ECE 555 Real-Time Embedded Systems Multi-Channel Communication in Wireless Sensor Networks(WSN)

1. ECE 555Real-Time Embedded SystemsMulti-Channel Communication in Wireless Sensor Networks(WSN) Presented by Rukun Mao Nov. 13th 2008

2. Reference • Yafeng W., Stankovic J.A., Tian H, Shan L, “Realistic and Efficient Multi-Channel Communications in Wireless Sensor Networks” • Hieu Khac Le, Dan Henriksson, and Tarek Abdelzaher, “A Practical Multi-Channel Media Access Control Protocol for Wireless Sensor Networks”

3. Outline • Introduction • Experiments on Multi-Channel Reality • Tree Based Multi-Channel Protocol (TMCP) • Minimum Interference Channel Assignment Problem • Performance Evaluation • Conclusion

4. Introduction • Current WSN hardware such as Micaz and Telos provide multiple channels • Improve network throughput • Provide reliable and timely communication services • Recently MAC layer multi-channel protocols are proposed • To assign different channels to two-hop neighbors and coordinate channel switching • Also called node-based schemes

5. Introduction • Practical issues for node-based scheme • A large number of orthogonal channels are needed in dense networks. • Require precise time synchronization at nodes. • Channel switching delay and scheduling overhead. • Complex.

6. Non-orthogonal Channel Interferences • Place three Micaz motes in a line • One transmitter, one receiver, and one jammer • The jammer’s transmission is synchronized with the transmitter. Transmitter : channel 11 Receiver : channel 11 Jammer : channel 12 (adjacent) channel 13 (2 channel away)

7. Interferences with 802.11 networks • Put 8 pairs of Micaz motes closely in a department office with 802.11 networks • Each pair uses unique channel and all 8 channels are orthogonal. • Multi-channel protocols must have capabilities to work well with a small number of available channels.

8. Impact of Time Synchronization Errors 1 2 3 4 5 • Each node with unique channel and all are synchronized. • A time period is divided into 2 time slots • 1st time slot • Nodes in odd positions send packets • Nodes in even positions receive packets • 2nd time slot is vice versa

9. TMCP • To overcome two problems in practical networks • The small number of available orthogonal channels. • Unavoidable time errors. • Data collection traffic • Multiple information flows generated at sensor nodes converge to the base station.

10. TMCP • Main idea • Partition the whole network into multiple vertex-disjoint sub-trees all rooted at base station • Allocate different channels to each sub-tree. • Forward each flow only along its corresponding sub-tree. • 3 components • Channel detection (CD) • Channel assignment (CA) • Data communication (DC)

11. TMCP • CD finds available orthogonal channels • Two motes are used to sample the link quality, and we selected good link qualities with non-adjacent channel. • Assume we have k channels at this point. • CA partitions the whole network into k sub-trees and assigns one unique channel to each sub-tree • Inter-tree interference is eliminated (non-adjacent) • Intra-tree interference is minimized ( same channel)

12. TMCP • DC manages the data collection through each sub-tree • Assume the base station is equipped with multiple radio transceivers. • Without time synchronization

13. Model and Problem Definition • The goal is to minimize intra-tree interferences. • Assume that a sensor network is a static. • The interference set of a node u is defined as • INT(u) = {v|vєD(v, Iv), where D(v, Iv) is the interference disk with node v in its center and radius Iv} (*) • Interference value: int(u) = |INT(u)| • The intra-tree interference value of a tree T is defined as • int(T) = max{int(u): u is a non-leaf of T} * M. Burkhart, P. V. Rickenbach, R. Wattenhofer, and A. Zollinger, “Does topology control reduce interference,” in ACM MobiCom, 2004.

14. PMIT Algorithm • Apply Breadth-First search algorithm from the base station to construct a fat tree. • Nodes keep height and have multiple parents on the fat tree. • The tree is a shortest path tree. • Execute the channel allocation one-by-one level from top to bottom on the fat tree • For each node, choose an optimal tree and add this node to bring the least interference to this tree. • Selects a parent which has the least interference value. • Nodes with fewer parents first, because they are less free to choose channels.

15. Evaluation of the PMIT Algorithm • Simulations parameters • 200m x 200m field • 250 nodes are uniformly distributed • Communication range is 10~35m • Interference range is 1.5 times as the communication range

16. Evaluation of the PMIT Algorithm

17. Performance with different node density Performance comparison of TMCP and MMSN

18. Evaluation in a Real Testbed • Experiment setup • A real testbed with 20 Micaz motes. • Four motes are laid closely together to act as a base station with four transceivers.

19. TMCP effectively reduces interferences and mitigates congestion at nodes. • TMCP works well in a real sensor network.

20. Conclusion • Multiple channels to improve network performance in WSNs. • Realities in WSNs • Small number of available channels • Synchronization errors • TMCP • Work with a small number of channels. • Work without the need of time synchronization. • Decrease potential radio interferences. • Three components: Channel Detection, Channel Assignment(CA), Data Communication(DC)

21. Critique Critique1: Communication between nodes in different sub-trees is blocked. Critique2: Adjacent channels are not used, and limit bandwidth is not fully utilized. Critique3: The assumption that interference sets of all nodes are already know is not applicable under certain circumstance. Critique 4: interference range/communication range ratio is set at 1.5 without justification.

22. A Practical Multi-channelMedia Access Control Protocolfor Wireless Sensor Networks

23. Outline • Introduction • Related Work • Protocol Analysis • Implementation • Experimental Results • Conclusion

24. Introduction • A typical sensor network • There is lots of collisions and interference

25. Introduction The solution is multi-channel Utilizing multi-channel at MAC layer give any application benefit from multi-channel for free.

26. Related Work • Many multi-channel MACs were developed for WSN and ad-hoc networks. • All of them have at least one of below drawbacks • Assume channel switching time is negligible • Require time synchronization • Require multi-radio or special radio interface • As a result • Most of them were only illustrated in simulation • The rest need special sensor node with multi-radio

27. Contribution • Uses only one half-duplex radio interface • Considers real channel switching overhead • Is lightweight with small memory and code footprint • Does not require any input from the upper layers

28. Protocol Analysis • Each node has a home channel • A node stays at its home channel while listening to incoming messages • If node A want to send a message to node B, A need to switch to B’s home channel • It is desirable to minimize cross-channel communication and maximize same channel traffic. This is directly related to the objective of the K-Way cut problem.

29. The K-Way Cut Problem A K-Way Graph Partitioning: • Given a graph G(V, E) • Given a number K • Divide the graph in to K sub-graphs such that the total weight of edges across the sub-graphs is minimum.

30. “K-Way Cut Problem” Algorithm Some intuitions • More channels are only allocated when needed • Nodes with better view of network traffic will initiate the cut • Nodes with less information should act locally to minimize cross-channel communication

31. Three Rules • Rule 1: Nodes start at channel F0 • Rule 2: Channel Advancement • Each node jperiodically broadcasts <#successful channel access = sj, #failed channel access fj> to its neighborhood. • Node j estimates the probability of successful access the channel α If α is below a threshold, node considers advancing channel with the probability β • Rule 3: Channel Follow If a node A figures out its neighbor B switches channel, A consider going to the channel in which it has highest data flow.

32. The switch fluctuation problem Use a simple control scheme • Consider a node at channel i having successful access probability αi(k) at time k - Probability to switch from channel i to channel i+1 is βi,i+1(k) at time k If αi(k) < αref βi,i+1(k) βi,i+1(k-1) + Kc(αref – αi(k)) If αi(k) ≥ αref βi,i+1(k) βi,i+1(k-1) – Kc1(αi(k)- αref )

33. Implementation • Implemented on MicaZ mote • Same code works for both TOSSIM and MicaZ • Channel switching time is ~50ms • Code footprint is around 9KB and RAM footprint is less than 1KB

34. Experimental Results

35. Experimental Results

36. Experimental Results

37. Conclusion Presented MAC has the following advantages • Using only one half-duplex radio interface • Having channel switching time conforming to reality • Lightweight enough to run on MicaZ mote with small memory and code footprint • Not requiring any input from the upper layers

38. Critique • There is no discussion about interference among K available channels. • Threshold γ is determined based on worst-case delay d, which will change during various network status. • Only compare with single-channel MAC. Should also compare with other multi-channel MAC

39. comparison

40. Acknowledgement • I would like to thank Yu-Chun Chang(paper 1) and Hieu Khac Le (paper 2) for sharing PPT slices with me, which make this presentation possible.

41. Thank You • Questions?