1. 1 Using Multiple Channels and Spatial Backoff to Improve Wireless Network Performance Nitin Vaidya University of Illinois at Urbana-Champaign www.crhc.uiuc.edu/wireless MURI Review Meeting, September 12, 2006

2. 2 Sharing the Spectrum Classification of approaches  Temporal : Traditional contention resolution  Spatial : Spatial backoff  Spectral : Multi-channel systems

3. 3 Multi-Channel Wireless Networks: Capacity with Constrained Channel Assignment Joint work with Vartika Bhandari

4. 4 Channel Model  c channels available  Bandwidth per channel W Channel 1 Channel 2 Channel c

5. 5 Channel-Interface Scenarios  Common scenarios today 1 1 1 1 c c Single interfaceMultiple interfaces

6. 6 Fewer Interfaces than Channels  An interface can only use one channel at a time Channel 1 Channel c Single interface, multiple channels

7. 7 Interface Constraint  Throughput is limited by total number of interfaces in a neighborhood  Interfaces, a limited resource k nodes in the “neighborhood”  throughput ≤ k * W  (for single interface per node)

8. 8 Capacity with Multiple Channels  How does capacity scale when number of channels c is increased?  Depends on constraints on channel assignment to the interfaces Capacity as defined by [Gupta & Kumar]

9. 9 Unconstrained Channel Assignment Pre-MURI work [Kyasanur05MobiCom] Channels Network Capacity Single interface nodes can utilize multiple channels effectively

10. 10 Constrained Channel Assignment  Hardware limitations  Low cost, low power transceivers  Limited tunability of oscillator  Policy issues  Dynamic spectrum access via cognitive radio: secondary users in a band only when primary inactive 123456

11. 11 Network Model c channels W bandwith per channel n nodes randomly deployed over a unit area torus Interface can switch between f channels: 2 ≤ f ≤ c c = O(log n) 1 2 3 4 5 6 … … … c Each node has one interface s(1) s(2) s(f) … …

12. 12 Network Model  Motivated by [Gupta & Kumar]  Each node is source of exactly one flow  Chooses its destination as node nearest to a randomly chosen point

13. 13 Impact of Switching Constraints Connectivity: A device can communicate directly with only a subset of the nodes within range Bottleneck formation: Some channels may be scarce in certain regions, causing overload on some channels/nodes (1, 2) (2, 3) (1, 3) (2, 5) (7, 8 ) (6, 7) (3, 6) (5, 6) (4, 5)

14. 14 Proposed Models  Adjacent (c, f) assignment –A node can use adjacent f channels –Model encompasses untuned radio model  Random (c, f) assignment –A node can use randomly chosen f channels  Spatially correlated assignment

15. 15 Adjacent (c, f) Assignment f=2 c=8  Each node assigned a block of adjacent f channels  c – f + 1 possibilities  A node can use channel i with probability = minimum {i, c-i+1, f, c-f+1} /c

16. 16 Random (c, f) Assignment  Each node uses a random f-subset of channels  A node can use channel i with probability f/c f=2 c=8

17. 17 Spatially Correlated Assignment  N randomly located pseudo-nodes, each assigned a channel  Nodes close to a pseudo-node blocked from using the pseudo-node’s channel  Captures primary- secondary users 1 2 R R

18. 18 Results at a Glance Unconstrained assignment Adjacent (c,f)Random (c,f) Use c channels Use f common channels f

19. 19 Adjacent (c, f) Assignment Necessary condition on range r(n)  Capacity upper bound = c

20. 20 Lower Bound Construction Divide torus into square cells of area a(n) Cell structure based on [El Gamal] r(n) Transmission range r(n)

21. 21 Lower Bound Construction  Notion of preferred channels:  Probability that a node has that channel is at least f/2c  Includes most channels (except the fringe)  Each node has at least f/2 preferred channels  By choice of a(n): Every cell has Θ(log(n)) nodes capable of switching on each preferred channel w.h.p.

22. 22 Routing of Flows Straight-line routes for long flows. Detour routing for short Flows Ensure  (c/f) hops S D P

23. 23 Channel Transition Strategy (1, 2, 3) (4, 5, 6) Adjacent (6, 3) assignment Preferred channels : 2,3,4,5 (3, 4, 5) (4, 5, 6) (2, 3, 4) (1, 2, 3) 2 2 3 4 5 5 (4, 5, 6) (2, 3, 4) ( 3, 4, 5) Use randomly chosen preferred channel available at source (channel 2) Start transitions to get to a preferred channel at destination (channel 5)

24. 24 Random (c,f) Channel Assignment  Required range for connectivity smaller than adjacent (c,f)  However, at minimum range, all channels not sufficiently represented in each cell  Our lower bound construction is not tight: Uses larger range than minimum for connectivity

25. 25 Conclusion: Multi-Channel Networks Unconstrained assignment Adjacent (c,f)Random (c,f) Use c channels Use f common channels f Even when f=2, get capacity benefit of √c

26. 26 Conclusion: Multi-Channel Networks Unconstrained assignment Adjacent (c,f)Random (c,f) Use c channels Use f common channels f Even when f=2, get capacity benefit of √c cf

27. 27 Conclusion: Multi-Channel Networks  Constrained channel assignment may be mandated by cost/complexity/policy constraints  Possible to get significant benefits with little flexibility in channel switching  Open issues  Closing the gap for random assignment  Spatially correlated assignment  Protocol design

28. 28 Sharing the Spectrum Classification of approaches  Temporal : Traditional contention resolution  Spatial : Spatial backoff  Spectral : Multi-channel systems

29. 29 Spatial Contention Resolution with Carrier Sense Protocols Joint work with Xue Yang

30. 30 Contention Resolution  Temporal Approach: Adapt channel access probability  number of transmissions in a contention region = 1  Spatial Approach: Adapt contention region  number of transmissions in a contention region = 1

31. 31 Contention Resolution Temporal Approach: Adapt access probability Number of transmissions in a contention region = 1 Spatial Approach: Adapt contention region

32. 32 Larger Occupied Space  Fewer concurrent transmissions at higher rate

33. 33 Smaller Occupied Space  More concurrent transmissions at lower rate

34. 34  D perceives idle channel although A is transmitting A B C D distance Signal Strength CS Threshold Carrier Sense Multiple Access (CSMA)

35. 35 A B C D distance Signal Strength CS Threshold How Carrier-Sensing Controls Occupied Space E F

36. 36  Larger CS threshold by other stations leads to smaller occupied space by station A’s transmissions A B C D distance Signal Strength CS Threshold How Carrier-Sensing Controls Occupied Space E F

37. 37 A B C D distance Signal Strength CS Threshold Transmission Rate Needs to Be Adjusted Suitably E F  Larger CS threshold leads to higher interference  Transmission rate depends on Signal-to-Interference-Noise Ratio  Lower rate

38. 38 Adaptation of Occupied Space  Occupied Space == Contention Region  Occupied space can be adapted by joint adaptations: Rate-CS threshold Power-CS threshold Power-rate Power-rate-CS threshold

39. 39 Analytical Motivation for Protocols Pre-MURI work [Yang05Infocom] Cellular Model + Carrier Sense SINR

40. 40 β = CS th / Rx th (dB) Normalized Aggregate Throughput Network Aggregate Throughput (curves for different network parameters) For fixed power, optimal needs joint rate and CS threshold adaptation

41. 41 Dynamic Spatial Backoff For fixed power, optimal needs joint rate and CS threshold adaptation Joint adaptation of other parameters can be justified similarly

42. 42 Dynamic Spatial Backoff Joint Rate and CS Threshold Adaptation  Adaptation as search CS[1]CS[k] Rate[2] Rate[k] CS[2]CS[k-1] Rate CS Threshold Rate[k-1] Rate[1] 2 dimensional space

43. 43 Towards a Protocol: Reduce Search Space CS[1]CS[k] Rate[1] Rate[k] CS[2]CS[k-1]  Reduce search space using a lower bound on suitable CS threshold for a given rate Rate CS Threshold

44. 44 Towards a Protocol: Dynamic Search Using Transmission Success/Failure History Rate CS Threshold CS[1]CS[2]CS[3]CS[4] rate[1] rate[2] rate[3] rate[4] V V V >>> Rate CS Threshold CS[1]CS[2]CS[3]CS[4] rate[1] rate[2] rate[3] rate[4] V V V >>> Success Failure

45. 45 Towards a Protocol: Other Components  Determining success or failure using current parameters  Using history to guide search Successful combination of parameters cached for future use

46. 46 Towards a Protocol  We have proposed a dynamic spatial backoff protocol that adapts rate and CS threshold  Similar mechanisms can be used for other joint adaptations

47. 47 Performance of Dynamic Spatial Backoff (Random Topology: 40 nodes) β = CS th / Rx th (dB) Aggregate Throughput (Mbps) 101% of static optimal

48. 48 Performance of Dynamic Spatial Backoff (Random Topology: 16 nodes) β = CS th / Rx th (dB) Aggregate Throughput (Mbps) 92% of static optimal

49. 49 Conclusion: Dynamic Spatial Backoff  Significant potential to optimize performance using distributed mechanisms  Challenges remain: Accurately determining success versus failure Fully distributed mechanisms can be sub-optimal Interactions with higher layers Integration with temporal contention resolution

50. 50 Thanks! www.crhc.uiuc.edu/wireless

51. 51 Thanks! www.crhc.uiuc.edu/wireless

52. 52 Random (c,f) Channel Assignment

53. 53 Channel Transition Strategy (1, 2, 3) (4, 5, 6) Adjacent (6, 3) assignment Preferred channels : 2,3,4,5 (3, 4, 5) (4, 5, 6) (2, 3, 4) (1, 2, 3) 2 2 3 4 5 5 (4, 5, 6) (2, 3, 4) ( 3, 4, 5) Use randomly chosen preferred channel available at source (channel 2) Start transitions to get to a preferred channel at destination (channel 5)