Wireless Sensor Network Deployment Lessons Learned Steven Lanzisera Environmental Energy Technologies Division, LBNL 21 January 2011

2 Wireless Sensor Networks in 2002

3 Project Overview – Commercial Buildings LBNL Building 90 – 90,000 s.f. office Plug-in device metering network 6 months of data collection (on going)

4 Current Building 90 Deployment 300+ ACmes installed throughout building 500 at full build-out – 0.5 nodes per 100 s.f , CSMA, 6LowPAN, RPL (draft), SMAP (custom) Power, apparent power, energy every 10s 245 ft / 75 m

5 Residential Deployments 5 Houses (4 bay area, 1 Boston area) – ~80 nodes installed per house – 1 gateway – Data reported every 10s – 6 months of data collection (ongoing) Gearing up for 70 homes in next year – nodes per home

6 Overview Zigbee & Standards Context Why Wireless Networks Fail – Communication Issues – Other Issues Final Thoughts

7 Overview Zigbee & Standards Context Wireless Network Characteristics Final Thoughts

8 IEEE – Overview Emphasis of IEEE is: – low-cost, low-speed ubiquitous communication between nearby devices with little to no underlying infrastructure – Nominal communication at 250 kb/s – 10m communication range assumed – to meet embedded constraints, several PHY layers are available Key technology features are: – collision avoidance through CSMA/CA – integrated support for secure communications (128-bit AES encryption) – power management functions such as link quality and energy detection – 16 channels in the 2.4 GHz band – star and mesh topologies can theoretically be built

9 IEEE – MAC Layer There are two general channel access methods: Non-Beacon Network: – simple, traditional multiple access system used in simple peer networks – standard CSMA conflict resolution – positive acknowledgement for successfully received packets Beacon-Enabled Network – can be used in beacon-request mode without superframes – superframe structure - network coordinator transmits beacons at predetermined intervals – dedicated bandwidth and low latency – low power consumption mode for coordinator

10 Mischa Dohler & Thomas ICC 2009 IEEE – MAC Layer Super-Frame Structure for Beacon-Enabled Mode:

11 IEEE MAC in Practice Beacons are rarely used Contention based networks are common Zigbee doesn’t require one or the other – Often implemented without beacons

12 ZigBee ZigBee in short: – international alliance for wireless control applications; SIG certifies platforms – based on IEEE PHY & MAC – millions of products today are embedding a chipset of the ZigBee family – Small numbers of ZigBee certified products are available Provides network through application layers Most devices listen all the time (and must be mains powered)

13 Full function device Reduced function device Communications flow Mesh for full function Listen all the time Star for reduced function Sleep between transmissions Zigbee General Topology

14 Overview Zigbee & Standards Context Why Wireless Networks Fail – Communication Issues – Other Issues Final Thoughts

15 Assumptions Multi-hop network of low power wireless sensors Communicating using IEEE radio chips (16 frequency channels in the 2.4GHz band) 2.4 GHz Channels GHz 5 MHz 2.4 GHz PHY A B C D E F

16 Single Channel Solutions The quality of a link varies with frequencywith time there is no “best channel”!

17 Channel Success Probability vs. RSSI

18 Second Challenge: Multipath Fading

19 Second Challenge: Multipath Fading ch.11

20 Second Challenge: Multipath Fading ch.11ch.12 0% reliability100% reliability

21 Second Challenge: Multipath Fading ch.11 ch.13 ch.15 ch.17 ch.12 ch.14 ch.16 ch.18 ch.19 ch.21 ch.23 ch.25 ch.20 ch.22 ch.24 ch.26

22 Impact of Interference Noise Interference 2.4 GHz Channels GHz 5 MHz 2.4 GHz PHY Relative Noise Power

23 Mischa Dohler & Thomas ICC 2009 Interference continued

Spectrum & Interference

25 BT & WLAN interfere with ZigBee Theoretical results indicate that interference is an issue [SPC07]:

26 Mischa Dohler & Thomas ICC 2009 Reservation vs. Contention MAC Example of throughput versus offered load: Offered Load Normalized Throughput reservation based

27 Data Collection Network Reliability % of Possible Packets

28 Latency Multihop latency suffers because of communication failures 1-hop latency is < 10ms if it works Backoff after failure increases latency Tests w/50 ms backoff & 5 hops – Average latency ~100ms – 90% of packets arrive by 500ms – MAC time out occurs before 99% (1s)

29 Link Length & Routing Stability In B90 (Office building) Typical links 30ft Longest (reliable) links 50ft 60% of routes didn’t change this week 20% of routes changed >5 times (Check daily)

30 Overview Zigbee & Standards Context Why Wireless Networks Fail Final Thoughts

31 Why Zigbee? Zigbee is a “new” protocol Limited industry experience Known for interoperability, reliability problems Latency, packet size are far from ideal Very few products on the market Plus side: could be cheap(er) – Somewhat lower power

32 Consider WiFi Over 2M WiFi chips shipped every day Same MAC, but better coexistance SEP 2.0 is not linked to a PHY SEP 2.0 and other Zigbee will work on IP Power difference isn’t large (0.3W vs 0.1W) Cost difference negligible

33 Recommendations Early study of Zigbee in intended environment – Multihop network – Test latency, reliability, etc Consider draft SEP 2.0 (available on the web)

34 Summary Lots of Zigbee-like networks deployed Lots of problems negatively impact the network Need to study Zigbee performance – Because it’s not well known like WiFi – Will kill the project if performance is poor