Many-to-Many Beam Alignment in Millimeter Wave Networks

Transcription

1 Many-to-Many Beam Alignment in Millimeter Wave Networks Suraj Jog Jiaming Wang, Junfeng Guan, Thomas Moon, Haitham Hassanieh, Romit Roy Choudhury

2 Bandwidth requirement of wireless applications is growing

3 VR and AR Restricts Mobility

4 VR and AR

5 VR and AR

6 Robotic Automation and Collaboration Tasks

7 Millimeter Wave Technology Huge bandwidth available at millimeter wave frequencies Millimeter Wave Bands 0 GHz 10 GHz 20 GHz 30 GHz 40 GHz 50 GHz 60 GHz 70 GHz 80 GHz 90 GHz 100 GHz Currently we operate here >14 GHz of Unlicensed Spectrum Millimeter Wave can support data rates of multi-gbps

8 How to scale mmwave networks while maintaining multi-gbps throughput per user?

9 Today s Networks : Broadcast

10 mmwave changes how wireless systems operate mmwave: Narrow-beam Antennas

11 mmwave changes how wireless systems operate mmwave: Narrow-beam Antennas

12 mmwave changes how wireless systems operate mmwave: Narrow-beam Antennas

13

14

15 Past work focuses on quickly finding best alignment for a single communication link [NSDI 17, SIGCOMM 18, INFOCOM 15, SIGMETRICS 15]

16 Multi-Link Beam Alignment is challenging!

17 Client 1 Collision Carrier Sense does not work in directional networks Client 2 [MOBICOM 02, SIGCOMM 09] AP1 AP2

18 Collision Reflector We cannot align the beams of each AP and client independent of other APs and clients in the network

19 BounceNet A many-to-many beam alignment protocol that can enable many links to operate in parallel in confined spaces without interfering.

20 Many-to-Many Beam Alignment What is the best alignment of beams that densely packs as many links as possible?

21 Many-to-Many Beam Alignment What is the best alignment of beams that densely packs as many links as possible? Leverage sparsity in the mmwave channel!

22 Millimeter Wave Channels are sparse There are only a few number of paths between any TX and RX

23 Millimeter Wave Channels are sparse There are only a few number of paths between any TX and RX

24 Millimeter Wave Channels are sparse There are only a few number of paths between any TX and RX

25 Millimeter Wave Channels are sparse Physical Signal Routing

26 Collision Cannot Transmit at the same time

27 Can both Transmit at the same time

28 Physical Signal Routing enables more efficient beam alignment Can both Transmit at the same time

29 Many-to-Many Beam Alignment AP-Client Pair 1 Conflict Graphs AP-Client Pair 1 DP Direct Path Super Node DP IP Direct Path Indirect Path

30 Many-to-Many Beam Alignment AP-Client Pair 1 Conflict Graphs AP-Client Pair 1 IP Indirect Path DP Super Node DP IP Direct Path Indirect Path

31 Many-to-Many Beam Alignment Conflict Graphs AP-Client Pair 1 AP-Client Pair 2 IP IP2 IP1 DP DP IP IP2 IP1 DP DP AP-Client Pair 3 AP-Client Pair 4

32 Many-to-Many Beam Alignment Conflict Graphs AP-Client Pair 1 AP-Client Pair 2 IP IP2 IP1 DP DP IP IP2 IP1 DP Maximum Weighted Independent Set AP-Client Pair 3 AP-Client Pair 4 DP

33 Many-to-Many Beam Alignment Conflict Graphs AP-Client Pair 1 AP-Client Pair 2 IP IP2 IP1 DP DP IP IP2 IP1 DP Maximum Weighted Independent Set à NP-Hard AP-Client Pair 3 AP-Client Pair 4 DP

34 Many-to-Many Beam Alignment Key Idea Direct paths are highest data rate paths à Prioritize routing along direct path Decouple routing along direct and indirect paths

35 Many-to-Many Beam Alignment Conflict Graphs AP-Client Pair 1 AP-Client Pair 2 IP IP2 IP1 DP DP IP IP2 IP1 DP DP AP-Client Pair 3 AP-Client Pair 4

36 Many-to-Many Beam Alignment Conflict Graphs Direct Path Conflict Graphs AP-Client Pair 1 AP-Client Pair 2 DP DP DP Sparse Graph à Chordal with high probability DP AP-Client Pair 3 AP-Client Pair 4

37 Many-to-Many Beam Alignment Fairness among links Create Multiple Many-to-Many Alignments 1. Maximize number of nodes that transmit simultaneously 2. Ensure each client its fair share on highest data rate path

38 Many-to-Many Beam Alignment How to quickly learn the paths and interference patterns to adapt the alignment?

39 Learn paths between nodes AP Direct Path Client Indirect Path

40 Learn paths between nodes AP! 1 Direct Path ɸ 1 Client Indirect Path

41 Learn paths between nodes AP! 2 Direct Path ɸ 2 Client Indirect Path

42 Learn paths between nodes AP Direct Path Client Direct Path: Indirect Path: (! 1 (! 2,, ɸ 1 ) ɸ 2 ) Indirect Path

43 Learn paths between nodes AP Direct Path Client Direct Path: Indirect Path: (! 1 (! 2,, ɸ 1 ) ɸ 2 ) Indirect Path

44 Learn paths between nodes AP Direct Path Client Direct Path: Indirect Path: (! 1 (! 2,, ɸ 1 ) ɸ 2 ) Indirect Path

45 Learn paths between nodes AP 1 AP 2 Client 1 Client 2 AP 3 Client 3

46 Learn paths between nodes AP 1 AP 2 Client 1 Client 2 AP 3 Client 3

47 Learn paths between nodes AP 1 AP 2 AP 3 Client 1 Client 2 Client Overhead scales as O(N 2 ) for N links AP N Client N

48 Learn paths between nodes Ethernet Backbone AP 2 Client 2 AP 1 Client 1 Reflector

49 Ethernet Backbone Learn paths between nodes AP 2 AP 1 70 o Client 2 Client 1 AP1 can align towards me along 70 degree AP1 can align towards me along degree Reflector

50 Learn paths between nodes Ethernet Backbone AP 2 Client 2 AP 1 Client 1 Reflector

51 Learn paths between nodes Ethernet Backbone AP 2 Client 2 AP 1 Client 1 Reflector

52 Learn paths between nodes Ethernet Backbone AP 2 Client 2 AP 1 Client 1 Controller Require only O(N) scans to find all paths between every pair Reflector of nodes

53 Overall Flow Clients Access Points Controller

54 Overall Flow Clients Access Points Controller

55 Alignment 1 Clients Access Points Controller

56 Alignment 2 Clients Access Points Controller

57 Alignment 3 Clients Access Points Controller

58 Many-to-Many Beam Alignment 1. How to quickly learn the paths and interference patterns to adapt the alignment? 2. What is the best alignment of beams that densely packs as many links as possible?

59 BounceNet Evaluation Clock (a) 60 GHz Radios Clock (c) 12 o and 3 o Antennas I Q I Q (b) 24 GHz radios with phased arrays

60 BounceNet Evaluation

61 Evaluation Methodology Compare Schemes ad Standard 2. Baseline: Independent Alignment with Carrier Sense

62 Total Network Data Rate (Gbps) Total Network Data Rate (Gbps) Total Network Data Rate Phased Array BounceNet ad Baseline Number of Clients Number of Links

63 Total Network Data Rate (Gbps) Total Network Data Rate (Gbps) Total Network Data Rate Phased Array BounceNet ad Baseline Number of Clients Number of Links

64 Total Network Data Rate (Gbps) Total Network Data Rate (Gbps) Total Network Data Rate Phased Array BounceNet ad Baseline Number of Clients Number of Links

65 Total Network Data Rate (Gbps) Total Network Data Rate (Gbps) Total Network Data Rate Phased Array BounceNet ad Baseline Number of Clients Number of Links 3x Gain

66 Total Rate (Gbps) Total Network Data Rate (Gbps) Total Network Data Rate 3 Degree BounceNet ad Baseline Number of Clients Number of Links

67 Total Rate (Gbps) Total Network Data Rate (Gbps) Total Network Data Rate 3 Degree BounceNet ad Baseline Number of Clients Number of Links 6.6x Gain

68 Data Rate of Worst Case Client Baseline BounceNet Gain 3 Degree 1.7 Gbps 2.9 Gbps 1.7x 12 Degree 0.2 Gbps 2.6 Gbps 13.5x Phased Array 0.13 Gbps 0.98 Gbps 7.5x BounceNet scales network throughput and ensures fairness, outperforming compare schemes

69 To conclude 1. BounceNet enables many-to-many beam alignment by exploiting dense spatial reuse 2. mmwave opens up a new paradigm and requires rethinking of wireless network design