1. Mesh Capacity and Multi-channel Meshes Y. Richard Yang 2/10/2009

2. Admin. r I am out of town on Thursday and Firday m Thursday class: by Richard Alimi on superposition coding m I will make up missed office hours on Thursday and Friday later in the semester 2

3. 3 Recap: Mesh Networks infrastructure mode ad-hoc (mesh) mode AP wired network AP: Access Point

4. 4 Recap: Two Constraints  transmission successful if there are no other transmitters within a distance (1+  )r of the receiver Interference constraint r a single half-duplex transceiver at each node Radio interface constraint

5. 5 Capacity Bound Note: Let L be the average (direct-line) distance for all T end-to-end bits.

6. 6 Recap: Random Networks r Uniform distribution of n nodes r n origin-destination (OD) pairs r Each node chooses same power level P, and thus equal radius r(n) r Equal throughput (n) bits/sec for all OD pairs

7. 7 Recap: Random Networks Supply/Demand r Required bit transmissions per second: r What is the maximum number of transmissions (of bits) in one second? m space used per transmission (interference limited): at least ¼  r(n)/2] 2 =  2 r 2 (n)/16 m number of simultaneous transmissions at most (interference limited): m total bits per second

8. 8 Random Networks: Capacity Required ≤ offered  

9. 9 Connectivity Constraint r Need routes between origin-destination pairs - places a lower bound on transmit range r(n) Not connected Connected A D A D To maintain connectivity with a high probability, requires r(n) on the order:

10. 10 Random Networks: Capacity Required ≤ offered   

11. 11 Measurement r Measured scaling law: throughput declines worse with n than theoretically predicted: 1/n 1.68 r Remaining story line m mesh networks with wide-area traffic may have low scalability, and need techniques to increase capacity

12. 12 Mesh Network Capacity: Arbitrary  transmission successful if there are no other transmitters within a distance (1+  )r of the receiver Interference constraint r a single half-duplex transceiver at each node Radio interface constraint rate*distance capacity:

13. 13 Improving Capacity: Approximate Ideal Model r Transmission power control r MAC scheduling r Routing r …

14. 14 Approximate Ideal Control: Power/Carrier-Sense Control ABCD A BCD Transmit Spatial Power Rate reuse High High Low Low Low High

15. 15 Improving Capacity: Change Traffic Pattern r To make communications local m node placement: change the demand patterns (thus L) e.g. base stations/access points with high-speed backhaul m use mobility F E A BC D BS1BS2 S T infrastructure

16. 16 Improving Capacity: Reduce Radio Interface Constraint r Multiple radio interfaces/codes 1 m

17. 17 Improving Capacity: Reduce Interference Constraint r Antenna design: steered/switched directional antennas r Non-interfering channels A D C B A B D C

18. Outline r Admin r Capacity r Improving capacity using all available spectrum 18

19. 19 Using All Available Spectrum r Goal: Using right channels at right nodes at right time may optimize network throughput r Problem domains m Wireless LANs APs determine the channel Clients share the same channel as their associated Aps m Mesh networks Nodes use multiple channels to increase spatial reuse 1 m 1 c 1 c

20. 20 Distributed Sharing of Unlicensed Spectrum r Federal Communications Committee (FCC) is increasing unlicensed spectrum allocation: m Industry, Science, and Medicine (ISM) m Unlicensed Personal Communication Service (UPCS): 1910- 1930 MHz and 2390-2400 MHz (30 Mhz) m National Information Infrastructure (NII) Band: 350 Mhz m 59-64 Ghz Millermeter Wave band r Coordination in unlicensed spectrum is difficult

21. 21 One Proposal for UPCS Spectrum Etiquette r Upper bound on energy level [official] r A station must “Listen Before Talk” (LBT) [official] m be quite for a monitoring time M after the previous energy level stops r Penalty for using a channel [non-official] m if a station holds a channel for a duration H, the station cannot transmit for P(H) amount of time as a penalty Question: what property do you want P(H) to hold?

22. 22 How to Design P(H)? r Assume an arrived bit of a user is transmitted immediately if the user is having the channel; otherwise the bit has to wait until the next interval r What is the average delay MHP(H)H

23. 23 Discussion r Spectrum sharing is largely still an open field r There are several proposals and evaluations m for example, see http://dx.doi.org/10.1023/A:1019129906297 r A potential term project topic

24. Outline r Admin r Capacity r Improving capacity using all available spectrum r Improving capacity using multiple channels in 802.11 mesh 24

25. 25 Multi-Channel 802.11 Mesh r Wireless LANs m APs determine the channel m Clients share the same channel as their associated APs r 802.11 mesh networks m Each node can choose operating 802.11 channels to increase spatial reuse 1 1 2 2

26. 26 Operating Channels for 802.11b 2400 [MHz] 24122483.524422472 channel 1channel 7channel 13 Europe (ETSI) US (FCC)/Canada (IC) 2400 [MHz] 24122483.524372462 channel 1channel 6channel 11 22 MHz

27. 27 Operating channels for 802.11a / US U-NII 5150 [MHz] 5180 5350 5200 3644 16.6 MHz center frequency = 5000 + 5*channel number [MHz] channel 404852566064 149153157161 522052405260528053005320 5725 [MHz] 5745 5825 5765 16.6 MHz channel 57855805

28. 28 Multi-Channel 802.11 Mesh: Goal Goal: Using right channels at right nodes at right time to improve network throughput 1 m 1 c 1 c 1 1 2 2

29. 29 Multi-Channel Network Capacity Channels Network Capacity c: number of channels

30. 30 Setup r A radio can use multiple orthogonal channels (e.g., 12 channels in 802.11a) 26 MHz100 MHz200 MHz100 MHz 2.45 GHz 915 MHz 5.25 GHz 5.8 GHz 3 channels8 channels4 channels 1 1 c Key question: how to assign channels to interfaces?

31. 31 Interface Assignment Strategies r Static assignment m an interface is assigned a fixed channel r Dynamic assignment m interface assignment changes with time r Hybrid interface assignment m some interfaces use static assignment, others use dynamic assignment  To focus on the key issue: assume a single interface 1 1 c

32. 32 Key Challenge r Connectivity vs using multiple channels Multiple channels not used Network is disconnected A 11 1 1 1 BC D A 1 2 BC D Additional constraints: intermediate relay nodes need to share the same channel as the upstream and downstream node

33. 33 SSCH: Slotted Seeded Channel Hopping – Overview r A dynamic assignment algorithm m divides the time into equal sized slots (e.g. 10 ms) and switches each radio across multiple orthogonal channels on the boundary of slots in a distributed manner r Main aspect of SSCH m channel scheduling self-computation of tentative schedule communication of schedules synchronization with other nodes

34. 34 SSCH – Desired Properties r No Logical Partition: Ensure all nodes come into contact occasionally so that they can communicate their tentative schedule r Synchronization: Allow nodes that need to communicate to synchronize r De-synchronization: Infrequently overlap between nodes with no communication A 1 2 BC D

35. 35 Channel Scheduling -Self-Computation r Each node uses (channel, seed) pairs to represent its tentative schedule for the next slot r Seed: [1, number of channels -1] Initialized randomly r Focus on the simple case of using one pair r Update rule: new channel = (old channel + seed) mod (number of channels) 10210210 A: Seed = 2 01201201 B: Seed = 1 Example: 3 channels, 2 seeds

36. 36 Channel Scheduling – Logical Partition 12012012 A: Seed = 1 01201201 B: Seed = 1 r Are nodes guaranteed to overlap? m same init channel, same seed (always overlap) m same init channel, different seeds (overlap occasionally) m different init channels, different seeds (overlap occasionally) m special case: Nodes may never overlap if they have the same seed but different channels

37. 37 Channel Scheduling – Solution to Logical Partition r Parity slot m every (number of channels) slots, add a parity slot m in parity slot, the channel number is the seed A: Seed = 1 B: Seed = 1 1201201211 0120120111 Parity Slot

38. 38 Channel Scheduling - Communication of Schedules r Each node broadcasts its tentative schedule (represented by the pair) once per slot

39. 39 Channel Scheduling - Synchronization r If node B needs to send data to node A, it adjusts its (channel, seed) pair to be the same as A. A B 1201201211 0211201221 111111111 Seed 222111112 Flow starts Sync starts upon the parity slot

40. 40 Channel Scheduling – Channel Congestion r It is likely various nodes will converge to the same (channel, seed) pair and communicate infrequently after that. (1,2)

41. 41 Channel Scheduling – Solution to Channel Congestion r De-synchronization r To identify channel congestion: compare the number of the synchronized nodes and the number of the nodes sending data. De- synchronize when the ratio >= 2 r To de-synchronize, simply choose a new (channel, seed) pair for each synchronized and non-sending nodes

42. 42 Channel Scheduling – Synchronizing with Multiple Nodes r Examples m a sender with multiple receivers m a forwarding node in a multi-hop network r Solution: Use multiple seeds per node m use one seed to synchronize with one node m add a parity slot every cycle ( = number of channels * number of seeds); the channel number of the parity slot is the first seed. 221011022100 Green slots are generated by seed 1 Yellow slots are generated by seed 2 1

43. 43 Channel Scheduling – Partial Synchronization 2210110221001 2012110201021 A B Seed 1211211222112 2122222222222 Flow starts Partial Sync Sync the second seed only

44. 44 Evaluations of SSCH r Simulate in QualNet r 802.11a, 54Mbps, (used) 13 orthogonal channels r Slot switch time = 80 µs r 4 seeds per node, slot duration = 10 ms r UDP flows: CBR flows of 512 bytes sent every 50 µs (enough to saturate the channel)

45. 45 Evaluation – Throughput (UDP)

46. 46 Evaluation – Multi-hop Mobile Networks

47. Backup Slides 47

48. 48 Preview: Network Layer r Transport packet from source to dest m network layer protocol in every host, router Basic functions: r Control plane: location management, path determination m locates hosts; determines route taken by packets from source to dest., and quality of service packets receive r Data plane: forwarding m move packets from router’s input to appropriate router output; use the setup from the control plane

49. 49 Packet Scheduling – Main Idea r Send packets to receivers in the same channel and delay sending packets to receivers in other channels

50. 50 Packet Scheduling – Basic Scheme r Within a slot, a node transmits packets in a round robin fashion among all flows r For a single flow, the packet is transmitted in FIFO order r Failed transmission causes the relevant flow to be inactive for half a slot. An inactive flow does not participate in transmission unless there are no active flows.

51. 51 Packet Scheduling – Absent Destination r Problem: The destinations are in other channel r Solution: Retransmission m broadcast: 6 transmission m unicast: Until successful or the cycle ends r Question: Can SSCH distinguish m destinations in other channels? m failure because of bad channel condition or node crash m collision

52. 52 Types of Conflicts in WLAN D <= R APs within communication range of each other

53. 53 Types of Conflicts in WLAN R < D <= 2R APs within interference range of each other

54. 54 Types of Conflicts in WLAN 2R < D <= R BSS

55. 55 Types of Conflicts in WLAN R BSS < D <= R BSS +R

56. 56 Channel Assignment Algorithm r Objective: maximize # conflict free clients m Client c is conflict-free if at least one AP in the range set is assigned a color j and no other AP in the range set is assigned color j r Random compaction Initialize AP assignment X’= random-permute(X) While true do ncf = num_conflict_free(T,Θ) for each AP i in X’ Θ(xi) = compaction_step(T, Θ, xi, k) end if num_conflict_free(T, Θ) == ncf then stop end

57. 57 Channel Assignment Algorithm r Objective: maximize # conflict free clients m Client c is conflict-free if at least one AP in the range set is assigned a color j and no other AP in the range set is assigned color j r Random compaction Initialize AP assignment X’= random-permute(X) While true do ncf = num_conflict_free(T,Θ) for each AP i in X’ Θ(xi) = compaction_step(T, Θ, xi, k) end if num_conflict_free(T, Θ) == ncf then stop end r Any drawback?

58. 58 Protocol Separation Implementation Routing and/or Interface assignment Interface switching and buffering IP stack Interface User Space Kernel Space