Multichannel Reliability Assessment in Real World WSNs

Multichannel Reliability Assessment in Real World WSNsJorge Ortiz and David CullerUniversity of California at Berkeley

9th Int. Conf. on Information Processing in Sensor Networks (SPOTS Track)

April 12-16, 2010Stockholm, SwedenMotivationChannel diversity seen as necessary in industrial setting for reliable communicationStandards802.15.4eSP100.11aWirelessHART

It is thought that for sensor network deployments in industrial settings, multichannel communication is absolutely necessary for reliable packet delivery. This setting is characterized by surfaces with lots of metal and concrete that causes multipath induced narrow-band fading.

Three standards bodies have formed specifically to address this concern.2Our results demonstrate the contraryOpportunities where multichannel provides communication where single-channel multihop routing cannot are rareEvent when those opportunities exists are rarely importantWell-connected network on single channel provides enough diversity for reliable communicationOur contributionThis work formalizes assumptions that motivate multichannel in industrial settingsEvaluate multichannel utility in context of routingQuantify the opportunity where multichannel necessaryMultichannel often unnecessary for reliable delivery when routing is an alternative

Our results demonstrates that the utility of multichannel is small when taking into account multihop routing.

We formalize the set of assumptions used to motivate multichannel communication and show that the opportunity for multichannel communication is extremely rare in practice. We conclude that multichannel is not necessary for reliable packet delivery in the context of multihop routing.4RoadmapDiversity hides link variabilityStandards goals and assumptionsMotivating studyFormalize assumptionsIntroduce network facets to test assumptionsMultichannel Links (MCL) and Multichannel Triangles (MCT)ResultsQuantify the MCL and MCT occurrencesShow multichannel rarely helps when there is routingijcc2c1Sources of Loss:Collisions and External Interference

802.11802.15.4Animation to demonstrate collision:

InterferenceCollisionsCSMA, TDMAExternal noiseMicrowaves, cordless phones, Bluetooth

6Sources of Loss:Non-line of site communication

Animation to demonstrate this as well: Show Rob Poors loss graphs

Non-line of site communicationWeakens signal over spaceDestructive interference at certain locations (multipath loss)

7Diversity HelpsSpatial diversityUse multiple receiversMultiple antennasMultiple next-hop routing choicesFrequency DiversitySignal modulationDSSSChannel hoppingFHSSTime Diversity

Talk about the use of diversity technique show pictures of MIMO access points, a network8Standards Diversity RecommendationsTo address multipath and external interfaceMultichannel provides level of immunity against both loss sourcesInterference on current channelThe sender has load to offerInterference spans narrow bandTo support end-to-end reliabilityMultihop routingTopology-formation recommendations made

Conditions for multichannel may or may not hold in practice; how often do they actually occur?

9Current Claim:Multichannel diversity is required1Partly motivates standards decision to include multichannelDirectly motivates ISA SP100.11aEvaluates the WSN radio channel quality in industrial environmentLink quality varies substantially over timeMultipath induced narrow-band fading negatively affects link qualityMultichannel necessary for reliabilityD. Sexton, M. Mahony, M. Lapinski, and J. Werb. Radio channel quality in industrial wireless sensor networks. In SICON 05 Sensors for Industry Conference, 2005.Experimental methodology1D. Sexton, M. Mahony, M. Lapinski, and J. Werb. Radio channel quality in industrial wireless sensor networks. In SICON 05 Sensors for Industry Conference, 2005.6 motes with CC2420 Radio40 ft x 66 ft (12 m x 20 m)Round-robin transmission with local logging4 hours of continuous probingRecorded packet reception rates (PRR)

11No single channel provide best set of links1D. Sexton, M. Mahony, M. Lapinski, and J. Werb. Radio channel quality in industrial wireless sensor networks. In SICON 05 Sensors for Industry Conference, 2005.

Notes by animation.

0. This graph is from the motivating study behind the SP100.11a standards decision to use multichannel for communication. Generated by 6 motes using the CC2420 radio in an industrial machine room setting that is 43 ft x 66 ft (12 m x 20 m).

1. Links between pairs of nodes 12No single channel provide best set of links1D. Sexton, M. Mahony, M. Lapinski, and J. Werb. Radio channel quality in industrial wireless sensor networks. In SICON 05 Sensors for Industry Conference, 2005.

Links (1,2) ,(2,1)416132521343632454153565264621Notes by animation.

0. Lets examine a specific links and extract some of the observed properties1. Links between pairs of nodes 13No single channel provide best set of links1D. Sexton, M. Mahony, M. Lapinski, and J. Werb. Radio channel quality in industrial wireless sensor networks. In SICON 05 Sensors for Industry Conference, 2005.

Links (1,2) ,(2,1)Channel 13, 25 High lossChannel 25,~30% lossChannel 13,~70% loss21Notes by animation.

0. Lets examine a specific links and extract some of the observed properties1. Links between pairs of nodes 14No single channel provide best set of links1D. Sexton, M. Mahony, M. Lapinski, and J. Werb. Radio channel quality in industrial wireless sensor networks. In SICON 05 Sensors for Industry Conference, 2005.

Links (1,2) ,(2,1)21PRR(12, 25)=30%PRR(21, 25)=100%PRR(12, 13)=70%PRR(21, 13)=100%No single channelgood for all linksNotes by animation.

0. Lets examine a specific links and extract some of the observed properties, links between pairs of nodes are asymmetricNo channel is good for all links; Examining all channels for all links, this is observed.

15and from this they conclude1D. Sexton, M. Mahony, M. Lapinski, and J. Werb. Radio channel quality in industrial wireless sensor networks. In SICON 05 Sensors for Industry Conference, 2005.There was no channel that allowed for reliable communications over all paths for all units throughout the entire test period None of the paths were very symmetric for all channels. The results of these experiments clearly show that a frequency agile approach might be more robust than a single channel approach. . . These implicitly identify instances where multichannel provides reliable delivery where single channel cannotConclusion is soundbut only if we consider one-hop, direct communicationNotes by animation.

0. This graph is from the motivating study behind the SP100.11a standards decision to use multichannel for communication. Generated by 6 motes using the CC2420 radio in an industrial machine room setting that is 43 ft x 66 ft (12 m x 20 m).

1. Links between pairs of nodes 16Formalizing the observations:Multichannel Links (MCL)What is an asymmetric link?A link for distinct nodes i and j is asymmetric if the link PRR(i,j) >= T and PRR(j,i) < T for some usability threshold, T.How can frequency agility improve problems with asymmetric links?We refer to this link as a Multichannel Link

ijcc2c1Formalizing the observation: Multichannel Triangle (MCT)There was no channel that allowed for reliable communications over all paths for all units throughout the entire test period .Formally: 3 distinct nodes, i, j, and k, can all communicate bi-directionally on some channel, but no channel where all 3 can communicate.

c1 c2 and c1 c3Note path from i to j when c2 = c3 Routing can be used as an alternative to multichannel communicationBoth MCLs and MCTs are instances in the networks believed to provide connectivity where single channel cannot. This is true in the case where only direction communication is possible.

However, if we consider routing as an alternative to switching channels, we find that MCT instances may not be as important for identify an opportunity for multichannel communication.Note that if c2=c3, then there is a two-hop path from the node i to node j through k, with the same route and transmission stretch as the multichannel solution.

We test for the existence of MCLs and MCTs in live networks and evaluate their importance for providing reliable communication.18Environments testedIndustrial machine room95 ft x 40 ft (28.9 m x12.2 m)20 TelosB motesComputer room28 ft x 28 ft (8.5 m x 8.5 m)23 TelosB motesOffice setting128 ft x 128 ft (39 m x 39 m)55-60 MicaZ motes

Experimental MethodologyMotes with CC2420 RadioEach mote sends 100 packets at 20 millisecond inter-packet interval, round robin, on each channelListening motes log packets to flashMultiple 802.15.4 channels probedMultiple experimental runsLog AnalysisReliability determined through path existenceLogs contain source and sequence numberConnectivity graph constructed on each channelLink PRR calculated for each observed linkUsability threshold, T, applied in the construction of each graphMCL and MCT locator processed over every connectivity graph~1.7 million packets sent, >3500 graphs examined3 runs (12 hours) in machine room, 2 (8 hours) in computer room, 17 in office setting (10 days) continuous probingEnvironmental Comparisons6 random nodes selectedSimilar patterns observed in all 3 environments

Link(36, 13)=53% lossLink(63, 13)=2% lossLink(36, 22)=70% lossLink(63, 22)=4% lossComputer Room LinksEnvironmental ComparisonLoss pattern observed similar but less narrowMay affect multichannels opportunity to find an alternative frequency

Industrial EnvironmentTestbed EnvironmentExperimental Results: MCL CountMany unidirectional links foundVaries from 8-70% of the links being unidirectional on some graphConnectivity still maintained throughout in machine room and computer room, ~95% of time on testbed2-6% of links in all graphs for all settings are MCL linksThis may suggest a small gray region in our deployments[24, 27] locations in the network where connectivitybetween the sender and receiver are at the edge of radio connectivity.Generally, the population of links in the gray regionis small, since reliable communication is desirable andgray-region links have more unpredictable link quality [13].Sparser networks have more links at the edge of networkconnectivity and thus there may be some links that are abovethe goodness threshold on some channels and not on others,raising the likelihood that they are MCLs. In practice,sparse network deployments should be avoided. The WirelessHART deployment guides specifically suggest that eachnode have at least 3 neighbors [9]. If these are followed,the network should be very well connected, decreasing thelikelihood of MCLs24The key questionAre these links important? Our data show:Never important in machine roomNever important in computer room1.8% of occurrences on testbed prevent network partition

ijcc2c1c important for preventing network partitionWhich tradeoff do youwant to live with?Each time there was a partition in the network, there was no channel available to connect the remaining connected components. We explored this further, by examining all the connectivity graphs, for all experimental runs and for all thresholds. On the testbed, only 1.8% of the connectivity graphs (2720 graphs examined) had a corresponding connected multichannel graph.

This may indicate much wider noise correlation across channels in the operating frequency band. This also directly addresses the assumption made in the standards about FHs ability to avoid interference. It can only improve reliabilityif there is an opportunity to transmit on some channel that is interference free. According to our data, that opportunity is quite rare. Furthermore, this number is optimistic since there must not only be a free channel, but the protocol must find it to be successful. The wider the interference is across channels, the smaller the probability of finding a free channel for transmission.25Experimental Results: MCT CountSingle channel set (SC set)All distinct node triples connected on a single channelMultichannel set (MC set)All distinct node triples connected on any channelMCT setAll distinct node triple in the multichannel set and not in the single channel setMCT occurrence rate = |MCT set|/|MC set|MCT Count: Industrial Machine room6-hop network diameterMaximum occurrence rate is 3500 graphs examined3 runs in machine room, 2 in computer room, 17 in office setting~12 days worth of experimental runsPath existence in underlying connectivity graph within a time-slice assumes successful packet delivery.39Standards Address Concerns for Industrial SettingsThree main standards bodies formed to address concerns in industrial settings802.15.4eSP100.11aWireless HARTGeneral goalsReliable packet deliveryLong deployment lifetimeAdjustable QoSSP100.11a Frequency Hopping SimulationSupports 5 hopping patternsWe ran pattern index 1 19, 20, 24, 16, 23, 18, 25, 14, 21, 11, 15, 22, 17, 13, 26Connectivity graph much worse without backlistingConnectivity graph same to remaining on single channel with backlistingRandom nodes selected to transmit on random channel