Mitigating Radio Interference in Large IoT Networks .... Sparber, C.A. Boano, S.S. Kanhere, and K. Romer: Mitigating Radio Interference in Large IoT Networks through Dynamic CCA Adjustment¨

c© 2017 by the authors; licensee RonPub, Lubeck, Germany. This article is an open access article distributed under the terms and conditions ofthe Creative Commons Attribution license (http://creativecommons.org/licenses/by/4.0/).

Open Access

Open Journal of Internet of Things (OJIOT)Volume 1, Issue 1, 2017

http://www.ronpub.com/ojiotISSN 2364-7108

Mitigating Radio Interference in Large IoTNetworks through Dynamic CCA Adjustment

Tommy Sparber A, Carlo Alberto Boano A, Salil S. Kanhere B, Kay Romer A

A Institute for Technical Informatics, Graz University of Technology, Inffeldgasse 16/1, 8010 Graz, Austria,[email protected], [email protected], [email protected]

B School of Computer Science and Engineering, The University of New South Wales, Building K17, Gate 14,Barker Street, Kensington NSW 2052, Sydney, Australia, [email protected]

ABSTRACT

The performance of low-power wireless sensor networks used to build Internet of Things applications often suffersfrom radio interference generated by co-located wireless devices or from jammers maliciously placed in theirproximity. As IoT devices typically operate in unsupervised large-scale installations, and as radio interference istypically localized and hence affects only a portion of the nodes in the network, it is important to give low-powerwireless sensors and actuators the ability to autonomously mitigate the impact of surrounding interference. Inthis paper we present our approach DynCCA, which dynamically adapts the clear channel assessment thresholdof IoT devices to minimize the impact of malicious or unintentional interference on both network reliability andenergy efficiency. First, we describe how varying the clear channel assessment threshold at run-time using onlyinformation computed locally can help to minimize the impact of unintentional interference from surroundingdevices and to escape jamming attacks. We then present the design and implementation of DynCCA on top ofContikiMAC and evaluate its performance on wireless sensor nodes equipped with IEEE 802.15.4 radios. Ourexperimental investigation shows that the use of DynCCA in dense IoT networks can increase the packet receptionrate by up to 50% and reduce the energy consumption by a factor of 4.

TYPE OF PAPER AND KEYWORDS

Regular research paper: Clear Channel Assessment, CCA, Contiki, Internet of Things, radio interference, RPL

1 INTRODUCTION

Networks of low-power wireless sensors are an integralpart of the Internet of Things (IoT) and enable a largenumber of applications with high societal relevanceand impact. Miniature low-cost wireless sensors and

This paper is accepted at the International Workshop on VeryLarge Internet of Things (VLIoT 2017) in conjunction with theVLDB 2017 Conference in Munich, Germany. The proceedings ofVLIoT@VLDB 2017 are published in the Open Journal of Internetof Things (OJIOT) as special issue.

actuators are indeed increasingly being used, amongothers, to build smart cities and make life in dense urbanenvironments more comfortable, to control and optimizeproduction processes in smart factories, to monitor thevital functions of patients in hospitals, or to maximize thecomfort of inhabitants in residential buildings and officeswhile reducing their monthly energy bill.

Several of these IoT applications employ aconsiderable number of devices and can be deployedon a very large scale (e.g., across several districts of

103

http://creativecommons.org/licenses/by/4.0/

http://www.ronpub.com/ojiot

Open Journal of Internet of Things (OJIOT), Volume 1, Issue 1, 2017

a city [22], or across several wards in hospitals [13]).Despite the scale and the number of nodes, the networkis expected to operate reliably and efficiently forextended periods of time. On the one hand, the senseddata or any actuation command needs to be reliably andtimely delivered (e.g., alarms due to intrusion detectionor deteriorating vital signs of patients). On the otherhand, the energy consumption of the network needsto be minimized, as wireless sensors and actuators aretypically powered by batteries with limited capacity. Ahighly energy efficient network implies a longer systemlifetime and avoids a frequent battery replacement.

A major threat to the reliability and energy efficiencyof low-power wireless networks used in the IoT is radiointerference. Most of the commercial wireless IoTdevices use indeed the increasingly crowded and lightlyregulated ISM radio bands, freely-available portions ofthe radio spectrum reserved worldwide for industrial,scientific and medical purposes. The 2.4 GHz frequencyspectrum is a notorious example of a crowded ISMband: Wireless devices specifically marketed for the IoT,such as IEEE 802.15.4, Bluetooth low-energy (BLE) andWireless-HART, communicate using these frequenciesand have not only to co-exist with each other, but alsowith other wireless devices and home appliances thatcommunicate or emit noise in this frequency range [30].The latter includes IEEE 802.11 (Wi-Fi) devices andmicrowave ovens, which are nowadays ubiquitous inhouseholds and residential or public buildings.

Figure 1 shows an example of wireless technologiesemploying the same frequencies for communication.IEEE 802.11, IEEE 802.15.4 and BLE use overlappingchannels and their communications may henceexperience disturbances from surrounding devices.The presence of neighboring devices transmittingat higher power may lead to unpredictable mediumaccess contention times and high delays as well as to asignificant increase in the packet loss rate. In addition,interference from surrounding devices may significantlyworsen the energy efficiency of the system, as well asincrease network traffic due to packet re-transmissions.As more and more IoT devices are being deployednowadays, and as their number will grow exponentiallyin the coming years, it is to be expected that the sharedfrequency spectrum will become increasingly morecrowded and that interference from surrounding deviceswill represent a major threat for the dependability of IoTapplications deployed in large-scale installations.

An orthogonal problem to unintentional interferencefrom surrounding wireless devices are maliciousjamming attacks to IoT devices. The shared nature of thewireless medium makes indeed it easy for an adversary

to launch denial of service attacks on low-power wirelessdevices, and these attacks can be easily accomplishedalso by using off-the-shelf equipment [20]. The presenceof malicious jammers in the surroundings of a low-power wireless sensor or actuator can easily block thetransmission and reception of packets, as well as quicklydeplete a battery if no proper mechanisms are in place atthe medium access control (MAC) layer [19].

The problem of denial of service attacks is evenmore significant given that most IoT devices are leftunattended during their operation. For this reason, itis important to give low-power wireless sensors andactuators the ability to autonomously mitigate – whenpossible – the impact of surrounding interference.

In this paper we develop an approach, DynCCA [25],which dynamically adapts the clear channel assessment(CCA) threshold of low-power MAC protocolsemployed in common IoT applications. DynCCAuses only information computed locally to adjust theclear channel assessment threshold and can be used byall low-power MAC protocols based on carrier sensemultiple access with collision avoidance (CSMA/CA).We show experimentally that this mechanism cansignificantly help in minimizing the impact of maliciousor unintentional interference on both network reliabilityand energy efficiency. In particular, we demonstratethat varying the clear channel assessment threshold atrun-time allows to filter the (malicious) noise generatedby surrounding nodes, allowing a large-scale dense IoTnetwork to sustain a high packet reception ratio and highenergy efficiency even in the presence of interference.

The rest of this paper is organized as follows. Thenext section describes the body of work that has studiedthe role of clear channel assessment on the performanceof IoT networks. Section 3 describes in detail how theCCA threshold can be used to tune network densityand minimize the impact of malicious or unintentionalinterference on both network reliability and energyefficiency. We present the design of the DynCCAalgorithm in Section 4, along with a description of itsimplementation on top of ContikiMAC. In Section 5 weevaluate experimentally the performance of DynCCA ina network of 30 nodes and show that a large portionof the devices in the network can efficiently escapethe interference in their surroundings and sustain ahigh packet reception ratio. We finally summarize ourcontributions and conclude the paper in Section 6, alongwith a discussion of future work.

104

T. Sparber, C.A. Boano, S.S. Kanhere, and K. Romer: Mitigating Radio Interference in Large IoT Networks through Dynamic CCA Adjustment

BLEDataAdvertising

2 MHz11

23

12

13 14

15

16

17

18

19

20

21

22

24

25

26

IEEE 802.15.4

37

39

38

0 1 2 3 4 5 6 7 8 9 10

11

12

13

14

15

16

17

18

19 20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

1 2 3 4 5 6 7 8 9 10

11

12

13

14

22 MHz2412 2484

IEEE 802.11

Figure 1: Wireless technologies used to build IoT applications, such as IEEE 802.15.4 and BLE, have toco-exist with each other as well as with co-located Wi-Fi networks [26]

2 RELATED WORK

The influence of clear channel assessment on theperformance of low-power wireless networks hasattracted a large interest in the research community.

A large body of work has analyzed the impactof different clear channel assessment modes [29]and parameters such as back-off times [2] on theperformance of IEEE 802.15.4 networks, providinghelpful hints on how to optimize the use of the CCAalgorithm. Wong et al. [27] and Kim [16] haveanalyzed the benefits and drawbacks of multiple clearchannel assessments following the detection of a busychannel. Kiryushin et al. [18] have shown that low-power CSMA/CA-based MAC protocols suffer from ahigh number of collisions when transmitters can heareach other and start their transmissions almost at thesame time, and suggested to select a CCA threshold closeto the noise floor to reduce the number of such collisions.

Another body of work has proposed the useof algorithms that dynamically change the clearchannel assessment threshold to improve networkperformance. In the remainder of this section, we reviewthese algorithms, highlighting the key differences incomparison to our work.

Adaptive CCA to reduce channel access failures.King et al. [17] have shown that employing a back-offstrategy when colliding with traffic generated by nonIEEE 802.15.4 devices decreases network throughputand does not contribute to a collision resolution. Theirfindings confirm the experiments of Bertocco et al. [3]that have previously shown that the best performancein a crowded spectrum is obtained when disablingboth the channel sensing the back-off mechanism ofIEEE 802.15.4 devices. To alleviate the problem, King

et al. [17] propose to dynamically differentiate CCAto ignore non IEEE 802.15.4 traffic during and beforepacket transmission, and to immediately re-transmit apacket without a back-off in case no acknowledgementis received and non IEEE 802.15.4 traffic has beendetected. Similarly, Yuan et al. [28] have proposed analgorithm that dynamically adapts the CCA threshold toreduce the amount of discarded packets due to channelaccess failures, and validated it in simulation. In contrastto this body of work, our solution does not focus onchannel access failures during transmissions, but insteadon improving the efficiency of packet reception in thepresence of radio interference.

Adapting the CCA threshold to minimize the numberof false wake-ups. A few studies have experimentallyshown that false-wake ups caused by a sub-optimalCCA mechanism can significantly affect the energy-efficiency of low-power listening protocols, especiallyin noisy environments. King et al. [17] have proposedan enhancement of ContikiMAC – Contiki’s defaultMAC protocol – that lets a node keep its radio onto receive a packet only if IEEE 802.15.4 traffic hasbeen detected. The authors exploit the modulationdetection of carrier sense (i.e., the one reporting thechannel busy only if an IEEE 802.15.4 compliant signalis detected) and ignore any other activities, henceimmediately returning the radio to its sleeping state.Sha et al. [24] have designed AEDP, an adaptive energydetection protocol that dynamically adjusts a node’sclear channel assessment threshold to improve networkreliability and duty cycle based on application-specifiedbounds. In contrast to the work we describe in this paper,AEDP is a reactive approach that focuses on application-specific requirements (e.g., whether the current ETX ishigher than a given threshold) and does not carry outa pro-active enhancement of network performance as

105


soon as (malicious) radio interference is detected in thesurroundings.

Adapting the CCA threshold to temperaturevariations. Researchers have also analyzed theimpact of temperature variations on the performanceof low-power MAC protocols, highlighting how thefunctionality of clear channel assessment on traditionallow-power listening protocols drastically decreases athigh temperatures [4, 10, 23]. In particular, Bannisteret al. [1] and Boano et al. [5, 7, 11] have shown that thereceived signal strength attenuates at high temperaturesdue to the impact of temperature on the radio’s low-noise and power amplifiers, which can cause a completedisruption of a wireless link when static clear channelassessment thresholds are employed. To alleviate theproblem, the authors model the attenuation of thereceived signal strength on common low-power radiosas a function of temperature variations, and leveragethese models to dynamically adapt the CCA threshold ofContikiMAC. The adaptive algorithm described in thispaper is orthogonal to this body of work, and can becombined with the aforementioned adaptive algorithm tomaximize the reliability of IoT protocols in the presenceof both temperature variations and radio interference inthe surroundings.

3 ADJUSTING THE CCA THRESHOLDTO ESCAPE INTERFERENCE

The clear channel assessment algorithm plays afundamental role in low-power CSMA-CA MACprotocols with respect to reducing the number ofwasteful transmissions and preserving the limited energybudget of the nodes in the network. In particular, CCA istraditionally employed for two main tasks:

1. Collision avoidance during transmission.Low-power CSMA-CA MAC protocols relyon clear channel assessment to determinewhether another device is already transmittingon the same frequency channel, and defertransmissions that may otherwise collide withongoing communications. In case no ongoingtransmissions are detected, a packet is immediatelysent, otherwise the MAC protocol defers thetransmission using different back-off strategies [9].

2. Wake-up of nodes. Duty-cycled protocols such asContikiMAC [14], B-MAC [21], and X-MAC [12]typically employ CCA to trigger wake-ups, i.e., todetermine if a node should stay awake to receivea packet or whether it should remain in low-powermode. Towards this goal, an inexpensive CCAcheck is performed: If the channel is detected to be

busy, the transceiver is kept on in order to receivethe incoming packet, otherwise the radio returns tosleep mode.

The CCA check is traditionally carried out usingenergy detection or carrier sense. The latter consists insampling the energy level in the wireless channel andcomparing the measured signal strength with a givenCCA threshold as shown in Figure 2. Most protocolsemploy fixed CCA thresholds and rely on the defaultsystem settings. This typically implies that the defaultvalue of the radio device is used, e.g., −77 dBm for thewidely used TI CC2420 transceiver [7].

3.1 Varying the CCA threshold

Changing the default CCA threshold can have a strongimpact on the performance of duty-cycled CSMA-CA MAC protocols, especially in the presence ofinterference in the surroundings.

Impact on energy consumption. On the one hand,lowering the CCA threshold, i.e., picking a value closerto the sensitivity threshold of the transceiver, may causethe radio to remain active for a large portion of time andhence reduce the energy efficiency of the system. Havinga low CCA threshold maximizes indeed the chancesto hear RF noise generated by surrounding devices,which increases the probability of backing-off duringtransmission as well as the number of false wake-upsduring reception (see Figure 2(b)). This applies to RFnoise generated by transmissions from other nodes inthe same network (internal interference) and to RF noisegenerated by neighboring devices that do not belong tothe same network (external interference) [6].

Impact on network density. On the other hand,increasing the CCA threshold allows to minimize theenergy expenditure, but maximizes the risk of having adisconnected network. The CCA threshold has indeeda high impact on network density: If a node Kreceives packets from a neighbor N with a receivedsignal strength RS that is lower than the selectedCCA threshold CT , its radio is never woken up fromlow-power mode and no link can be established (seeFigure 2(a)). By decreasing CT to a value below RS ,Node K can establish a connection with N , but mayincrease the number of false wake-ups, as previouslydiscussed. Consequently, increasing the CCA thresholdhelps in minimizing the number of false wake-ups, butmay also cause the number of connected links in thenetwork to drastically decrease.

Figure 3 shows the impact of different CCAthreshold values on connectivity in a network of29 Advanticsys MTM-CM5000-MSP nodes (TelosB

106


-100

-80

-60

-40

-20

0 0.5 1 1.5 2

Transmitter: start transmissionReceiver: keep radio off

RS

SI [

dBm

]

Time [ms]

Measured RSSICCA Threshold

(a) Received signal below CCA threshold

-100

-80

-60

-40

-20

0 0.5 1 1.5 2

Transmitter: defer transmissionReceiver: keep radio on

RS

SI [

dBm

]

Time [ms]

Measured RSSICCA Threshold

(b) Received signal above CCA threshold

Figure 2: The CCA algorithm is traditionally used in low-power CSMA-CA MAC protocols to performcollision avoidance during transmission and to wake-up nodes from their sleep state

10014

10213

-71/-73 dBm46/52%

10314

-78/-79 dBm43/11%

10620

-78/-77 dBm58/63%

10715

-78/-81 dBm16/5%

1089

-100/-78 dBm0/90%

10912

-100/-76 dBm0/89%

1149

-76/-78 dBm68/24% 115

10

-67/-70 dBm79/93%

11722

-69/-77 dBm69/88%

11814

-77/-82 dBm55/6%

12011

-72/-74 dBm88/89%

12311

-75/-78 dBm81/1%

12516

-78/-78 dBm54/91%

12710

-62/-61 dBm52/90%

-71/-69 dBm53/38%

-100/-71 dBm0/40%

-70/-71 dBm81/50%

-76/-77 dBm64/38%

-75/-73 dBm42/44%

10413

-76/-75 dBm30/22%

12916

-60/-61 dBm70/54%130

13

-76/-74 dBm51/38%

1317

-78/-76 dBm19/41%

11112

-78/-77 dBm12/40%

11215

-77/-75 dBm55/51%

11612

-76/-75 dBm49/49%

-78/-78 dBm1/40%

-100/-74 dBm0/61%

-74/-76 dBm71/62%-61/-62 dBm

92/62%

-72/-70 dBm47/64%

-69/-70 dBm29/47%

11313

-68/-69 dBm57/62%

-76/-80 dBm67/5%

-67/-67 dBm52/62%

-69/-70 dBm56/61%

-67/-67 dBm57/61%

11011

-73/-76 dBm83/62%

-68/-71 dBm60/52%

-100/-68 dBm0/35%

-74/-77 dBm74/59%

-73/-76 dBm58/62%

-72/-80 dBm73/2%

-68/-73 dBm60/62%

-78/-80 dBm26/1%

-60/-63 dBm72/61%

-76/-75 dBm72/63%

-74/-76 dBm63/58%

-73/-74 dBm57/60%

-71/-72 dBm58/63%

-76/-76 dBm48/63%

-70/-75 dBm77/63%

1268

-76/-78 dBm86/6%

1228

-65/-66 dBm70/61%

1242

-77/-79 dBm70/4%

-100/-70 dBm0/55%

-100/-73 dBm0/50%

-73/-79 dBm74/7%

-70/-72 dBm58/67%

-62/-63 dBm73/65%

-73/-69 dBm69/66%

-76/-78 dBm66/63%

-80/-78 dBm4/59%

-70/-68 dBm59/67%

-80/-76 dBm1/66%

-72/-73 dBm68/55%

-66/-64 dBm72/66% -77/-100 dBm

69/0%

-76/-100 dBm79/0%

-68/-100 dBm69/0%

-71/-100 dBm63/0%

-76/-100 dBm51/0%

-75/-100 dBm90/0%

-66/-100 dBm73/0%

-76/-100 dBm76/0%

-77/-100 dBm49/0%

-76/-100 dBm69/0%

-70/-100 dBm53/0%

-73/-100 dBm53/0%

-74/-100 dBm77/0%

-73/-73 dBm89/73%

-64/-65 dBm67/81%

-70/-68 dBm51/78%

-73/-70 dBm57/80%

-76/-82 dBm70/3%

-77/-76 dBm91/66%

-80/-76 dBm2/75%-82/-78 dBm

3/26%

-76/-74 dBm47/49%

-71/-67 dBm52/70%

-75/-64 dBm73/75%

-79/-71 dBm3/73%

-76/-72 dBm70/71%

-78/-70 dBm34/72%

-70/-64 dBm85/70%

-77/-70 dBm64/74%

1192

-100/-70 dBm0/11%

-74/-67 dBm73/58%

-80/-75 dBm1/56%

-78/-74 dBm10/55%

-100/-76 dBm0/56%

-57/-60 dBm69/86%

-81/-78 dBm6/23%

-83/-78 dBm4/34%

-76/-81 dBm64/2%

-76/-78 dBm49/36%

-65/-69 dBm88/53%

-78/-82 dBm37/2%

-71/-73 dBm52/20%

1019

-100/-74 dBm0/63%

-78/-77 dBm45/58%

-76/-81 dBm69/6%

-72/-68 dBm52/65%

-80/-77 dBm2/62%

-81/-79 dBm7/11%

-67/-68 dBm67/72%

-100/-76 dBm0/77%

-78/-74 dBm59/81%

-79/-80 dBm11/2%

-73/-76 dBm74/30%

-70/-78 dBm73/29%

-69/-70 dBm89/30%

-78/-77 dBm21/31%

-69/-72 dBm68/28%

-78/-100 dBm17/0%

-76/-76 dBm52/30%

-78/-100 dBm22/0%

-75/-76 dBm85/29%

-71/-78 dBm73/47%

-78/-82 dBm55/5%

-69/-69 dBm91/66%

-77/-80 dBm76/2%

-80/-78 dBm10/49%

-76/-78 dBm70/12%

-77/-75 dBm47/62%

-73/-74 dBm87/61%

-70/-70 dBm62/71%

-59/-59 dBm48/53%-67/-62 dBm

65/51%

-51/-50 dBm63/48%

-65/-73 dBm74/71%

-76/-80 dBm59/2%

-71/-74 dBm77/52%

-77/-75 dBm75/40%

-78/-78 dBm59/27%

-72/-72 dBm61/73%

-62/-68 dBm72/51%

-66/-71 dBm56/60%

-76/-77 dBm90/28%

-70/-68 dBm65/61%

-77/-80 dBm65/4%

-78/-78 dBm18/15%

-76/-76 dBm81/55%

-78/-88 dBm19/7%

-72/-77 dBm60/19%

-77/-80 dBm89/3%

-75/-75 dBm73/46%

-74/-75 dBm49/49%

-66/-68 dBm86/48%

-75/-80 dBm75/6%

-75/-76 dBm58/70%

-77/-77 dBm77/76%

-82/-77 dBm5/72%

-79/-77 dBm3/60%

1059

-71/-68 dBm59/7%

-68/-68 dBm61/5%

-74/-79 dBm74/1%

-74/-76 dBm59/7%

-68/-68 dBm76/7%

-76/-71 dBm72/7%

-75/-74 dBm58/7%

-78/-78 dBm40/5%

-64/-62 dBm65/7%

-68/-71 dBm81/65%

-80/-78 dBm1/25%

1210

(a) CCA threshold: −80dBm

1005

1025

-71/-73 dBm46/52%

1151

-67/-70 dBm79/93%

11715

-69/-77 dBm69/88%120

5

-72/-74 dBm88/89%

1275

-62/-61 dBm52/90%

1299

-60/-61 dBm70/54%

1038

-71/-69 dBm53/38%

1093

-100/-71 dBm0/40%

1144

-70/-71 dBm81/50%

-79/-71 dBm3/73%

-76/-72 dBm70/71%

1306

-78/-70 dBm34/72%

1227

-77/-70 dBm64/74%

1256

-75/-64 dBm73/75%

1186

-71/-67 dBm52/70%

1263-70/-64 dBm

85/70%

1191

-100/-70 dBm0/11%

-57/-60 dBm69/86%

-71/-73 dBm52/20%

1012

-72/-68 dBm52/65%

-67/-68 dBm67/72%

-59/-59 dBm48/53%

1313

-67/-62 dBm65/51%

-61/-62 dBm92/62%

-72/-70 dBm47/64%

1044

-69/-70 dBm29/47%

1115

-69/-70 dBm56/61%

1127

-67/-67 dBm57/61% 113

4

-68/-69 dBm57/62%

-67/-67 dBm52/62%

-66/-100 dBm73/0%

-70/-100 dBm53/0%

-64/-65 dBm67/81%

-70/-68 dBm51/78%

-73/-70 dBm57/80%

-70/-78 dBm73/29%

-69/-70 dBm89/30%

-69/-72 dBm68/28%

-65/-73 dBm74/71%

-72/-72 dBm61/73%

1235 -71/-74 dBm

77/52%

-62/-68 dBm72/51%

-70/-68 dBm65/61%

-66/-71 dBm56/60%

-71/-78 dBm73/47%

-69/-69 dBm91/66%

-70/-70 dBm62/71%

-51/-50 dBm63/48%

1055

1069

-71/-68 dBm59/7%

1079

-68/-68 dBm61/5%

-64/-62 dBm65/7%

-68/-68 dBm76/7%

-76/-71 dBm72/7% -72/-80 dBm

73/2%

-71/-72 dBm58/63%

-68/-71 dBm60/52%

-65/-66 dBm70/61%

-60/-63 dBm72/61%

1084

-100/-68 dBm0/35% 110

2-70/-75 dBm

77/63%

-68/-73 dBm60/62%

-70/-68 dBm59/67%

-66/-64 dBm72/66%

-62/-63 dBm73/65%

-73/-69 dBm69/66%

-100/-70 dBm0/55%

-72/-73 dBm68/55%

-70/-72 dBm58/67%

-68/-71 dBm81/65%

-65/-69 dBm88/53%

-68/-100 dBm69/0%

-71/-100 dBm63/0%

-74/-67 dBm73/58%

1162

-72/-77 dBm60/19%

-66/-68 dBm86/48%

1210

1240

(b) CCA threshold: −73dBm

1001

1271

-62/-61 dBm52/90%

1021

1295

-60/-61 dBm70/54%130

2 -59/-59 dBm48/53%

1312

-67/-62 dBm65/51%

1031

1202

-61/-62 dBm92/62%

-57/-60 dBm69/86%

1051

1222

-64/-62 dBm65/7%

1061

1232 -60/-63 dBm

72/61%

1072

-66/-64 dBm72/66%

-62/-63 dBm73/65%

1121

1173

-62/-68 dBm72/51%

1251

-75/-64 dBm73/75%

1261-70/-64 dBm

85/70%

1141

-64/-65 dBm67/81%

-51/-50 dBm63/48%

1010

1040

1080

1090

1100

1110

1130

1150

1160

1180

1190

1210

1240

(c) CCA threshold: −65dBm

Figure 3: Varying the clear channel assessment threshold has a direct effect on the density of the network

replicas) deployed in an office building and transmittingpackets periodically with an output power of −22 dBm.When selecting a low CCA threshold such as −80 dBm,the network is fully connected, with a total number oflinks between all nodes equals to 175 (Figure 3(a)).When increasing the clear channel assessment thresholdof all nodes to −73 dBm, most of the nodes in thenetwork are still connected (with the exception of node124), but the total number of links is less than halfcompared to the previous case (Figure 3(b)). If theCCA threshold is increased further (e.g., to −65 dBm),most of the nodes in the network are isolated and cannotconnect to any neighbor (Figure 3(c)).

3.2 Impact of Malicious Interferers onCCA Operation

A notable case is the one in which the selected CCAthreshold is equal to or lower than the measured noisefloor, i.e., the RSSI in absence of packet transmissions.In this case the medium is detected to be busy essentiallyat all times, postponing all transmissions and keeping theradio unnecessarily active to listen for incoming packetsand causing a very quick battery depletion. This is forexample the case when a malicious jammer is constantlyactive in proximity of a node.

This worst-case scenario is depicted in Figure 4(a),which shows the RSSI measured by an Advanticsys

107


-100

-80

-60

-40

-20

0

0 3000 6000 9000 12000

Malicious Jamming Sequence Wi-Fi transmission

RS

SI [

dBm

]

Time [µs]

(a) Measured RSSI with different interference sources

0 3 6 9 12 15Time [min]

100

90

80

70

60

50

40

RSSI

[dBm

]

103123

(b) Measured RSSI in the presence of a malicious jammer (node 121)

Figure 4: Comparison of the different RSSI measurements obtained in the presence of a malicious jammerand a Wi-Fi station active nearby (a) and RSSI measurements on all nodes in the network in the presence ofa malicious jammer placed in their proximity (b)

MTM-CM5000-MSP sensor node in the presence ofdifferent types of interference. Initially, the noise flooris close to the sensitivity threshold of the radio (about−95 dBm, since the TI CC2420 is used). As soonas a malicious jammer is active nearby, the RSSI ispersistently increased: If the received signal is abovethe CCA threshold, we have the aforementioned case inwhich the medium is detected to be busy essentially atall times. In the presence of Wi-Fi interference, instead,the RSSI is not constantly above the CCA threshold, butonly for a fraction of time that is dependent on the typeand frequency of Wi-Fi transmissions. It is importantto highlight that smart malicious jammers could alsoemulate Wi-Fi transmissions (e.g., using JamLab [8]),and that it is hence important to be resilient both to ajammer constantly active and to intermittent and burstyinterference.

The impact of interference is not the same across anetwork, as IoT networks can be very large in scale.Interference often affects indeed only a portion of thenodes in the network. To observe the spatial impact of amalicious interferer, we reuse the same network used toperform the density experiments shown in Figure 3 to runa data collection using Contiki’s RPL, where all nodesforward data to a central sink (node 108, highlightedin black in Figure 3), and where all nodes periodicallyread their noise floor. One of the nodes in the network(node 121, highlighted in red in Figure 3), acts as amalicious jammer and emits constant noise by means ofa continuous carrier [8]. Figure 4(b) shows the increaseof noise floor at each of the nodes in the network afterthe malicious jammer is activated (i.e., after 5 minutes).Node 103 is closer to the jammer and measures an RSSIclose to −60 dBm, whereas node 123 is rather far awayand measures an RSSI below −77 dBm. For example, if

the CCA threshold would be selected to be −77 dBm,Node 103 would be persistently blocked, whereas theoperations of Node 123 would not be affected.

Based on the aforementioned observations, to mitigatethe impact of interference it is necessary to (i) performan accurate measurement of the noise floor on all nodesin the network, and to (ii) adapt the CCA threshold ofeach individual node such that most of the interferenceis avoided, while maintaining connectivity with the restof the network. The next section presents a lightweightalgorithm that dynamically changes the CCA thresholdof a node based on the measured noise floor.

4 DYNCCA: DESIGN AND IMPLEMENTATION

This section describes the design and implementationof DynCCA, an algorithm that dynamically adapts theclear channel assessment threshold of 802.15.4 radiosto minimize the impact of malicious or unintentionalinterference on both network reliability and energyefficiency.

Requirements. As shown in Section 3, setting theCCA threshold just above the noise floor can helpin escaping interference. This requires the ability toperform a periodic measurement of the noise floor.Such measurement should give an accurate pictureof interference in the surroundings, but minimize theamount of time during which the radio is activeto maximize energy-efficiency. As interference canoccur in different forms, the algorithm to be designedshould be effective against both malicious jammingand unintentional background interference. Finally, thealgorithm to be developed should also be transparentto the application, i.e., the adaption of the clear

108


Algorithm 1: DynCCA’s dynamic threshold adaptation

1: procedure DYNCCA2: xt ← find noise floor(t) + ε3: xt ← max (xt, CCAfix)4: CCAt ← min (xt, xt−1, xt−2, . . . , xt−n−1) + β5: end procedure

channel assessment should have minimal impact on theapplication running on the nodes.

Obtaining a good noise floor estimate. Due to theRSSI readings of low-cost IoT radios being noisy, ahigh number of RSSI samples at high frequency istraditionally required to get a good estimate of thesurrounding interference. To reduce the number ofmeasurements as well as the memory used, but stillobtain a good picture of the surrounding interference, wesample RSSI values and build a histogram of the RSSIoccurrences. This allows us to easily estimate the noisefloor by identifying the highest observed RSSI level, theRSSI value occurring most often, or the minimum RSSIvalue recorded by at least a given portion of the readings(percentile).

DynCCA algorithm. DynCCA builds on top of theaforementioned RSSI estimation and is sketched inAlgorithm 1. After deriving the noise floor xt at time tfrom the RSSI measurements, a constant value ε is addedto it in order to account for the co-channel rejectionability of low-power radios1. The chosen noise floor xtis then capped at to a fixed threshold CCAfix. This isan optional step, but important to reduce the number offalse wake-ups in the networks and especially to allowan optimal tree formation in data collection protocols.For example, in the IPv6 Routing Protocol for Low-Power and Lossy Networks (RPL), the default objectivefunction ETX tries to minimize the number of hopsto the sink and hence prefers nodes sustaining a highpacket reception ratio that can cover large distances.When selecting very low CCA thresholds, there is anincreased chance to create links that are easily affectedby radio interference or that can fall in the transitionalregion [31], causing a large number of parent switchesand reducing network performance. Hence, by cappingxt at CCAfix, we make sure to select nodes with anRSSI sufficiently higher than the transitional region.Please note that the selection of CCAfix is especiallycritical in sparse networks, as high values may lead to apartitioned network (see Section 3.1).

After this step, a filtered baseline CCAt is calculatedand used as a lower bound for the CCA threshold.

1 Low-power radios can receive a valid packet only if it is higher thanthe noise floor by a factor specified as co-channel rejection ratio.

Filtering the measured noise floor is important to keepthe network stable. As only the lower bound is ofinterest, the filter uses the minimum value of the lastn samples. Adding a constant value β to CCAt canhelp data collection protocols in selecting better parentsby forcing a reduced set of neighbors (β > 0) or bylowering the CCA threshold to ensure that the networkis connected (β < 0). Obtaining a sufficiently accurateknowledge of the current network performance and thenumber of neighbors to properly select β may, however,come at a higher communication overhead or energyexpenditure, and we therefore keep this feature optional.

Implementation. We implement DynCCA onthe popular Contiki operating system, and keep itsimplementation lightweight and energy efficient, asrequired to support constrained networked embeddedsystems. We implement DynCCA as a separate Contikiprocess, running every 10 s. The current implementationis optimized for Contiki OS’ sky platform and measuresthe current noise floor for approximately 50ms at asampling frequency of about 20 kHz. It then determinesa filtered value by computing the minimum value of thelast four measurements. We further optimize the RSSIreadings by implementing access to the SPI betweenthe micro-controller and the CC2420 radio in assembler.The array employed to store the RSSI histogram uses 2bytes per index and stores values in the range [-100,0]dBm.

5 EVALUATION

We evaluate the performance of DynCCAexperimentally and show that it helps to significantlyimprove both network reliability and energy efficiency.

5.1 Experimental Setup

We run a set of experiments on 30 Advanticsys MTM-CM5000-MSP nodes deployed in our local testbed. Allnodes run a data collection application using RPL withETX as objective function and form a mesh network asshown in Figure 3. Each node periodically sends a UDPmessage with a payload of 46 bytes to the sink (markedin black in Figure 3). Transmissions are scheduled every10 seconds, with a random offset of ±10 seconds. Thetransmission power of the nodes has been set to 4 toensure multiple hops in our dense testbed setup and itis assumed that nodes cannot increase their transmissionpower to escape interference.

We use Contiki’s default MAC protocol,ContikiMAC [14], with a channel check rate (CCR) of32 Hz to better factor out packet losses due to internalinterference. Using a lower CCR would decrease the

109


Table 1: Summary of our evaluation resultsIn

terf

eren

cety

pe

Dyn

CC

Aac

tive?

PRR

[%]

Pow

erco

nsum

ptio

n[m

W]

No.

ofpa

rent

chan

ges/

node

Nod

esw

ithPR

R>

90%

None NO 94 3.45 4.00 27None YES 97 3.29 3.83 28Wi-Fi NO 86 8.16 13.24 18Wi-Fi YES 89 3.53 4.24 25

Malicious NO 9 14.52 0.52 2Malicious YES 61 4.47 10.97 15

energy consumption, but also increase the packet lossand latency. Similar to the setup used in Section 3.2,we run JamLab [8] on node 121 (highlighted in red inFigure 3) at transmission power 11 to emulate either aWi-Fi device or a malicious interferer. We collect thepacket reception rate (PRR) between each node andmeasure the energy consumption of all the nodes inthe network using Energest [15]. We further collecta number of low-level metrics such as the number ofparent changes per node and the RSSI of each packet.All experiments last one hour and are repeated multipletimes.

5.2 Results

We compare the performance of data collection usingRPL in our network with Contiki’s default CCAthreshold and with DynCCA by performing experiments(i) in absence of controlled interference, (ii) in thepresence of emulated Wi-Fi interference, and (iii) in thepresence of malicious interference. Table 1 summarizesour results.

Power consumption. The use of DynCCAsignificantly helps in reducing the false wake-up rate inthe presence of Wi-Fi interference. We have measureda decrease in the average power consumption of thenodes in the network from 8.16mW to 3.53mW whenusing DynCCA, i.e., an improvement of 56%. DynCCAachieves even better results in the presence of maliciousinterference: The average power consumption in thenetwork is reduced from 14.52mW to 4.47mW, i.e., adecrease of 69%. By comparing the power consumptionrecorded in absence of interference (3.29mW), we can

0 1 2 3 4 5 6Epsilon [dB]

012345678

Powe

r [m

W]

(a) Power [mW]

0 1 2 3 4 5 6Epsilon [dB]

0102030405060708090

PRR

[%]

(b) PRR [%]

Figure 5: Performance of DynCCA as a function of ε

100

500

800

1000

2000

5000

1000

0Samples

0

1

2

3

4

5

6

7

Powe

r [m

W]

(a) Impact of number of samples

83 85 88 91 94 97 100

Percentile [th]

012345678

Powe

r [m

W]

(b) Impact of noise floor choice

Figure 6: Power consumption of DynCCA as afunction of the number of RSSI samples and noisefloor choice

also conclude that DynCCA is efficient and does notnegatively affect the overall energy consumption.

Packet reception rate. As discussed in Section 4,the use of DynCCA helps RPL in forming an optimaltree by avoiding unreliable links close to or inside thetransitional region. This is shown by the improvement inthe packet reception rate of 3% in absence of interferenceas well as in the presence of Wi-Fi interference. In caseof malicious interference, Contiki’s default performanceis very poor, with an average PRR in the network below10% and with only two nodes being able to sustain aPRR higher than 90%. When using DynCCA, the PRRin the network is increased to 61%, with 15 nodes beingable to sustain a PRR higher than 90%, i.e., DynCCAcould allow 13 nodes to escape the malicious jammer byautomatically adapting their CCA threshold.

Parent changes. Our experimental results have alsoshown that, in the presence of Wi-Fi interference,nodes experience a significantly lower number of parentchanges per node (from 13.24 down to 4.24), reachinga value very close to the one observed when no

110


interference is present (3.8). This further confirmsthat the use of DynCCA helps RPL in forming anoptimal tree. Please note that in the case of maliciousinterference, the number of parent changes per node isincreased from 0.51 to 10.96 due to the fact that, withoutDynCCA, the nodes’ communication was blocked andthus no parent change could be performed.

Impact of specific parameters. We finally evaluatethe performance of DynCCA as a function of specificparameters. First, we run experiments comparing thepacket reception rate and power consumption in thenetwork while changing ε in the range [0, 6]. Figure 5shows our experimental results. When using ε = 0,no packet is being received in the network, whereasselecting ε = 6 leads to a less connected network andhigher loss. A value of 3 dB represents the best trade-off,as it minimizes power consumption and maximizes thepacket reception rate. This is perhaps not surprising, as3 dB is exactly the declared co-channel rejection ratio ofthe employed radio transceiver – the TI CC2420. Pleasenote that the experiments summarized in Table 1 wereconducted with the optimal value of ε = 3dB.

We also analyze whether the number of samples orthe noise floor percentile used have an influence on theefficiency of the DynCCA algorithm. Our experimentalresults summarized in Figure 6 show that a highernumber of samples increases the accuracy but alsothe overhead: Using a noise floor percentile higherthan the 88th and a number of RSSI samples between500 and 2000 provides the best trade-off in terms ofpower consumption. Please note that the experimentssummarized in Table 1 were conducted with 1000 RSSIsamples and using the 100th percentile.

6 CONCLUSIONS

The CCA threshold used by radio-duty-cycled protocolssuch as ContikiMAC is found to be an adjustableknob to improve network performance under heavyinterference. In this paper, we propose and implementDynCCA: An approach to dynamically change the CCAthreshold in order to mitigate both unintentional andmalicious interference in the surroundings of a node. Anexperimental evaluation shows that the use of DynCCAcan increase the packet reception rate in a networkfrom 9% to about 60%, while also reducing the energyconsumption by 69%. DynCCA is particularly usefulfor large IoT installations where the deployed nodes areunattended and vulnerable to denial of service attacksand interference from surrounding devices such as Wi-Fi access points.

In the future, we plan to carry out experimentsby generating interference using real Wi-Fi devices

instead of JamLab. Future work also includes theimplementation of β: This would integrate data collectedfrom RPL such as the number of available neighborsinto the CCA threshold adaption algorithm to enforce asmaller set of (better) parents.

ACKNOWLEDGEMENTS

This work was performed within the LEAD-Project “Dependable Internet of Things in AdverseEnvironments”, funded by Graz University ofTechnology, Austria.

REFERENCES

[1] K. Bannister, G. Giorgetti, and S. K.Gupta, “Wireless Sensor Networking for HotApplications: Effects of Temperature on SignalStrength, Data Collection and Localization,” inProceedings of the 5th International Workshop onEmbedded Networked Sensors (HotEmNets), Jun.2008.

[2] M. Bertocco, G. Gamba, and A. Sona,“Experimental Optimization of CCA Thresholdsin Wireless Sensor Networks in the Presence ofInterference,” in Proceedings of the IEEE Europethe Workshop on ElectroMagnetic Compatibility(EMC), Jun. 2007.

[3] M. Bertocco, G. Gamba, and A. Sona, “IsCSMA/CA really Efficient against Interferencein a Wireless Control System? An ExperimentalAnswer,” in Proceedings of the 13th IEEEInternational Conference on EmergingTechnologies and Factory Automation (ETFA),Sep. 2008.

[4] C. A. Boano, “Dependable Wireless SensorNetworks,” Ph.D. dissertation, Graz University ofTechnology, Graz, Austria, Nov. 2014.

[5] C. A. Boano, J. Brown, N. Tsiftes, U. Roedig,and T. Voigt, “The Impact of Temperature onOutdoor Industrial Sensornet Applications,” IEEETransactions on Industrial Informatics, vol. 6,no. 3, Aug. 2010.

[6] C. A. Boano and K. Romer, “External RadioInterference,” in Radio Link Quality Estimation inLow-Power Wireless Networks, ser. SpringerBriefsin Electrical and Computer Engineering -Cooperating Objects. Springer InternationalPublishing, Jul. 2013.

[7] C. A. Boano, K. Romer, and N. Tsiftes, “Mitigatingthe Adverse Effects of Temperature on Low-PowerWireless Protocols,” in Proceedings of the 11th

111


IEEE International Conference on Mobile Ad hocand Sensor Systems (MASS), Oct. 2014.

[8] C. A. Boano, T. Voigt, C. Noda, K. Romer,and M. A. Zuniga, “JamLab: AugmentingSensornet Testbeds with Realistic and ControlledInterference Generation,” in Proceedings ofthe 10th IEEE International Conference onInformation Processing in Sensor Networks(IPSN), Apr. 2011.

[9] C. A. Boano, T. Voigt, N. Tsiftes, L. Mottola,K. Romer, and M. A. Zuniga, “Making SensornetMAC Protocols Robust Against Interference,” inProceedings of the 7th European Conference onWireless Sensor Networks (EWSN), Feb. 2010.

[10] C. A. Boano, H. Wennerstrom, M. A. Zuniga,J. Brown, C. Keppitiyagama, F. J. Oppermann,U. Roedig, L.-A. Norden, T. Voigt, and K. Romer,“Hot Packets: A Systematic Evaluation of theEffect of Temperature on Low Power WirelessTransceivers,” in Proceedings of the 5th ExtremeConference on Communication (ExtremeCom),Aug. 2013.

[11] C. A. Boano, M. A. Zuniga, J. Brown, U. Roedig,C. Keppitiyagama, and K. Romer, “TempLab:A Testbed Infrastructure to Study the Impact ofTemperature on Wireless Sensor Networks,” inProceedings of the 13th ACM/IEEE InternationalConference on Information Processing in SensorNetworks (IPSN), Apr. 2014.

[12] M. Buettner, G. V. Yee, E. Anderson, and R. Han,“X-MAC: A Short Preamble MAC Protocolfor Duty-cycled Wireless Sensor Networks,”in Proceedings of the 4th ACM InternationalConference on Embedded Networked SensorSystems (SenSys), Nov. 2006.

[13] O. Chipara, C. Lu, T. C. Bailey, and R. Gruia-Catalin, “Reliable Clinical Monitoring usingWireless Sensor Networks: Experiences in a Step-down Hospital Unit,” in Proceedings of the 8th

International Conference on Embedded NetworkedSensor Systems (SenSys), Nov. 2010.

[14] A. Dunkels, “The ContikiMAC Radio DutyCycling Protocol,” Swedish Institute of ComputerScience, Kista, Sweden, Tech. Rep. T2011:13, Dec.2011.

[15] A. Dunkels, F. Osterlind, N. Tsiftes, and Z. He,“Software-based On-line Energy Estimation forSensor Nodes,” in Proceedings of the 4th

International Workshop on Embedded NetworkedSensors (EmNetS), Jun. 2007.

[16] W. Kim, “Short Clear Channel Assessmentin Slotted IEEE 802.15.4 Networks,” WirelessPersonal Communications, vol. 71, no. 1, Sep.2012.

[17] A. King, J. Brown, and U. Roedig, “DCCA:Differentiating Clear Channel Assessment forImproved 802.11/802.15.4 Coexistence,” inProceedings of the 10th IEEE InternationalConference on Wireless and Mobile Computing,Networking and Communications (WiMob), Oct.2014.

[18] A. Kiryushin, A. Sadkov, and A. Mainwaring,“Real-World Performance of Clear ChannelAssessment in 802.15.4 Wireless SensorNetworks,” in Proceedings of the 2nd InternationalConference on Sensor Technologies andApplications (SENSORCOMM), Aug. 2008.

[19] Y. W. Law, M. Palaniswami, L. V. Hoesel,J. Doumen, P. Hartel, and P. Havinga, “Energy-efficient Link-layer Jamming Attacks AgainstWireless Sensor Network MAC Protocols,” ACMTransactions on Sensor Networks, vol. 5, no. 1,Feb. 2009.

[20] K. Pelechrinis, M. Iliofotou, and S. V.Krishnamurthy, “Denial of Service Attacks inWireless Networks: The Case of Jammers,” IEEECommunications Surveys Tutorials, vol. 13, no. 2,May 2010.

[21] J. Polastre, J. L. Hill, and D. E. Culler, “VersatileLow Power Media Access for Wireless SensorNetworks,” in Proceedings of the 2nd ACMInternational Conference on Embedded NetworkedSensor Systems (SenSys), Nov. 2004.

[22] San Francisco Municipal Transportation Agency,“SFpark: Putting Theory Into Practice,” Aug. 2011.

[23] F. Schmidt, M. Ceriotti, N. Hauser, and K. Wehrle,“If You Can’t Take the Heat: Temperature Effectson Low-Power Wireless Networks and How toMitigate Them,” in Proceedings of the 12th

European Conference on Wireless Sensor Networks(EWSN), Feb. 2015.

[24] M. Sha, G. Hackmann, and C. Lu, “Energy-efficient Low Power Listening for Wireless SensorNetworks in Noisy Environments,” in Proceedingsof the 12th ACM/IEEE International Conferenceon Information Processing in Sensor Networks(IPSN), Apr. 2013.

[25] T. Sparber, “Experimental Analysis and Evaluationof RPL under Radio Interference,” Master’s thesis,Graz University of Technology, Graz, Austria, Jan.2017.

112


[26] U. Wetzker, I. Splitt, M. Zimmerling, C. A.Boano, and K. Romer, “Troubleshooting WirelessCoexistence Problems in the Industrial Internet ofThings,” in Proceedings of the 14th IEEE/IFIPInternational Conference on Embedded andUbiquitous Computing (EUC), Aug. 2016.

[27] C.-M. Wong and W.-P. Hsu, “An Additional ClearChannel Assessment for IEEE 802.15.4 SlottedCSMA/CA Networks,” in Proceedings of theIEEE International Conference on CommunicationSystems (ICCS), Nov. 2010.

[28] W. Yuan, J.-P. M. Linnartz, and I. G. Niemegeers,“Adaptive CCA for IEEE 802.15.4 WirelessSensor Networks to Mitigate Interference,” inProceedings of the IEEE Wireless Communicationand Networking Conference (WCNC), Apr. 2010.

[29] M. Zeghdoud, P. Cordier, and M. Terre, “Impactof Clear Channel Assessment Mode on thePerformance of ZigBee Operating in a Wi-FiEnvironment,” in Proceedings of the 1st IEEEWorkshop on Operator-Assisted Wireless-MeshCommunity Networks (OpComm), Sep. 2006.

[30] G. Zhou, J. A. Stankovic, and S. H. Son,“Crowded Spectrum in Wireless Sensor Networks,”in Proceedings of the 3rd Workshop on EmbeddedNetworked Sensors (EmNets), May 2006.

[31] M. A. Zuniga and B. Krishnamachari, “Analyzingthe Transitional Region in Low-Power WirelessLinks,” in Proceedings of the 1st IEEEInternational Conference on Sensor, Mesh and AdHoc Communications and Networks (SECON),Oct. 2004.

AUTHOR BIOGRAPHIES

Tommy Sparber is anembedded software engineerat MEDS in Graz, Austria.He received his Bachelor andMaster degree in Informationand Computer Engineering fromGraz University of Technologyin 2013 and 2017, respectively.His Master’s thesis analyzedexperimentally the performanceof RPL under radio interference.

His current area of work spans from developingembedded Linux applications to real-time applicationson small micro-controllers.

Carlo Alberto Boano isan assistant professor atthe Institute for TechnicalInformatics of Graz Universityof Technology, Austria. Hereceived a doctoral degreesub-auspiciis praesidentisfrom Graz University ofTechnology in 2014 with athesis on dependable wirelesssensor networks, and holds

a double Master degree from Politecnico di Torino,Italy, and KTH Stockholm, Sweden. His researchinterests encompass the design of dependable networkedembedded systems and the robustness of networkingprotocols against environmental influences.

Salil S. Kanhere is an associateprofessor at the Schoolof Computer Science andEngineering at the Universityof New South Wales in Sydney,Australia. He obtained his B.E.in Electrical Engineering fromVJTI, Bombay, India in 1998and his M.S. and Ph.D., bothin Electrical Engineering from

Drexel University in Philadelphia, USA, in 2001 and2003 respectively. Salil’s current research interests arein the areas of sensor networks, mobile networking,vehicular communications and network security.

Kay Romer is a professorand director of the Institutefor Technical Informatics atGraz University of Technology,Austria. He held positions ofProfessor at the University ofLubeck in Germany, and seniorresearcher at ETH Zurich inSwitzerland. Prof. Romerobtained his Doctorate incomputer science from ETHZurich in 2005 with a thesis

on wireless sensor networks. His research interestsencompass wireless networking, fundamental services,operating systems, programming models, dependability,and deployment methodology of networked embeddedsystems, in particular Internet of Things, Cyber-PhysicalSystems, and sensor networks.

113

Documents

Mitigating Radio Interference in Large IoT Networks .... Sparber, C.A. Boano, S.S. Kanhere, and K. Romer: Mitigating Radio Interference in Large IoT Networks through Dynamic CCA Adjustment¨