Wireless network measurementUnderstanding Packet Delivery Performance in Dense Wireless Sensor NetworksMotivationWhy focus on packet delivery?Experiment designExperiment environmentSensor nodes: Mica motesExperiment softwarePacket delivery at physical layerPhysical layer encoding schemePacket loss in three environmentsPacket loss with different transmit powerPacket delivery using different coding schemesHow does reception rate vary with distance from the transmitter?Are the results representative?ImplicationsCan signal strength predict link quality?Can sophisticated physical layer coding mask the gray area?Packet delivery correlationTemporal characteristics of packet deliveryPacket delivery at MAC LayerTopologyTraffic patternPacket loss distribution w/ retransmissionPacket delivery efficiencyAsymmetry in packet deliveryConclusions1Wireless network measurementWe’ll look at three papersWireless LAN (WLAN) usage study•at Dartmouth collegeLink measurement in multihop network•RoofnetPacket delivery in sensor networks•Medium-size sensor networks2Understanding Packet Delivery Performance in Dense Wireless Sensor NetworksJerry Zhao & Ramesh GovindanSenSys ‘033Motivationwireless sensor networks deployed in harsh environmentusing low power radio (not much frequency diversity)densely deployed quantitative understanding of packet deliveryphysical-layer measurement (w/o interfering transmission)•Dependence on environment, physical-layer coding scheme, and receiver?MAC layer measurement (w/ interfering transmission)•Effect of carrier sense, MAC layer retransmission?4Why focus on packet delivery? Very basic Packet delivery ratio determines energy efficiency & network lifetimePoor packet delivery may degrade application performanceImportant for evaluating almost all communication protocols5Experiment designExperiment environmentIndoor environment (office building)Natural habitatEmpty parking lotSensor nodesMica motesExperiment software6Experiment environmentI: office building2m x 40m hallwayharsh environment: multipath reflectionH: habitat150m x 150m segment of a state parkdownhill slopemulti-path due to foliage & rocksO: open parking lot150m x 150m open parking lot“benign” environment7Sensor nodes: Mica motes4MHz Atmel processor, 512KB flash memoryASK (amplitude shift keying)low-power 433Mhz radioomni-directional antennanominal throughput of 20KbpsTinyOSphysical-layer error detection/correctionMAC: CSMA/CA, link-layer ACK8Experiment softwareTraffic generatorPeriodic generationGeneration following a distributionUpload experiment parametersInformation logger (in TinyOS)9Packet delivery at physical layertopology: 60 motes placed in a linesingle transmitter: head of line0.5m apart0.25m apart near the edge of the comm. rangeremove some nodes near the transmittertraffic: periodic transmission, 1 pkt/secdisable TinyOS MAC & retransmissionphysical-layer codingthree coding schemestransmit powerhigh, medium, low10Physical layer encoding schemeSECDED (Single Error Correction & Double Error Detection)TinyOS defaultconvert each byte into 24 bitscan detect 2 bit errors & correct one bit errorManchester encodingconvert a byte into 16 bitsdetect an error out of 2 bits4-bit/6-bit scheme (4b6b)encode one byte into 12 bitsdetect 1 bit error out of 6 bits11Packet loss in three environments4b6b coding, high Tx powerIHO12Packet loss with different transmit powerbetter delivery under lower power (possibly due to reduced multi-path problems)HML4b6b coding, indoor environment13Packet delivery using different coding schemesSECDED is much better (however also consumes more bandwidth)4b6b & Manchester coding similar performancehigh tx power, indoor environment14How does reception rate vary with distance from the transmitter?Gray area due to multipath problemsWidth of gray area significantIOHhigh tx power, 4b6b coding15Are the results representative?losses caused by multipath difficult to be overcome by low-power radioLow frequency diversity16Implicationslikelihood of links falling into gray area is highshortest path (in hop count or geographic distance) may not be gooda long hop may have high loss rateother bad consequencesnodes need to carefully select neighbors based on measured packet delivery rate17Can signal strength predict link quality?Unfortunately, NOIndoor, high Tx Power18Can sophisticated physical layer coding mask the gray area?Theoretical calculationNot necessarily, SECDED has the lowest effective bandwidth19Packet delivery correlationI & O show noticeably higher correlation than HImplication: at the physical layer, independent losses are a reasonable assumptionI O H20Temporal characteristics of packet deliveryLarge variations in average reception rate time varying packet losses in gray areain 40sec windowsstdev of average delivery ratios21Packet delivery at MAC LayerTinyOS MACCSMA/CA: random backoff upon carrier senseno RTS/CTSlink layer ACK: send 4 byte ACK to the senderretransmit up to 3 times, when there’s no ACKExperiment methodologyfix physical-layer coding (4b6b)three environmentstopologytraffic pattern22Topologymultihop networkmedium transmit powerindoor office building62 motes, one every officenode degrees: 15-18 open parking lot similar topologynode degrees: 17-20natural habitat4 x 12 grid, 0.75 m between two nodesnode degrees: 6-823Traffic patterneach node sends k packets per secondk=0.5, 1, 2, 4 pkts/secinter-packet intervalexponentially distributedavoid packet synchronizationeach node unicasts packets (36 bytes) to neighbors in round-robin fashionneighbor table: periodic broadcastnot intended to model application traffic24Packet loss distribution w/ retransmissionmany packet lossespacket losses due to environment noise or collision?better MAC (e.g., S-MAC) is requiredIOH30% of links have loss rate > 0.525Packet delivery efficiency# of distinct pkts received / # of pkts transmitted Efficiency low - better MAC is required26Asymmetry in packet deliverySignificant asymmetry, should try to avoid such linksabsolute difference of packet loss rates in two directions of a linkindoor27Conclusionsunderstand packet delivery in dense