1. VWID: Variable-Width Channels for Interference Avoidance Brad Karp UCL Computer Science CS M038 / GZ06 26 th January, 2009

2. Context: Sharing of Spectrum 2 Finite RF spectrum available for use by nodes in a wireless network MACAW/802.11 approach to sharing of spectrum: –Each node uses entire channel (full width, in Hz) for each packet transmission –Try to schedule senders that interfere so that they don’t send concurrently Interference plagues 802.11-style sharing in mesh networks –MAC can’t perfectly serialize interfering senders –Result: interference reduces throughput (evidence: Roofnet’s ETT overpredicts throughput…) Alternatives to serializing transmissions by mutually interfering nodes? Alternatives to devoting entire channel to each node’s transmissions?

3. Another Way: Orthogonal Channels 802.11a allows use of different channel widths: –20 MHz (default): 54 Mbps nominal –10 MHz: 27 Mbps nominal –5 MHz: 13.5 Mbps nominal Idea: assign non-overlapping (orthogonal) channels to mutually interfering links –In principle, should prevent interference –Under certain assumptions, increases total capacity vs. single-channel CSMA! 3

4. Modeling Capacity: Assumptions Consider 2-client 802.11 network with one base station, all traffic from clients to base station Base station has one radio, one antenna Clients each have one radio, one antenna All nodes in mutual range Both clients send continuously Client 1 received at BS with power P 1, client 2 received at BS with power P 2 4

5. Understanding Two-Node Capacity 5 R1R1 R2R2 (bits/s/Hz) R 1 < l og 2 ( 1 + P 1 N ) b i t s / s / H z ; R 2 < l og 2 ( 1 + P 2 N ) b i t s / s / H z ; l og 2 ( 1 + P 1 N ) l og 2 ( 1 + P 2 N ) Optimum sum-capacity Transmitter 1’s Rate R 1 + R 2 < l og 2 ( 1 + P 1 + P 2 N ) b i t s / s / H z : [Ramki Gummadi] Transmitter 2’s Rate Sum of Rates

6. 6 VWID throughput R1R1 R2R2 (bits/s/Hz) A B Optimum throughput at α = P 1 P 1 + P 2 l og 2 ( 1 + P 2 N ) l og 2 ( 1 + P 1 N ) R 1 < α l og 2 ( 1 + P 1 α N ) b i t s / s / H z ; R 2 < ( 1 - α ) l og 2 ( 1 + P 2 ( 1 - α ) N ) b i t s / s / H z : l og 2 ( 1 + P 2 N ) l og 2 ( 1 + P 1 N ) α = 0 α = 1 [Ramki Gummadi]

7. Finding Optimal Channel Widths Want to maximize sum of two rates: R = R1 + R2 = Setting gives maximum: i.e., to maximize total throughput, assign each node channel width proportional to its share of total power received at AP 7

8. Example: CSMA vs. Orthogonal Channels Two clients, each of which is received by base station with SNR of 1 Under CSMA, one client alone achieves throughput: –so when alternating, each gets 0.5 bits/s/Hz If we assign half of channel to each and allow concurrent transmissions, each gets: 8 0.79 bits/s/Hz

9. VWID Prototype Automated system for channel assignment –For each link, assign sender one of {5, 10, 20} MHz channel Chooses assignment of channels that maximizes aggregate throughput across all links Additional constraint: don’t decrease a sender’s channel width if doing so reduces that link’s throughput (vs. 20 MHz channel width) 9 Exhaustive search: worst case cost is exponential in number of interfering links!

10. VWID Experimental Evaluation Outdoor 802.11a testbed: –6 nodes; 10 links, 8 of which 1-2 km long Bit-rate for each node fixed; chosen so that node gets reasonable throughput on its links Results given only for UDP traffic; all nodes send as fast as they can Experiments have carrier sense enabled because “gives higher throughput” (!?) 10

11. Link Throughput Improvement: Point-to-Point Links 11 no VWID with VWID [Ramki Gummadi]