the source is the minimumof the transmitter and receiver CWs.

5

0 10 20 30 40 50 60 70 80 90 1000

20

40Tr

ansm

itter

Con

gest

ion

Win

dow

size

(pac

kets

)

0 10 20 30 40 50 60 70 80 90 1000

10

20

0 10 20 30 40 50 60 70 80 90 1000

10

20

0 10 20 30 40 50 60 70 80 90 1000

10

20

Time (s)

Rx CW =2

Rx CW =3

Rx CW =5

Rx CW = 10

Fig. 5. Transmitter congestion window evolution in a ten-node linearnetwork.

slow-start phases quite often3, which result in TCP at node 1often releasing multiple packets in response to an ack packet.This increases the probability of overlapped transmission. Theresults in Table II verify that the number of overlapped trans-missions increases with an increase in the CW size; yet, thefull potential of overlapped transmissions is not realized. Forinstance, for a CW size of 16, the ratio of CTT/(NPT+CTT) isonly 48%. This is mainly due to the lack of interaction betweenOCSMA and TCP, which we further investigate below.

It is important to note that in real-world conditions, TCPCW size is a dynamic parameter that is not controlled by thesource. The destination advertises the TCP CW, which is areflection of the congestion of the network as perceived by thedestination. Hence it may not be always possible to operate ata CW size that provides maximum end-to-end throughput.

B. Techniques to mitigate collisions and improve throughput

In the previous section, we observed that the throughput ofthe network increases with an increase in the TCP CW size.However, an increase in CW size increases the number ofcollisions in the network (refer to Fig. 5 and Table II), whichcan be attributed to an increase in the number of control framesin OCSMA (vs. IEEE 802.11). In this section, we considerstrategies to mitigate the impact of higher collision rates inOCSMA, which is particularly severe for large CWs.

We first note that higher collision rates will result in morepackets being dropped by the MAC. To compensate for thisbehavior, the STA Short Retry Count (SSRC), and STA LongRetry Count (SLRC) limits [20, Section 9.2.5.3] should beincreased. In our previous work [12], we evaluated the effectsof increasing these retry limits on the throughput of TCPin linear networks. We observed that the throughput gainsof OCSMA over IEEE 802.11 increase as the retry limitsincrease. Thus, in the remainder of this work, we set the SSRCand SLRC limits to 20 and 10, respectively.

Here we consider the use of delayed-ack(n) in TCP, inwhich one ack is sent for every n received packets [24].This changes the granularity at the source: for each ack

3The TCP variant TCP-Reno is used for congestion control.

that the source receives, TCP will release multiple packetsto the interface queue. Since fewer acks are transmitted, weexpect this strategy to decrease the number of collisions in thesystem. Since multiple packets are released simultaneously atthe source, we expect that this may increase the number ofoverlapped transmissions and, hence, the throughput.

0 5 10 15 20 25 309

10

11

12

13

14

15

TCP CW size

End!

to!e

nd T

hrou

ghpu

t (pa

cket

s/s)

n=1n=2n=3n=4

Fig. 6. End-to-end throughput comparison in a ten-node network for severalvalues of delayed-ack granularity, n.

The results in Fig. 6 illustrate the end-to-end throughput of aten-node network for several values of delayed-ack granularity,n. An important design parameter that impacts the systemperformance is the ack-timeout parameter. This determines themaximum tolerable delay between the first and nth successfulreception of packets at the receiver before which an ack isgenerated. This interval value is chosen to be slightly greaterthan n times the average time for a packet to reach thedestination when no delayed-ack strategy is employed. Theresults indicate that there is considerable throughput gain fromn = 1 to n = 2. However, increasing the value of n beyond2 does not provide significant gains. Note that as n increases,the sensitivity of the network to an ack packet loss increases,which negates the benefit of the reduced collision rate.

TABLE IIIPERFORMANCE COMPARISON OF OCSMA AND OCSMA DA(2).

OCSMA OCSMA OCSMA OCSMAType DA(2) DA(2)

(CW=2) (CW=2) (CW=16) (CW=16)TPUT 9.3 9.3 13.0 15.1RTS 84.9 66.3 190.1 196.1CTS 76.3 65.8 107.1 112.1PTS 59.3 51.6 79.6 84.6RTT 52.0 51.3 56.6 61.7CTT 11.2 19.8 21.5 30.0NPT 49.6 32.0 23.4 18.0

DATA 63.8 65.8 95.8 105.4ODATA 11.2 19.8 21.5 30.0COLL 3.4 0.0 8.8 5.4CTT 18.4% 38.2% 47.9% 62.5%

/(CTT+NPT)ODATA 17.6% 30.1% 22.4% 28.5%/DATA

6

101 2 3 4 5 6 7 8 9

Fig. 7. Ten-node linear network under OCSMA LA.

The end-to-end throughput and the MAC layer events underOCSMA and OCSMA with delayed-ack (OCSMA DA(2)) aretabulated in Table III. When the CW size is 2, the end-to-end throughput (TPUT) under OCSMA and OCSMA DA arethe same. However, the number of collisions in the case ofOCSMA DA(2) is much lower than in the case of OCSMA.The number of overlapped transmissions is also greater inthe case of OCSMA DA(2). When the CW size is 16, OC-SMA DA(2) provides 16% throughput gain over OCSMA.The ratio CTT/(CTT+NPT) is 62.5% for OCSMA DA(2)with CW size 16. Also note the increase in overlappedtransmissions and reduction in collisions in the case of OC-SMA DA(2). However, the proportion of ODATA to DATAis still relatively small at 28.5%. Thus, the performance ofOCSMA is still very limited compared to its potential.

IV. OCSMA WITH LOOK AHEAD CAPABILITY(OCSMA LA)

In the previous sections, we analyzed the impact of OCSMAon the performance of TCP flows in wireless networks. Thesimulation results suggest that although OCSMA provides sig-nificantly better end-to-end throughput than the IEEE 802.11protocol, the full potential of overlapped transmissions is notrealized. We attribute this to packet starvation and lack ofinteraction between the two layers. In this section, we modifythe OCSMA protocol and incorporate the observations of theprevious sections to address the issue of packet starvation.

Motivated by the work in [25], we introduce the conceptof look-ahead. Upon the completion of an overlapped trans-mission, both the primary and secondary receivers contendfor the channel access. The OCSMA protocol was designedto allow for the secondary receiver to backoff for a greaterduration and allow for the primary receiver to have a greaterchance for channel access ( cf. [12] for more details). However,this design doesn’t always guarantee the occurrence of anoverlapped transmission. In networks with linear flows, theprobability of an overlapped transmission can be increased byensuring that the primary receiver of the current overlappedtransmission always gets access to the channel before thesecondary receiver. This is accomplished with the help of thelook-ahead feature, as explained below.

A. OCSMA with Look-ahead Protocol Description

OCSMA with look-ahead (OSCMA LA) is based on theOCSMA protocol, so here we highlight only the differencesbetween the two protocols. We will use the example networkof Fig. 7 to describe the design of the OCSMA LA protocol.The differences between OCSMA and OCSMA LA are duringthe acknowledgment phase, as described below.

In the network of Fig. 7, assume that node 3’s transmis-sion to node 4 is the primary transmission, and node 1’stransmission to node 2 is the secondary transmission. After

!"#$%&'()

*!%&'()"+,-&+. %"/01&'-2+,

%&'()"+,-&+.

*!%&'()"+,-&+. %"/01&'-2+,

*!%&'()"+,-&+. %"/01&'-2+,

%&'()"+,-&+.

%&'()"+,-&+.

%&'()"+,-&+. 01&'-2+, *! 3! 0! %"/450

678-)-9: 6 ; ; ; < <

43/$%&'()

%&'()"+,-&+.

78-)-9: 6 6 ; ;$ <

*! 3! %"/01&'-2+,

6 6 ; <78-)-9:

*3/$%&'()

"3/$%&'()

78-)-9: 6 6 ; ; ; <

*33$%&'()

78-)-9: 6 6 ; <

"33$%&'()

78-)-9: 6

0!3!=70!3!$%&'()

78-)-9: 6 6 ; <

01&'-2+, *! 3! %"/4!

6 ; ; ; 6 ;

01&'-2+,=50

!>>&)99?!>>&)996 !>>&)99@/)A1),8)"+,-&+. !>>&)99<

%&'()$B+>C %"/

D!6@?6 <

678-)-9: 6 ; ; ; 6 <

!#E$%&'()

01&'-2+, *! 3! F! %"/F%0

Fig. 8. ACK and ACKM frame formats for the OCSMA LA protocol.

the completion of the secondary handshaking phase, node 3commences transmission of the DATA frame to node 4. Uponsuccessful reception of the DATA frame, node 4 acknowledgesit with a modified ACK, the AKM (ACK Modified) frame.The frame format of the AKM frame is shown in Fig. 8.4

The AKM frame, in addition to the Receiver Address (RA)field, contains the additional fields TA, NA and NFD. TheTransmitter Address (TA) field contains the address of thenode transmitting the AKM frame. The Next Address (NA)field contains the address of the node for which the presentnode (the node transmitting the AKM) has a DATA frame,and the Next Frame Duration (NFD) contains the durationinformation of the DATA frame. Node 4 uses the contents ofthe first available DATA frame in its queue to fill the fieldsNA and NFD. If node 4 does not have a DATA frame in itsqueue and if the present frame is to be forwarded on by D,the present frame is used to generate the required informationbefore sending it to the higher layers. (This requires the MACto have access to the routing tables.)

When the primary transmitter, node 3, receives the AKM, itresets its retry limits and performs backoff just like in the caseof the reception of an ACK frame. When the next hop receiver,node 5, receives the AKM frame, it waits for a duration equalto the transmission of an ACK frame (to allow for node 2’stransmission of ACK to node 1), and transmits a CTS frameif the medium is free. Note that the information necessary forupdating the fields RA and Duration of the CTS frame (referto Fig. 8) are available through the TA and NFD fields of theAKM frame (refer to Fig. 8). When node 3 receives the CTSframe, it ensures that this frame is in response to either anRTS frame or an AKM frame. If this is true, it proceeds withthe secondary handshaking phase of OCSMA.

Since the next hop receiver (node 5 in the present example)requests for the DATA frame even before the secondary andprimary receivers have a chance to contend for the channelaccess, the secondary receiver, node 3, has a suitable framefor an overlapped transmission when node 4 transmits theDATA frame to node 5. Once an overlapped transmissionoccurs in the linear network, with high probability, the ca-pability to perform overlapped transmission is retained untilthe DATA/ODATA frames reach the destination. For instance,in the example network of Fig. 7, when node 4 transmits aDATA frame in response to the CTS sent by node 5 (whichresponds to an AKM frame sent by node 4), node 3 has asuitable frame for an overlapped transmission, and in the next

4The other frame formats for OCSMA LA are the same as for OCSMA [12]

7

transmission interval, when node 5 transmits the DATA framein response to the CTS sent by node 6, node 4 would have asuitable frame (the ODATA frame that it received from node 3)for an overlapped transmission, and so on.

The probability of overlapped transmission remains highwhen the primary transmission is successful and the AKMis successfully received. For the protocol model used inthis paper, this will be true as long as the collision rate islow. However, in more practical scenarios, fading may causeadditional degradation to the performance increase from thelook-ahead modification.

B. Simulation Results

5 10 15 20 25 309

10

11

12

13

14

15

16

TCP Congestion Window size (packets)

End!

to!e

nd T

hrou

ghpu

t (pa

cket

s/s)

OCSMAOCSMA_LAIEEE 802.11

Fig. 9. Throughput comparison of OCSMA, OCSMA LA and IEEE 802.11in a ten-node linear network.

We simulated the OCSMA LA protocol using ns2, and wecompare its performance to OCSMA and IEEE 802.11 forthe ten-node linear network of Fig. 2. The parameters usedfor the simulation are tabulated in Table I, except that theshort and long retry limits are set to 20 and 10, respectively,and OCSMA LA utilizes delayed-acks with n = 2. Theresults in Fig. 9 compare the end-to-end throughput of theOCSMA DA(2), OCSMA LA, and IEEE 802.11 protocols asa function of the TCP CW size (also see Fig. 3). The end-to-end throughput under IEEE 802.11 increases until the CWsize equals 6, beyond which it decreases. The throughputsunder OCSMA and OCSMA LA increase with an increase inCW size, saturating for CW sizes greater than 16. For CWsize greater than 20, OCSMA DA(2) provides a throughputgain of 31% to 34% over IEEE 802.11, while OCSMA LAprovides a throughput gain of 39% to 41% over IEEE 802.11.

The MAC-layer events under OCSMA LA are given inTable IV for three different values of the CW size. Note thatunder OCSMA LA, the number of CTS frames received canbe greater than the number of RTS frames received. A CTSis transmitted either in response to an RTS or an AKM. Forsmall CW sizes, the ratio CTS/RTS is very high. For instance,

TABLE IVMAC-LAYER EVENTS IN A TEN-NODE LINEAR NETWORK UNDER

OCSMA LA.

FrameType CW=2 CW=4 CW=8 CW=16 CW=32

TPUT 11.54 13.7 15.1 15.7 15.71RTS 19.0 55.5 103.8 118.6 119.6CTS 71.9 97.6 115.3 115.7 116.2PTS 58.2 80.0 88.8 74.3 74.6RTT 57.2 73.5 76.4 74.0 74.9CTT 38.0 31.8 30.7 36 37.2NPT 15.7 31.0 32.5 23.0 20.0

DATA 65.9 91.1 103.5 103.3 104.2ODATA 38.0 34.0 30.7 36.0 37.2COLL 0.0 6.6 12.4 14.4 15.4CTS/RTS 3.8 1.8 1.1 0.98 0.97CTT

/(CTT+NPT) 70.8% 50.6% 48.6% 61.0% 65.0%

for CW size 2, CTS/RTS is 3.8, which indicates that the look-ahead feature is often successful in scheduling transmissionsby the next radio in the linear network. However, for largeCWs, the CTS/RTS ratio is smaller, but still much larger thanwithout the look-ahead feature. For example for a CW sizeof 16, OCSMA LA has a CTS/RTS ratio of 0.98, whereas,from Table III, OCSMA DA(2) has a CTS/RTS ratio of 0.57.Thus, the look-ahead feature is still playing a prominent role,but just not as much as for small CW sizes. The reason is thatlarge CWs increase the number of packets in the network. Thistranslates into a larger collision rate in the network, which candisrupt the performance of the look-ahead protocol.

We note that when the CW size is 2, the ratio ofCTT/(CTT+NPT) is 70.8%, and DATA loss due to collisionsis zero. As the CW size increases, the end-to-end throughputunder OCSMA LA increases; however, the number of over-lapped transmissions (ratio of CTT to CTT+NPT) decreasesuntil a CW size of 8, and then increases. This behavior isin contrast to the behavior of OCSMA DA (cf. Table III).The non-monotonic behavior of the number of overlappedtransmissions can be attributed to the non-monotonic behaviorof the availability of packets for overlapped transmissionat secondary transmitters, as is evidenced by the NPT ratein Table IV. As the CW size is increased from 2 to 8,the additional packets in the network contribute primarily toadditional DATA transmissions: the number of received DATApackets/s goes from 65.9 up to 103.5. At the same time,the proportion of overlapped transmission opportunities forwhich a secondary transmitter does not have an appropriatepacket rises from 21% to 51%. However, as the CW size isincreased above 8, the additional packets in the network donot contribute significantly to additional DATA transmissions.Effectively, the rate for DATA transmissions has saturated.The additional packets do increase the availability of packetsfor overlapped transmissions, however, and the proportion ofoverlapped transmission opportunities for which a secondarytransmitter does not have an appropriate packet falls from 51%to 35% as the CW size goes from 8 to 32. At the same time,the collision rate increases, so the gain from increasing the CW

8

does not produce a significant change in end-to-end throughputfor CW sizes greater than 16.

10 15 20 25 30 35 40 45 508

10

12

14

16

18

20

22

Packet arrival rate (packets/s)

End!

to!e

nd T

hrou

ghpu

t (pa

cket

s/s)

OCSMA_LAIEEE 802.11

Fig. 10. Throughput comparison of OCSMA LA and IEEE 802.11 in aten-node linear network with CBR traffic.

The OCSMA LA protocol is designed to address the issueof packet starvation in TCP flows. However, we expect thelook-ahead feature of OCSMA LA to benefit UDP trafficalso. To investigate this, we simulated OCSMA LA in a ten-node linear network with constant bit-rate (CBR) UDP traffic.The results in Fig. 10 compare the end-to-end throughput ofOCSMA LA and IEEE 802.11 with CBR traffic as functionof packet arrival rate at the source, node 1. The packet sizeis 1400 bytes, and the short and long retry limits are 20and 10, respectively. We note that OCSMA LA has a higherthroughput than IEEE 802.11, and when the packet arrivalrate increases beyond 22 packets/s, the degradation in thethroughput is more gradual compared to IEEE 802.11. Forpacket arrival rates greater than 22 packets/s, OCSMA LAprovides at least 98% gain over the IEEE 802.11 protocol.The maximum gain over the IEEE 802.11 protocol is 126%.

V. FAIRNESS ISSUES AND MEDIUM CONTENTION

The interaction between TCP and the MAC is a majorsource of unfairness in multihop ad hoc networks. Whendifferent flows experience different congestion issues, the re-sources allocated to them may be different. Thus, an importantconsideration when proposing new MAC protocols is theirimpact on the fairness in the allocation of the channel amongcompeting flows. In this section, we compare inter-flow con-tention issues for IEEE 802.11 and OCSMA LA5. Starvationis another major problem that is caused by the greedinessof the MAC protocols and TCP. To evaluate these issues inthe context of OCSMA LA, we consider the two networktopologies illustrated in Figs. 11(a) and 11(b). Fig. 11(a) showsa network with three parallel flows each traversing through sixnodes. The adjacent nodes in a flow are placed at a distanceof 200 m, and the adjacent flows are separated by a distanceof 400 m. Fig. 11(b) shows a network with two flows that

5The fairness results were not significantly different for OCSMA orOCSMA LA.

Flow310 12117 8 9

Flow1

16 181713 14 15

1 2 3 4 5 6Flow2

(a) Network with parallel flows

Flow

2

6

8

10

11

12

13

1 2 3 4 5 6 7

Flow 1

(b) Network with intersecting flows

Fig. 11. Networks with multiple flows.

intersect at a common node, with the adjacent nodes separatedby a distance of 200 m.

0 20 40 60 80 10010

12

14

16

0 20 40 60 80 1000

2

4

6

0 20 40 60 80 10010

15

20

Time (s)

Thro

ughp

ut (p

ackt

es/s

)

IEEE 802.11, flow 1OCSMA_LA, flow 1

IEEE 802.11, flow 2OCSMA_LA, flow 2

IEEE 802.11, flow 3OCSMA_LA, flow 3

Fig. 12. Throughput comparison in a network with multiple flows.

The results in Fig. 12 show the end-to-end throughputevolution for the network with three parallel TCP flows thatis shown in Fig. 11(a). The throughput of each of the flowsunder the OCSMA LA and IEEE 802.11 MAC protocols isplotted for consecutive intervals of length 5 s. We chose aTCP CW size of 2 for each flow, which we observed to givethe best fairness performance under both OCSMA LA and

9

IEEE 802.11. We observe that under IEEE 802.11, flows 1and 3 have non-zero throughput at all times, whereas the end-to-end throughput in the case of flow 2 is zero. The nodes offlow 2 experience interference from both flows 1 and 3, whichresults in problems with the on-demand routing protocol andzero end-to-end throughput for flow 2.6

Under OCSMA LA, the end-to-end throughput of flow 2is non-zero, but still lower than that of flows 1 and 3. Sinceflow 2 experiences interference from nodes in flow 1 and 3,it is not surprising that the throughput of flow 2 is lower thanthat of flows 1 and 3. The non-zero throughput of flow 2under OCSMA LA is primarily due to the effect of increasedcollisions, and the ability to perform overlapped transmission.The collision rate under OCSMA LA is very high for thisnetwork topology, and thus nodes (including the nodes inflows 1 and 3) spend more time in backoff, which providesa greater chance for nodes in flow 2 to compete and succeedin accessing the channel. On the other hand, the increase incollisions across all the flows is offset to a large extent by anincrease in spatial re-use due to overlapped transmissions. Forinstance, in Fig. 12, note that the throughputs of flows 1 and 3under OCSMA LA are similar to the case of IEEE 802.11.

To further compare these protocols, we use Jain’s fairnessindex [26], which is defined as

f(x1, x2, · · · , xn) =(∑n

i=1 xi)2

n (∑n

i= x2i )

, (1)

where x1, x2, · · · , xn are the flow throughputs of each of the nflows, respectively. The closer the fairness index is to unity, themore equal the flow throughputs. Using the average throughputof each of the three flows over a simulation duration of1800 s, we have with IEEE 802.11, f802.11(x1, x2, x3) = 0.67and with OCSMA LA, fOCSMA LA(x1, x2, x3) = 0.81. Thefairness index of 0.67 for IEEE 802.11 is due to the channelresources being equally divided between flow 1 and flow 3,while flow 2 is almost completely deprived of the channelresources. For OCSMA LA, flow 2 has non-zero throughput,which is reflected by a higher value of the fairness index.

The collisions in OCSMA LA result in throttling of theTCP flows, which creates opportunities for flow 2 to senddata. When saturated CBR UDP traffic is sent over the links,fewer such transmission opportunities are created. For UDPtraffic with the IEEE 802.11 MAC protocol, no throughputis achieved for flow 2, while flows 1 and 3 achieve through-puts of approximately 20 packets/s. With the OCSMA LAprotocol, flows 1 and 3 achieve throughput of approximately28.4 packets/s, while flow 2 achieves throughput of 0.8 pack-ets/s. The values of the fairness index for UDP traffic aref802.11(x1, x2, x3) = 0.67 and fOCSMA LA(x1, x2, x3) =0.69. These results indicate that OCSMA LA does not allevi-ate inter-flow contention issues by itself; it is the interactionwith TCP that creates opportunities for flow 2 to achievesignificant throughput. In general, this suggests that regardlessof the specific MAC protocol or topology, overall fairnessmay be improved if the greedy nature of the TCP flows

6Using the destination-sequenced distance vector (DSDV) routing protocolcan produce a small, but non-zero, throughput of 0.3 packets/s for flow 2.

0 20 40 60 80 1000

5

10

15

0 20 40 60 80 1000

5

10

15

Time (s)

Thro

ughp

ut (p

acke

ts/s

)

IEEE 802.11, flow 2OCSMA_LA, flow 2

IEEE 802.11, flow 1OCSMA_LA, flow 1

Fig. 13. Throughput comparison in a network with multiple linear flows.

are occasionally interrupted to allow other flows a chance tocompete for the channel resources.

The results in Fig. 13 show the throughput evolution forthe network of Fig. 11(b) for TCP traffic. The end-to-endthroughput of each of the flows under OCSMA LA and IEEE802.11 is plotted for consecutive intervals of length 5 s. For theIEEE 802.11 protocol, we see that on any of the 5 s intervals,one of the flows captures the resources, while the other flow iscompletely deprived of the channel resources. This exemplifiesthe greediness of TCP flows. However, with OCSMA LA,the channel resources are more evenly distributed among bothflows, and the throughput of the flows is similar during theentire observation interval. We observed the same trend evenwhen the CW size is increased. Thus, OCSMA LA providesbetter fairness in situations involving inter-flow contention.

We evaluate short-term fairness by computing the averageJain’s fairness index based on the throughputs over consecutive10 s windows. The short-term fairness indices are f802.11 =0.50 and fOCSMA LA = 0.99. A fairness index of 0.5 underIEEE 802.11 indicates that one of the flows is completelydeprived of the channel resources. Under OCSMA LA, thefairness index is very close to unity, indicating a fair allocationof the channel resources. (The long-term fairness of both pro-tocols is close to 1.) When UDP traffic is sent over the networktopology in Fig. 11(b), we do not see the same problems withone flow capturing all of the channel resources. The short-term fairness measures under UDP traffic, f802.11 = 0.90 andfOCSMA LA = 0.92, show that fairness is not significantlycompromised with UDP traffic in this topology.

VI. CONCLUSION

We investigated the impact of overlapped transmissionson the throughput of TCP traffic in multihop networks withlinear flows. Through network simulations, we analyzed theinteractions between the OCSMA and TCP protocols. Weshow that flow control in TCP limits the availability of packetsfor overlapped transmission, which greatly reduces the perfor-mance of the OCSMA protocol. This is the “packet starvation”problem that we then try to address. We first modify some keyparameters at the MAC and transport layers but find that thefull potential of OCSMA is still not realized because of lost

10

opportunities for overlapped transmission. Thus, we modifiedthe OCSMA protocol to provide a look-ahead feature thatattempts to reserve the channel for the primary receiver toact as a transmitter after the completion of an overlappedtransmission. The resulting protocol, OCSMA LA, reducespacket starvation and improves the end-to-end performanceby up to 41% for TCP traffic and up to 126% for UDP trafficin a linear network. “Look ahead” and the other approachesconsidered in this paper can be applied to other overlappedtransmission schemes, such as PNC, ANC, and COPE, toimprove their performance with TCP flows. We also showedthat the OCSMA protocols offers significantly better inter-flowfairness for some topologies that cause poor performance forTCP flows with the IEEE 802.11 MAC protocol.

REFERENCES

[1] S. Verdu, Multiuser Detection. Cambridge Univ. Press,1998.

[2] S. Ghez, S. Verdu, and S. C. Schwartz, “Stability prop-erties of slotted Aloha with multipacket reception capa-bility,” IEEE Trans. Automat. Contr., pp. 640–649, July1988.

[3] J. Q. Bao and L. Tong, “Performance analysis of slottedAloha random access ad-hoc networks with multipacketreception,” in Proc. 1999 IEEE Military Commun. Conf.,vol. 1, Atlantic City, NJ, Nov. 1999, pp. 251–255.

[4] L. Tong, Q. Zhao, and G. Mergen, “Multipacket receptionin random access wireless networks: from signal process-ing to optimal medium access control,” IEEE Commun.Mag.,, vol. 39, no. 11, pp. 108–112, Nov. 2001.

[5] G. Mergen and L. Tong, “Receiver controlled mediumaccess in multihop ad hoc networks with multipacketreception,” in Proc. 2001 Military Commun. Conf., vol. 2,McLean, VA, Oct. 2001, pp. 1014–1018.

[6] Y. Sagduyu and A. Ephremides, “A game-theoretic lookat throughput and stability in random access,” in Proc.Military Commun. Conf., Oct. 2006, pp. 1–7.

[7] H. Sadjadpour, Z. Wang, and J. Garcia-Luna-Aceves,“The capacity of wireless ad hoc networks with multi-packet reception,” Communications, IEEE Transactionson, vol. 58, no. 2, pp. 600–610, Feb. 2010.

[8] J. J. Garcia-Luna-Aceves, H. R. Sadjadpour, andZ. Wang, “Challenges: towards truly scalable ad hocnetworks,” in Proc. ACM Int. Conf. Mobile Computingand Networking (MobiCom), 2007, pp. 207–214.

[9] J. Tranquilli, J. Farkas, J. Niedzwiecki, B. Pierce,L. Brothers, and J. DeBardelaben, “Real time implemen-tation of a multiuser detection enabled ad-hoc network,”in Proc. IEEE Military Commun. Conf., Nov. 2008, pp.1–6.

[10] S. Boppana and J. M. Shea, “Overlapped transmissionin wireless ad hoc networks,” in Proc. 2006 Int. Conf.Commun. Circuits and Systems, vol. 2, Guilin, China,Jun. 2006, pp. 1309–1314.

[11] S. Boppana, M. Sivakumar, and J. M. Shea, “The over-lapped carrier-sense multiple access (OCSMA) protocol,”in Proc. IEEE Military Commun. Conf., Orlando, FL,Oct. 2007, pp. 1–7.

[12] S. Boppana and J. M. Shea, “Overlapped carrier-sensemultiple access (OCSMA) in wireless ad hoc networks,”IEEE Trans. Mobile Computing, vol. 8, no. 3, pp. 369–383, Mar. 2009.

[13] S. Zhang, S. C. Liew, and P. Lam, “Physical-layernetwork coding,” in ACM MobiCom ’06, Los Angeles,CA, Sep. 2006, pp. 358–365.

[14] S. Katti, S. Gollakota, and D. Katabi, “Embracing wire-less interference: analog network coding,” in ACM SIG-COMM ’07, Kyoto, Japan, Aug. 2007, pp. 397–408.

[15] Y. Wu, P. A. Chou, and S.-Y. Kung, “Informationexchange in wireless networks with network codingand physical-layer broadcast,” Microsoft Research, Tech.Rep., Aug. 2004.

[16] ——, “Information exchange in wireless networks withnetwork coding and physical-layer broadcast,” in Proc.Conf. Inform. Science and Systems, Baltimore, March2005.

[17] S. Katti, D. Katabi, W. Hu, H. Rahul, and M. Medard,“The importance of being opportunistic: Practical net-work coding for wireless environments,” Proc. AllertonConf. Commun., Control, and Computing, 2005.

[18] S. Katti, H. Rahul, W. Hu, D. Katabi, M. Medard,and J. Crowcroft, “XORs in the air: practical wirelessnetwork coding,” in ACM SIGCOMM ’06, Pisa, Italy,Sep. 2006, pp. 243–254.

[19] S. Zhang, S. C. Liew, and P. P. Lam, “Physicallayer network coding,” 2007. [Online]. Available:http://arxiv.org/abs/0704.2475

[20] IEEE 802.11 WLAN Committee, “IEEE standard for in-formation technology-telecommunications and informa-tion exchange between systems-local and metropolitanarea networks-specific requirements - part 11: WirelessLAN medium access control (MAC) and physical layer(PHY) specifications,” IEEE Std 802.11-2007, June 122007.

[21] K. Fall and K. Varadhan, Eds., The ns Manual. TheVINT Project, 2008.

[22] J. Li, C. Blake, D. S. J. D. Couto, H. I. Lee, andR. Morris, “Capacity of ad hoc wireless networks,” inACM MobiCom ’01, Rome, Italy, Jul. 2001, pp. 61–69.

[23] K. Chen, Y. Xue, and K. Nahrstedt, “On setting TCP’scongestion window limit in mobile ad hoc networks,” inProc. 2003 IEEE Intl. Conf. Commun., Seattle, WA, May2003, pp. 1080–1084.

[24] R. Braden, “Requirements for Internet Hosts -Communication Layers,” RFC 1122 (Standard), Oct.1989, updated by RFCs 1349, 4379. [Online]. Available:http://www.ietf.org/rfc/rfc1122.txt

[25] C.-K. Toh, V. Vassiliou, G. Guichal, and C.-H. Shih,“MARCH: a medium access control protocol for mul-tihop wireless ad hoc networks,” in Proc. 2000 IEEEMilitary Commun. Conf., Los Angeles, CA, October2000, pp. 512–516.

[26] R. Jain, D. M. Chiu, and W. Hawe, “A quantitative mea-sure of fairness and discrimination for resource allocationin shared systems,” DEC-TR-301, Tech. Rep., Sep. 1984.

11

Surendra Boppana (S’02) received the B.Tech. de-gree in electronics and communication engineeringin 2003 from the Indian Institute of Technology(IIT), Guwahati, India and the M.S. and Ph.D.degrees in electrical and computer engineering fromthe University of Florida, Gainesville, in 2005 and2008, respectively. From May 2006 until January2007 he was a Graduate Technical Intern withthe Communications Circuit Lab, Intel Corporation,Hillsboro, Oregon. Dr. Boppana is currently a SeniorEngineer at Qualcomm, Inc., San Diego, California.

His research interests include wireless communications, information theory,and cross-layer design.

John M. Shea (S’92-M’99) received the B.S. (withhighest honors) in computer engineering from Clem-son University in 1993 and the M.S. and Ph.D.degrees in electrical engineering from Clemson Uni-versity in 1995 and 1998, respectively.

Dr. Shea is currently an Associate Professor ofelectrical and computer engineering at the Univer-sity of Florida. Prior to that, he was an AssistantProfessor at the University of Florida from July 1999to August 2005 and a post-doctoral research fellowat Clemson University from January 1999 to August

1999. He was a research assistant in the Wireless Communications Program atClemson University from 1993 to 1998. He is currently engaged in researchon wireless communications with emphasis on cross-layer protocol design,physical-layer security, cooperative diversity techniques, and applications oferror-control coding.

Dr. Shea was selected as a Finalist for the 2004 Eta Kappa Nu OutstandingYoung Electrical Engineer Award. Dr. Shea was a National Science Foun-dation Fellow from 1994 to 1998. He received the Ellersick Award from theIEEE Communications Society for the Best Paper in the Unclassified Programof the IEEE Military Communications Conference (MILCOM) in 1996. Hehas been an Editor for the IEEE Transactions on Wireless Communicationssince 2008 and an Editor for IEEE Wireless Communications magazine since2010. He was an Associate Editor for the IEEE Transactions on VehicularTechnology from 2002 to 2007. He was the Technical Program Chair for theUnclassified Program of MILCOM in 2010.