A Concentric Network Algorithm for Spatial Reuse in Networked Robotics

Victor Kuo and Robert FitchARC Centre of Excellence for Autonomous Systems

Australian Centre for Field Robotics (ACFR)The University of Sydney, NSW Australia{v.kuo, rfitch}@acfr.usyd.edu.au

AbstractExisting wireless network architectures are oftenpoorly suited to the growing diversity of up-comingnetworked robot systems. As networked robots de-pend on reliable wireless communication to operateeffectively as a team of networked robots, it is im-portant that the communication system is scalable.One of the existing challenges in wireless commu-nication is maintaining the bandwidth throughputof the network as the system size scales upward. Inthis paper we demonstrate a new method of apply-ing spatial reuse to improve bandwidth in our pro-posed network architecture and topology. We alsosuccessfully demonstrate experimental evidence ofour network architecture in both static and mobilescenarios using custom hardware and off-the-shelfradios.

1 IntroductionNetworked robots rely on a communication network for in-formation sharing and cooperation. An important class of au-tonomous networked robot systems uses wireless communi-cation. Although research in wireless communication is ex-tensive, networked robots present new challenges. We areinterested in the challenge of maintaining the level of band-width available between teams of robots in a network as thetotal number of robot teams increases.

It is envisioned that networked robots will play a criticalrole in future applications such as environmental monitoring,reconnaissance, and tele-operation. To support these applica-tions, the network needs to be scalable and provide efficientuse of the limited available bandwidth in wireless networks.In the case of video data, this bandwidth requirement is sig-nificant. However, standard wireless mesh networks do notsupport this requirement. As the number of robots increases,pair-wise bandwidth decreases [Gupta and Kumar, 2000]. Weare interested in the question of how to design bandwidth-aware architectures that address this problem.

In this paper we investigate an approach for achieving spa-tial reuse using a concentric network topology pictured in

Network Node

A1

B1

A2

B2 B3

A3

xR

x2R

R

Figure 1: Concentric network topology. R represents the baseradius and x denotes the increase in radius for each subse-quent circle. Ai and Bi represent lettered nodes on circle i.

Fig. 1. The concept of spatial reuse is to allow as many si-multaneous radio transmissions to occur as possible in a givenarea without causing corrupted transmissions. One of the lim-iting properties of wireless communication is that two radiostransmitting on the same frequency run the possibility of cor-rupting each other’s data at their respective receivers. In or-der for a pair of radios to communicate successfully, the re-ceiving radio must be able to receive the intended signal at ahigher power level than all other interfering signals and am-bient noise. Spatial reuse is important in wireless networks asit determines the bandwidth capacity of a network for a givenarea. Various strategies have been developed and investigatedto this end which we visit in Sec. 2. However existing spa-tial reuse strategies are unable to operate efficiently in ourproposed topology where interfering transmitters are locatedcloser to receiving radios than the intended signal transmit-ters. This is illustrated in Fig. 2 where S1 transmitting toa receiver R1 would prohibit S2 from transmitting to R2 inmost spatial reuse strategies. If S2 were to transmit concur-

R1 S1 S2 R2

Figure 2: Typically S2 cannot transmit to R2 whilst S1 istransmitting to R1. Our concentric network architecture al-lows for this.

rently, it would unknowingly interfere with the transmissionfrom S1 to R1. In addition, S1 would corrupt any transmis-sion from S2 to R2.

The concentric network is a special topology that allows usto achieve spatial reuse. There is no a priori application forsuch a topology; the idea is that a future networked roboticsapplication can realise the benefits of spatial reuse by adopt-ing the concentric ring topology in some way. The networkoperates on a single frequency and allows nodes on the sameconcentric ring to communicate with each other without in-terfering with the communication of radios on other networkrings. This is achieved by increasing the transmit power foreach successive ring such that radios on the outer rings havehigher transmit powers than radios on the inner rings. Thisallows all radios on the same ring to achieve the necessarySignal to Interference Ratio (SIR) to communicate with eachother without interfering with the transmissions of radios onthe other rings. The significance of our concentric networkarchitecture is that it solves the spatial reuse problem illus-trated in Fig. 2 in an efficient and simple manner and alsoworks when nodes are mobile. Furthermore nodes do notneed to know the exact location of other nodes in the samenetwork ring.

The paper is organised as follows. Sec. 2 discusses re-lated work in the area of wireless networks and spatial reuse.In Sec. 3 we formulate the equations that make our networktopology possible and describe the basic operation of the net-work architecture. In Sec. 4 we present experiments withcustom hardware and IEEE 802.15.4 radios that successfullyshow hardware validation for the presented network topologyand architecture. Our experiments are also executed in a re-alistic working environment with both stationary and mobilenetwork nodes. In Sec. 5 we provide suggestions for futureimprovements and conclude in Sec. 6.

2 Related WorkWhilst our network topology and architecture are novel, im-proving spatial reuse is a widely studied problem in many dif-ferent types of wireless networks. In wireless mesh networks(WMN’s), various strategies for improving spatial reuse havebeen studied and developed. In [Monks et al., 2001; Son etal., 2004; Wattenhofer et al., 2001; Yang and Vaidya, 2005;Tseng et al., 2001; Zhu et al., 2004; Kim et al., 2006;Zhou and Nettles, 2005], the carrier sense threshold and/or

power levels are tuned to optimise the trade-off between hid-den node collisions and permitting exposed nodes to trans-mit. However when applied to our topology, these strategieswould be unable to take advantage of the relative separationbetween network rings and hence miss opportunities to trans-mit concurrently.

In [Seung et al., 2009; Nadeem and Lusheng, 2007], lo-cation information such as GPS is used to optimise spatialreuse. However GPS does not work in all environmentsand the location information itself needs to be shared andupdated at regular intervals which adds to network over-head. In [Acharya et al., 2004; Vutukuru et al., 2008;Eisenman and Campbell, 2007] instead of using location in-formation, nodes actively build a map of nearby radios bylistening into nearby transmissions. By checking the sourceand destination addresses, a node can build a map of imme-diate neighbours and 1-hop neighbours allowing the node toidentify opportunities to transmit in parallel. These methodsare better suited to static networks as mobile networks maybe too dynamic for the maps to be useful.

The basis of spatial reuse is the capture effect which is thephenomenon where a receiver can correctly decode a packeteven in the presence of other concurrent co-channel transmis-sions. In analogue transmissions such as AM radios, interfer-ing signals on the same channel are actually merged togetherat the receiver resulting in the receiver hearing the two trans-missions overlayed on top of each. This however is not thecase in digital radios where the weaker of the two signals iscompletely suppressed as is observed in [Kochut et al., 2004;Ware et al., 2000; Lee et al., 2007]. In the capture effect, a ra-dio that is in the process of decoding a weaker signal can stillcorrectly capture and decode the data from a second strongersignal. Mechanisms for boosting the chances of packet cap-ture are also proposed in [Whitehouse et al., 2005].

3 Proposed Network ArchitectureThe concentric network topology consists of radio nodesplaced on circles sharing a common origin as illustrated inFig. 1. In the concentric network, nodes that reside on thesame circle are able to communicate with each other withoutinterfering with the communication of nodes on other circleswhilst operating on the same channel. As a result, the com-munication between concentric rings in the network are inde-pendent of each other in terms of bandwidth. For example inFig. 1, nodes A1 and B1 can communicate with each other atthe full data-rate of the radios whilst nodes A2 and B2 com-municate on the same channel and also at the full data-rate.Typically this is not possible as the transmissions between A2and B2 would interfere with the transmissions between A1and B1. Instead in a typical wireless network, the bandwidthwould have to be shared between the two pairs of radios.

Our concentric network topology achieves communicationconcurrency by exploiting the fact that radio signals decaywith the inverse-square of distance. This concept is analogous

to speaking to someone sitting next to you in a concert hall.During a concert you can easily talk to the person next to youeven if there are loudspeakers in the background because yourvoice only has to travel a short distance and hence remainslouder than the loudspeakers. At the same time your conver-sation is too soft to affect others around you from hearing theloudspeakers. By spacing network rings appropriately and in-creasing the transmit power of radios on successive rings, weachieve a network where the signal to interference ratio fromnodes within the same ring is always stronger than signalsfrom adjacent rings.

Using the Free Space Path Loss (FSPL) equation shownin Eqn. 1 we can calculate the necessary separation betweenconcentric circles and the relative transmit power levels. TheFSPL equation describes the rate at which radio signals decayin free space:

RXx,y = TXPy − 20 log dxy − k (1)

where RXx,y is the received signal power of a transmissionfrom y at x in decibels, TXPy is the transmit power of y, dxy

is the absolute distance between x and y and k is a constant.Looking at Fig. 1 we want to determine the necessary sep-

aration required between network rings such that a transmis-sion between A2 and B2 will be stronger than the transmis-sions from A3 and B3 when received at A2 and B2. At thesame time we also need to satisfy the requirement that thesignals between A3 and B3 are stronger than the interferencefrom A2 at A3 and from B2 at B3. We write these two re-quirements in Eqn. 2 and Eqn. 3 which we can then graph tofind the minimum transmit power and separation ratios.

RXA3,B3 > RXA3,A2 + CCRR (2)

RXA2,B2 > RXA2,A3 + CCRR (3)

In Eqn. 2 and Eqn. 3 the Co-Channel Rejection Ratio(CCRR) represents the minimum signal to noise ratio (SNR)that is required by the radio hardware to correctly decode asignal from any background interference or noise. Other per-mutations of Eqn. 2 and Eqn. 3 also exist but since the linksare symmetrical they are equivalent. We can then substituteEqn. 1 into Eqn. 2 and Eqn. 3 which provides us with equa-tions Eqn. 4 and Eqn. 5. In Eqn. 4 and Eqn. 5 we assume thatassume the transmit power of radios on common rings areequal such that TXPA2 = TXPB2 and TXPA3 = TXPB3.

TXPA3 − TXPA2 > 20 logdA3,B3

dA3,A2+ CCRR (4)

TXPA2 − TXPA3 > 20 logdB2,A2

dA3,A2+ CCRR (5)

If we now let TXPA2 be the reference power level of0dBm then we can determine the relative transmit power levelfor the next concentric ring (ie. TXPA3). We also convert the

distances between the radios to radial distance from the centreof the network to get Eqn. 6 and Eqn. 7.

TXPA3 > 20 log2x2R

x2R− xR+ CCRR

> 20 log2x

x− 1+ CCRR (6)

TXPA3 < −20 log2xR

x2R− xR− CCRR

< −20 log2

x− 1− CCRR (7)

In our discussion thus far we have assumed that the radiosare co-linear with the centre of the network. Basic geometryshows that this is the worst case scenario where the radios onthe same ring are furthest apart from each other whilst beingas close as possible to radios on the adjacent ring. Any otherplacement of radios would yield a higher separation ratio be-tween the signal transmitter and interfering transmitter andhence result in a higher SNR than calculated. Hence Eqn. 6represents the lower bound on the transmit power of the ra-dios on the outer network ring for varying separation ratios xand similarly Eqn. 7 represents the upper bound. The separa-tion ratio x in Eqn. 6 and Eqn. 7 is the ratio of the outer ringradius to the inner ring radius. We graph these two equationsin Fig. 3 for two different CCRR values.

The solid line in Fig. 3 represents Eqn. 7 which is the up-per bound and the broken line represents Eqn. 6 which is thelower bound. From Fig. 3 it clear there is a minimum sepa-ration ratio between two adjacent network rings that must bemet for Eqn. 6 and Eqn. 7 to hold true. There is also a cor-responding minimum relative transmit power correspondingto the minimum separation ratio. In Fig. 3 the intersectionof the two blue lines represents the minimum separation ra-tio and relative transmit power for a CCRR of 6dB and the

0 5 10 15 20 25 30 35 40 45 50-10

-5

0

5

10

15

20

25

30

Separation Ratio (x)

Tran

smit

Pow

er (d

B)

Minimum Separation Ratio and Transmit Power

Minimum Separation Ratio = 8.22Minimum Transmit Power = 9.15dB

at 2dB CCRR

Transmit Power Lower Bound

Transmit Power Upper Bound

Minimum Separation Ratio = 17.87Minimum Transmit Power = 12.52dB

at 6dB CCRR

Figure 3: Graph of separation ratio and transmit power re-quired between concentric circles.

r R

Network Node

Data Relay Path

Figure 4: Ring relay network for large concentric networks.

intersection of the two red lines represents the minimum sep-aration ratio and relative transmit power for a CCRR of 2dB.Relating this back to our network architecture in Fig. 1, if wechoose a CCRR of 6dB for our radios then x must be 17.87and radios at radius xR must transmit at 12.52dB relative toradios at radius R. And in the same way, radios at radius x2Rmust transmit at 12.52dBm relative to radios at radius xR.

3.1 Large Concentric NetworksAs the size of the network increase it is inevitable that thesize of the network grows larger than the transmission rangeof the radios. This occurs when the diameter of the networkring is greater than the transmission radius r of the radios. Inthis case nodes on the same ring may not necessarily be ableto communicate with each other directly. Instead messagesneed to be relayed via intermediate nodes on the same net-work ring in a similar fashion to ring networks. Each networkring must also contain enough nodes such that a continuouschain of relay nodes can be formed as illustrated in Fig. 4.Using simple geometry we can determine that the minimumnumber of nodes per ring in this case is n = 180

sin−1 r2R

whereR is the radius of the network ring. The minimum separationbetween each network ring is given by

d

r> 10

CCRR20 (8)

where d radius difference between two adjacent rings and ris the transmission radius. Here we make the assumption thatall radios on the outer rings transmit at the maximum powerlevel. In order to implement a ring network on a wireless net-work, networking protocols are also required. Whilst this isout of the scope of this paper, there are many readily avail-able wireless networking protocols that would demonstrategood performance.

3.2 Limited Transmit Power RangeDue to the generally limited dynamic range of transmit powerlevels on typical off-the-shelf radios, it is often impossible to

increase the transmit power to the required levels. This how-ever does not prevent the implementation of the concentricnetwork. The problem is easily overcome by reducing thedistance between inter-ring nodes and forcing data to flowaround the circumference of the ring in the same way as de-scribed in Sec. 3.1. By doing so we decrease the distance be-tween intra-ring nodes and whilst maintaining the minimumdistance to inter-ring nodes hence improving the SIR betweenintra-ring nodes. The disadvantage of this method is that theremust be enough radios to form a connected chain around theentire circumference of the network ring. There must also bea minimum separation between adjacent rings which is de-scribed by Eqn. 8. In this case d represents the separationbetween network rings and r represents the transmission ra-dius of the radios. If the transmission radii are not equal thenr is taken as the larger of the two. As can be seen by Eqn. 8the required intra-ring separation is significantly smaller thandescribed in Eqn. 6 and Eqn. 7 for the same CCRR.

3.3 Filling Inter-Ring GapIn Fig. 3 we see that the minimum separation between ringscan be quite large particularly for large CCRRs. If this isthe case and it is necessary to operate nodes within the gapthen two options are available. Firstly we can operate thenetwork in the gap on an orthogonal channel. Since radiotransmissions on orthogonal channels do not interfere witheach other, we can simply ignore the presence of the othernetwork.

In the case that there are insufficient orthogonal channelson the radio hardware, rings can still be placed closer togetherprovided that direct communication is not required across theentire ring. Information can be relayed around the circum-ference of the network in the same manner as discussed inSec. 3.1 and Sec. 3.2. However unlike using orthogonal chan-nels, there is still a minimum separation distance required be-tween adjacent rings using the same channel which is alsodescribed by Eqn. 8. Similarly d represents the separationbetween network rings and r represents the larger of the trans-mission radii of the radios. We can also see that the densityof the network is a function of the transmission radius of theradios. As the transmission radius increases then the networkdensity decreases.

3.4 Inter-Ring CommunicationCommunication between network rings is also possiblethrough software protocols provided that the inter-ring sep-aration does not exceed the transmission radius of the radiosin both network rings. To allow for inter-ring communica-tion, intra-ring signals must cease for the duration of the datatransfer. Since intra-ring signals are received at higher energylevels than inter-ring signals, intra-ring transmissions wouldcause data corruption. In Fig. 1 if node A2 on ring two wishesto transmit to node A3 on ring three then all other nodes onrings two and three must not transmit for that period. Further-

more node A2 would also have to raise it’s transmit power sothe signal can reach node A3.

Inter-ring communication can also occur between non-adjacent network rings. For example node A3 in Fig. 1 canalso exchange data with node A1 in the same manner as dis-cussed already. However in this case all network rings be-tween the source and destination pair must also cease trans-mission. As we can see inter-ring communication proceeds inthe same manner as a wireless mesh network where a trans-mitting node blocks all of its neighbours from transmitting si-multaneously and the network bandwidth must then be sharedamongst all radios within transmission range of each other.

Inter-ring transfers can be initiated on an ad-hoc or pre-scheduled basis depending on the requirements of the net-work application. Ad-hoc initiations would require a sourcenode transmitting to an inter-ring node destination by in-creasing its transmit power or waiting for a gap in transmis-sions and notifying the destination node of an incoming inter-ring packet. Pre-scheduled transmissions would require net-work nodes cease transmissions for a pre-determine period oftime allowing the source and destination nodes to completethe data transfer without interruption. Whilst pre-scheduledtransmissions guarantee successful inter-ring transmissions,they also block all other transmissions from occurring andhence should be scheduled as infrequently as possible. Ad-hoc based initiations accommodate this however cannot guar-antee that a connection can be made depending on the inter-ring separation and available transmit power range.

A third alternative to initiating inter-ring communication isthe use of an alternate orthogonal channel that is reserved forinter-ring communication. Nominated nodes in each networkring can switch to this alternate channel at pre-negotiated in-tervals to communicate with other network-rings and estab-lish whether any data needs to be exchange. The disadvantageof this however is the poor scalability of the alternate channelif there is high inter-ring network traffic.

4 Hardware ExperimentsIn Sec. 4.1 we describe the hardware that was employed inour experimental setup. The static environment experiment inSec. 4.2 shows the basic operation of the network with nodesplaced at fixed positions. A second experiment in Sec. 4.3demonstrates that the network also works for nodes movingaround their designated network ring. Each experiment in-cludes a control condition.

4.1 Hardware ImplementationIn our experiments we chose to use IEEE 802.15.4 radios.Other radio standards such as IEEE 802.11 are also suit-able radio candidates however the IEEE 802.15.4 standardis significantly simpler and allowed faster development time.These radios operate in the 2.4GHz spectrum and have ac-cess to 16 orthogonal channels. Each radio is connected to

(a) (b)

Figure 5: (a) Microprocessor board containing theIEEE802.15.4 Xbee module and SD Card logging. (b) Mi-croprocessor board containing the IEEE802.15.4 Microchipmodule.

custom designed microprocessor boards which ran the neces-sary software and also contained an onboard Micro SD cardfor logging the received data. Two different brands of ra-dios were used for transmitting and receiving. The MicrochipMR24J40 radios as seen in Fig. 5b were used for transmis-sion as they had a larger range of transmit power levels rang-ing from 0dBm to -36dBm. However due to the poor CCRRof these radios we chose Digi Xbee radios for receiving datawhich has a CCRR of 6dB as seen in Fig. 5a.

4.2 Static NetworkThe static experiment consists of 6 transceiver radio pairsconfigured as 6 separate network rings and operating on twoorthogonal channels as shown in Fig. 6. The figure showsone sub-network of 3 rings operating on channel 17 and an-other sub-network operating on channel 12. The placementof the transmitter-receiver pairs is such that we test the worstcase scenario where the transmitting radios in one ring areas close as geometrically possible to the receiving radio onthe adjacent ring. All other radio placements would yield alarger separation ratio and hence better results. As the twosub-networks operate on different channels we demonstratethat the inter-ring spacing can easily be filled with radios op-erating on orthogonal channels.

Our network testbed was setup in the downstairs office areaat the ACFR building. This environment was ideal as we wereable to place the radios along the top of the cubicle partitionswhich elevated the radios 1.6m off the ground to a commonheight. The office environment is also sufficiently clutteredto make the test bed environment realistic in terms of channelfading properties.

As our Digi Xbee radios have a CCRR of 6dB, theoreti-cally we would require a minimum separation ratio of 17.87and transmit power step size of 12.52dB as seen in Fig. 3. In-stead however we use transmit power step size of 10dB and

0.1m – Ring 1 Tx Power = -20dBm

1m – Ring 2 Tx Power = -10dBm

10m – Ring 3 Tx Power = 0dBm

20m – Ring 3 Tx Power = 0dBm

2m – Ring 2 Tx Power = -10dBm

0.2m – Ring 1 Tx Power = -20dBm

Legend

Transmitter Ch12 Receiver Ch12 Transmitter Ch17 Receiver Ch 17

Figure 6: Placement of radio nodes in static network test.

Control Test Concentric Network0

0.2

0.4

0.6

0.8

1Static Network Test (Channel 17)

Nor

mal

ised

Thr

ough

put (

pkts

/s)

Ring 1(Inner) Ring 2 (Middle) Ring 3 (Outer)

(a)

Control Test Concentric Network0

0.2

0.4

0.6

0.8

1Static Network Test (Channel 12)

Nor

mal

ised

Thr

ough

put (

pkts

/s)

Ring 1 (Inner) Ring 2 (Middle) Ring 3 (Outer)

(b)

Figure 7: (a) Normalised throughput of network on chan-nel 17. Control Test group shows the throughput when allnetwork rings are set to the same transmit power. Concen-tric Network group uses the stepped transmission powers of10dBm. (b) Normalised throughput of network on channel12.

a separation ratio of 10 as seen in Fig. 6. Due to limited res-olution of the transmit power settings on our radios, 10dBmwas the closest available power level below 12.52dB henceslightly handicapping our experiment. We also used a sig-nificantly smaller separation ratio of 10 as we had to fit theexperiment in the confines of the test bed. Fortunately wewere still able to successfully carry out the experiment as thepath-loss exponent in an office environment is greater thanthe path-loss exponent of free space from which we basedour equations on.

To test the performance of the network the transmit radiosshown in blue in Fig. 6 continuously broadcasts dummy dataaddressed to the corresponding receiver radio at the full datarate of the hardware. The receiver radios then log the num-ber of packets addressed to themselves and received withouterror. The results for this experiment are shown in Fig. 7aand Fig. 7b for channel 17 and 12 respectively. Each bar inFig. 7 represents the normalised throughput rate for a partic-ular network ring featured in Fig. 6. The throughput rates areaveraged over a 20 minute period and normalised relative toa benchmark data rate of 146 packets/second. This bench-mark data rate is obtained by continuously sending dummydata from a transmitter to a receiver on a clear channel andseparated by a meter. The clear channel assessment was donewith an Integration Zigbee dongle which measured the ambi-ent noise across all 16 Zigbee channels.

For comparison we also do a control test to demonstratethat a naive implementation does not allow concurrent trans-missions. The control experiment uses the same layout andprocedure as described above except all radios are set thesame transmit power level of 0dBm. During the control test,we first validate that each pair of radios communicates nor-mally with all other radios switched off before turning on allthe radios. The results of the naive implementation are shown

(a)

Mobile

Ring with

iRobot Create

Stationary Ring

R0.09m

R1.5m

Transmitting

at -10dBm

Transmitting

at -20dBm

(b)

Figure 8: (a) Picture of iRobot Creates in mobile networksetup. (b) Layout of radios in mobile network setup.

alongside the actual results in Fig. 7a and Fig. 7b.At first glance, a comparison of the results with the con-

trol test shows that our proposed architecture performs sig-nificantly better. Whilst the inner most pair of radios in thecontrol test shows no throughput degradation, all subsequentnetwork rings show a throughput of 0 which is expected andreflects what happens in typical radio networks.

Our concentric network throughput on the other handshows a good level of throughput for all three networkrings. Whilst we do see some slight throughput degrada-tion compared to the benchmark throughput rate of 146 pack-ets/second, overall the throughput remains above 90% of the

benchmark except for one radio pair that achieved 80%. Thereduced throughput may be attributed to the reduced separa-tion ratio, smaller transmit power step size and from externalwireless networks operating in the 2.4GHz range such as theoverhead WLAN network. Overall results validate the basicprinciple of operation of our concentric network topology andarchitecture.

4.3 Mobile NetworkThe mobile network experiment consists of three separate ex-periments: a control test, a manual mobile test and a mo-bility test with iRobot Creates. All three tests consists oftwo transceiver pairs as illustrated in Fig. 8b. The static pairsits at the centre of the network with separation of 9cm andthe mobile pair are situated at a radius of 1.5m. In the firstcontrol experiment we demonstrate that a naive implemen-tation does not allow concurrent transmissions by setting allradios to the same transmit power level. For the control testwe do not use the iRobot creates but instead manually movethe outer ring radios around the inner pair. The outer pair iskept diametrically opposite at a radius of 1.5m and moved45 degrees every minute. The results for the control experi-ment, as seen in Fig. 9, shows that the static inner pair has nothroughput losses as we expect. However, since concurrenttransmissions cannot proceed under a naive implementationthe mobile outer pair has 0 throughput.

In the manual mobility test, we repeat the control exper-iment except this time we implement the stepped transmis-sions powers as per our network architecture. The mobile pairis set to transmit at−10dBm whilst the inner pair of radios re-main at −20dBm from the control test. The results shown inFig. 9 clearly illustrates that both pairs of radios, transmittingon the same channel, are able to communicate simultaneouslywithout any throughput degradation in contrast to the controltest.

Finally we repeat the experiment but this time mount the

Control Test Manual Mobility iRobot Create0

0.2

0.4

0.6

0.8

1

Mobile Network Test

Nor

mal

ised

Thr

ough

put (

pkts

/s)

Static Inner Ring Mobile Outer Ring

Figure 9: Throughout in a mobile network under a naive im-plementation and using stepped transmission powers.

mobile outer ring radios onto two iRobot Creates which areprogrammed to circle around the inner ring at a radius of1.5m. The results in Fig. 9 shows that the inner static ring per-forms very close to the benchmark throughput rate. Howeverthe mobile ring does not perform as well with approximately20% throughput losses over 5 consecutive 2 minute trial runs.We believe that the throughput losses can be attributed to thefact that the iRobot Creates had difficulty maintaining a 1.5mradius and frequently dropped below a radius of 0.5m. Never-the-less, the experiments demonstrate that exact location ofradios does not effect the throughput as long as the radiosremain on their designated network ring. This is a signifi-cant result as most of the spatial reuse strategies discussed inSec. 2 deal poorly with mobility.

5 Future DirectionsThus far our discussion of our network topology and architec-ture has been highly idealistic. We have assumed that wire-less signals decay according to the FSPL equation, radioshave circular transmission ranges and communication linksare symmetrical. In reality the propagation of radio wavesin the real world is extremely difficult to model due to issuessuch as multi-path fading, deep fading and antenna anisotropy[Kotz et al., 2003]. This is of particular importance to ournetwork as we depend on uniform decay of signals to ensurethat receiver radios have a sufficient SIR to operate. In or-der to bridge the gap between theory and practice we believethat a practical implementation should use antenna diversityin the radio hardware. Antenna diversity involves implement-ing multiple antennas onto a single radio separated by frac-tions of a wavelength to combat the effects of multi-path anddeep fading. Radio hardware with low and stable CCRR’sshould be chosen as well as a radio standard that offers largernumbers of orthogonal channels. In addition to this, separa-tion ratios should be made larger than necessary to provide ahigher SIR.

6 DiscussionWe successfully demonstrated the proposed concentric net-work topology and architecture with 12 off-the-shelf radios inan indoor office environment with the experiments perform-ing as predicted. Our network topology demonstrates a novelmethod of exploiting the inverse-square decay of radio sig-nals to improve spatial reuse in a radio network and increasenetwork throughput. We also demonstrate that our networkarchitecture is inherently suited to mobile applications wherenetwork nodes orbit a common point. Our network architec-ture proposal covers several basic scenarios that can proveuseful in practical applications and we qualitatively analysedthe issues associated with migrating to a practical implemen-tation.

AcknowledgmentsThis work is supported by the Australian Research Council(ARC) Centre of Excellence programme, funded by the ARC

and the New South Wales (NSW) State Government. Wewould like to thank Ritesh Lal for the support he providedwith the radio hardware.

References[Acharya et al., 2004] A. Acharya, A. Misra, and S. Bansal.

Design and analysis of a cooperative medium accessscheme for wireless mesh networks. In Proc. of BROAD-NETS, pages 621–631, 2004.

[Eisenman and Campbell, 2007] S.B. Eisenman and A.T.Campbell. E-csma: Supporting enhanced csma per-formance in experimental sensor networks using per-neighbor transmission probability thresholds. In Proc. ofIEEE INFOCOM, pages 1208 –1216, May 2007.

[Gupta and Kumar, 2000] P. Gupta and P. Kumar. The ca-pacity of wireless networks. IEEE Trans. on InformationTheory, 46(2):388–404, Mar 2000.

[Kim et al., 2006] T.-S. Kim, H. Lim, and J. Hou. Improvingspatial reuse through tuning transmit power, carrier sensethreshold, and data rate in multihop wireless networks. InProc. of MobiCom, pages 366–377, 2006.

[Kochut et al., 2004] A. Kochut, A. Vasan, A.U. Shankar,and A. Agrawala. Sniffing out the correct physical layercapture model in 802.11b. In Proc. of ICNP, pages 252 –261, Oct 2004.

[Kotz et al., 2003] D. Kotz, C. Newport, and C. Elliott. Themistaken axioms of wireless-network research. TechnicalReport TR2003-467, Dartmouth College, Computer Sci-ence, July 2003.

[Lee et al., 2007] J. Lee, W. Kim, S.-J. Lee, D. Jo, J. Ryu,T. Kwon, and Y. Choi. An experimental study on the cap-ture effect in 802.11a networks. In Proc. of WinTECH,pages 19–26, 2007.

[Monks et al., 2001] J. Monks, V. Bharghavan, and W. Hwu.A power controlled multiple access protocol for wirelesspacket networks. In Proc. of IEEE INFOCOM, volume 1,pages 219 –228, 2001.

[Nadeem and Lusheng, 2007] T. Nadeem and J. Lusheng.Location-aware ieee 802.11 for spatial reuse enhance-ment. Trans. on Mobile Computing, 6(10):1171 –1184,Oct 2007.

[Seung et al., 2009] M.-H. Seung, S. Mao, Y.-T. Hou, K.-H.Nam, and J.H. Reed. Exploiting location information forconcurrent transmissions in multihop wireless networks.IEEE Trans. on VTC, 58(1):314 –323, Jan 2009.

[Son et al., 2004] D. Son, B. Krishnamachari, and J. Hei-demann. Experimental study of the effects of transmis-sion power control and blacklisting in wireless sensor net-works. In Proc. of IEEE SECON, pages 289 – 298, Oct2004.

[Tseng et al., 2001] Y.-C. Tseng, S.-L. Wu, C.-Y. Lin, and J.-P. Sheu. A multi-channel mac protocol with power controlfor multi-hop mobile ad hoc networks. Int. Conf. on Dis-tributed Computing Systems Workshops, page 0419, 2001.

[Vutukuru et al., 2008] M. Vutukuru, K. Jamieson, andH. Balakrishnan. Harnessing exposed terminals in wire-less networks. In Proc. of NSDI, pages 59–72, 2008.

[Ware et al., 2000] C. Ware, J. Judge, J. Chicharo, andE. Dutkiewicz. Unfairness and capture behaviour in802.11 adhoc networks. In IEEE Int. Conf. on Commu-nications, volume 1, pages 159 –163, 2000.

[Wattenhofer et al., 2001] R. Wattenhofer, L. Li, P. Bahl, andY.-M. Wang. Distributed topology control for power effi-cient operation in multihop wireless ad hoc networks. InProc. of IEEE INFOCOM, volume 3, pages 1388 –1397,2001.

[Whitehouse et al., 2005] K. Whitehouse, A. Woo, F. Jiang,J. Polastre, and D. Culler. Exploiting the capture effectfor collision detection and recovery. In IEEE Workshop onEmNetS-II, pages 45 – 52, May 2005.

[Yang and Vaidya, 2005] X. Yang and N. Vaidya. On phys-ical carrier sensing in wireless ad hoc networks. In Proc.of IEEE INFOCOM, volume 4, pages 2525 – 2535, Mar2005.

[Zhou and Nettles, 2005] Y. Zhou and S.M. Nettles. Balanc-ing the hidden and exposed node problems with powercontrol in csma/ca-based wireless networks. In Proc. ofIEEE WCNC, volume 2, pages 683 – 688, Mar 2005.

[Zhu et al., 2004] J. Zhu, X. Guo, L.L. Yang, W.S. Conner,S. Roy, and M.H. Mousumi. Adapting physical carriersensing to maximize spatial reuse in 802.11 mesh net-works: Research articles. Wirel. Commun. Mob. Comput.,4(8):933–946, 2004.