1. 1 MMSN: Multi-Frequency Media Access Control for Wireless Sensor Networks Gang Zhou, Chengdu Huang, Ting Yan, Tian He John. A. Stankovic, Tarek F. Abdelzaher Department of Computer Science University of Virginia

2. 2 Outline Motivation State of the Art Overhead Analysis Contribution – New Protocol Framework Frequency Assignment Media Access Design Performance Evaluation Conclusions

3. University of Virginia 3 Ad Hoc Wireless Sensor Networks Sensors Actuators CPUs/Memory Radio Minimal capacity Self-organize

4. University of Virginia 4 Motivation Limited single-channel bandwidth in WSN 19.2kbps in MICA2, 250kbps in MICAz/Telos The bandwidth requirement is increasing Support audio/video streams (assisted living, …) Multi-channel design needed Hardware appearing Multi-channel support in MICAz/Telos More frequencies available in the future Collision-based: B-MAC Scheduling-based: TRAMA Hybrid: Z-MAC Software still lags behind

5. University of Virginia 5 State of the Art: Multi-Channel MAC in MANET  Require more powerful hardware/multiple transceivers Listen to multiple channels simultaneously [Nasipuri 1999], [Wu 2000], [Nasipuri 2000], [Caccaco 2002]  Frequent Use of RTS/CTS Controls For frequency negotiation Due to using 802.11 Examples: [Jain 2001], [Tzamaloukas 2001], [Fitzek 2003], [Li 2003], [Bahl 2004], [So 2004], [Adya 2004], [Raniwala 2005]

6. University of Virginia 6 Basic Problems for WSN Don’t use multiple transceivers Cost Form factor Packet Size 30 bytes versus 512 bytes (or larger) in MANET RTS/CTS Costly overhead

7. University of Virginia 7 RTS/CTS Overhead Analysis MMAC: RTS/CTS frequency negotiation 802.11 for data communication RTS/CTS are too heavyweight for WSN: Mainly due to small packet size: 30~50 bytes in WSN vs. 512+ bytes in MANET From 802.11: RTS-CTS-DATA-ACK From frequency negotiation: case study with MMAC

8. University of Virginia 8 Contributions A new multi-frequency MAC, specially designed for WSN; Single half-duplex radio transceiver; Small packets sizes; Developed four frequency assignment schemes Supports various tradeoffs Toggle transmission and toggle snooping techniques for media access control; An optimal non-uniform backoff algorithm, and a lightweight approximation;

9. University of Virginia 9 Frequency Assignment F1 F2 F3 F4 F5 F6 F7 F8 Reception Frequency Complications Not enough frequencies Broadcast

10. University of Virginia 10 Frequency Assignment When #frequencies >= #nodes within two hops When #frequencies < #nodes within two hops Exclusive Frequency Assignment Implicit-ConsensusEven SelectionEavesdropping Both guarantee that nodes within two hops get different frequencies The left scheme needs smaller #frequencies The right one has less communication overhead Balance available frequencies within two hops The left scheme has fewer potential conflicts The right one has less communication overhead

11. University of Virginia 11 Media Access Design F1 F2 F3 F4 F5 F6 F7 F8 Issues: Packet to Broadcast Receive Broadcast Send Unicast Receive Unicast No sending/no receiving

12. University of Virginia 12 Media Access Design Different frequencies for unicast reception The same frequency for broadcast reception Time is divided into slots, each of which consists of a broadcast contention period and a transmission period. T b c T tran T b c T …...

13. University of Virginia 13 Media Access Design Case 1: When a node has no packet to transmit

14. University of Virginia 14 Media Access Design Case 2: When a node has a broadcast packet to transmit

15. University of Virginia 15 Media Access Design Case 3: When a node has a unicast packet to transmit

16. University of Virginia 16 Toggle Snooping During “ “, toggle snooping is used

17. University of Virginia 17 Toggle Transmission  When a node has unicast packet to send  Transmits a preamble  so that no node sends to me  so that no node sends to destination  We let

18. University of Virginia 18 Simulation Configuration ComponentsSetting SimulatorGloMoSim Terrain(200m X 200m) Square Node Number289 (17x17) Node PlacementUniform Payload Size32 Bytes ApplicationMany-to-Many/Gossip CBR Streams Routing LayerGF MAC LayerCSMA/MMSN Radio LayerRADIO-ACCNOISE Radio Bandwidth250Kbps Radio Range20m~45m Confidence IntervalsThe 90% confidence intervals are shown in each figure

19. University of Virginia 19 Performance with Different #Physical Frequencies - With Light Load ① Performance when delivery ratio > 93% ② Scalable performance improvement ③ Overhead observed when #frequency is small ④ More scalable performance with Gossip than many-to-many traffic

20. University of Virginia 20 Performance with Different #Physical Frequencies – With Higher Load ① When load is heavy, CSMA has 77% delivery ratio, while MMSN performs much better ② MMSN needs less channels to beat CSMA, when the load is heavier

21. University of Virginia 21 Performance with Different System Load Observation: CSMA has a sharp decrease of packet delivery ratio, while MMSN does not. Reason: The non-uniform backoff in time-slotted MMSN is tolerant to system load variation, while the uniform backoff in CSMA is not.

22. University of Virginia 22 Conclusions First multi-frequency MAC, specially designed for WSN, where single-transceiver devices are used Explore tradeoffs in frequency assignment Design toggle transmission and toggle snooping Theoretical analysis of an non-uniform back-off algorithm MMSN demonstrated scalable performance in simulation

23. University of Virginia 23 The End! Thanks to anonymous reviewers for their valuable comments!

24. University of Virginia 24 Performance with Different Node Densities

25. University of Virginia 25 Backup Slides: Optimal Non-Uniform Backoff

26. University of Virginia 26 Even Selection Frequency Assignment Beacon (multiple times) to collect nodes’ IDs within two hops Frequency decision is made sequentially in the increasing order of nodes’ IDs When making a decision, randomly choose one of the least chosen frequencies (once no unique ones left) Notify neighbors of decision NOTE: Frequency assignment happens once (or a few times)

27. University of Virginia 27 Back Off Period - Slotted Backoff into a slot Transmit at end of a slot

28. University of Virginia 28 Non-Uniform Backoff: Motivation & an Optimal Solution Uniform backoff Non-uniform backoff Let 34 slices of length T TS ; 68 nodes compete for the channel --- a timer fires An optimal distribution is presented in the paper Uses recursive computation Distribution depends on node density A simple approximation is needed

29. University of Virginia 29 Non-uniform Backoff: A Simple Approximation implementation