Packet Mixing: Superposition Coding and Network Coding May 21, 2007 Packet Mixing: Superposition Coding and Network Coding Richard Alimi CS434 Lecture Joint work with: L. Erran Li, Ramachandran Ramjee, Harish Viswanathan, Y. Richard Yang 2/12/2009 Yale University

Wireless Mesh Networks May 21, 2007 Wireless Mesh Networks City- and community-wide mesh networks widely used New approach to the “last mile” of Internet service In United States alone [muniwireless.com, Jan 16, 2007]: 188 deployed 148 in-progress or planned Yale University

Mesh Network Structure May 21, 2007 Mesh Network Structure APs deployed, some connect directly to Internet Street lamps, traffic lights, public buildings Clients associate with nearest AP Traffic routed to/from Internet via APs (possibly multi-hop) Yale University

Limited Capacity of Mesh Networks May 21, 2007 Limited Capacity of Mesh Networks Current mesh networks have limited capacity [Li et al. 2001, dailywireless.org 2004] Increased usage will only worsen congestion More devices Larger downloads, P2P, video streaming Limited spectrum Network-wide transport capacity does not scale [Gupta and Kumar 2001] Must bypass traditional constraints Yale University

Packet Mixing for Increased Capacity May 21, 2007 Packet Mixing for Increased Capacity Multiple packets transmitted simultaneously Same timeslot Cross-layer coding techniques No spreading (unlike CDMA) Receiver(s) decode own packets Possibly use side-information (e.g., packets previously overheard) Yale University

Packet Mixing Objective May 21, 2007 Packet Mixing Objective Objective Construct a mixed packet with Maximum effective throughput Sufficiently-high decoding probability at receivers Mixture of coding techniques Currently consider two techniques Downlink superposition coding XOR-style network coding Yale University

Recap: Physical Layer Signal Modulation May 21, 2007 Recap: Physical Layer Signal Modulation Signal has two components: I and Q Represented on complex plane Sender Map bits to symbol (constellation point) Receiver Determine closest symbol and emit bits 00 01 11 10 QPSK Yale University

Downlink Superposition Coding May 21, 2007 Downlink Superposition Coding Basic idea Different message queued for each receiver Transmit messages simultaneously Exploit client channel diversity Example Yale University

Downlink Superposition Coding May 21, 2007 Downlink Superposition Coding Basic idea Different message queued for each receiver Transmit messages simultaneously Exploit client channel diversity Example “01” Weaker receiver Layer 1 “Low resolution” Stronger receiver Layer 2 “High resolution” “11” Yale University

Downlink Superposition Coding May 21, 2007 Downlink Superposition Coding Basic idea Different message queued for each receiver Transmit messages simultaneously Exploit client channel diversity Example “01” Weaker receiver Layer 1 “Low resolution” Stronger receiver Layer 2 “High resolution” “11” Yale University

SC Decoding: Successive Interference Cancellation May 21, 2007 SC Decoding: Successive Interference Cancellation Weaker Receiver “01” “11” Decode Layer 1 Yale University

SC Decoding: Successive Interference Cancellation May 21, 2007 SC Decoding: Successive Interference Cancellation Stronger Receiver “01” “11” (1) Decode Layer 1 Yale University

SC Decoding: Successive Interference Cancellation May 21, 2007 SC Decoding: Successive Interference Cancellation Stronger Receiver “01” “11” (1) Decode Layer 1 (2) Subtract and decode Layer 2 Yale University

May 21, 2007 SC Quantization Gains Discrete rates are common in standards (including 802.11) In other words... Channel qualities are quantized distance 36 Mbps 24 Mbps Yale University

SC Quantization Gains Idea May 21, 2007 SC Quantization Gains Idea Steal extra power from one receiver without affecting data rate Allocate extra power to second layer destined for a stronger receiver Use largest rate achievable with this extra power Can derive formula for computing these gains: where: Yale University

SC Quantization Gains in 802.11a/g May 21, 2007 SC Quantization Gains in 802.11a/g Distances R1: varies R2: 40 meters Path-loss Exponent: 4 Noise: -90 dBm Remaining parameters consistent with Cisco Aironet 802.11g card Yale University

SC Scalability Analysis May 21, 2007 SC Scalability Analysis Recall capacity analysis for arbitrary network: Radio Interface constraint Interference constraint First we'll show an upper bound: Yale University

SC Scalability Analysis May 21, 2007 SC Scalability Analysis 1 1 2 2 3 3 4 4 5 5 bit time t Without Superposition Coding With Superposition Coding Yale University

SC Scalability Analysis May 21, 2007 SC Scalability Analysis 1 1 2 2 3 3 4 4 5 5 bit time t We now have up to n-1 transmissions per bit-time! Without Superposition Coding With Superposition Coding Yale University

SC Scalability Analysis May 21, 2007 SC Scalability Analysis Up to n-1 concurrent transmissions, so ... Without Superposition Coding With Superposition Coding Yale University

SC Scalability Analysis May 21, 2007 SC Scalability Analysis Up to n-1 concurrent transmissions, so ... Total area due to interfering transmissions reduces by factor of n Without Superposition Coding With Superposition Coding Yale University

SC Scalability Analysis May 21, 2007 SC Scalability Analysis Changes to capacity analysis At most n-1 concurrent transmissions (superposition coding with n-1 layers) # of interfering transmissions reduced by a factor of n Modified constraints produce O(n) upper bound Is the upper bound achievable? Yes Yale University

XOR-style Network Coding May 21, 2007 XOR-style Network Coding Basic idea Nodes remember overheard and sent messages Transmit bitwise XOR of packets: Receivers decode if they already know n-1 packets Example 1 2 n 2 1 R1 1 2 R2 Yale University

Putting it Together: Packet Mixing May 21, 2007 Putting it Together: Packet Mixing 4 Flows Packet d has dest Rd Without packet mixing 8 transmissions required With packet mixing 5 transmissions required 2 4 R4 R3 R1 R2 3 1 Routing link Overhearing link Yale University

Putting it Together: Packet Mixing May 21, 2007 Putting it Together: Packet Mixing R2 sends Pkt 1 to AP 1 2 4 R4 R3 R1 1 R2 3 1 Routing link Overhearing link Yale University

Putting it Together: Packet Mixing May 21, 2007 Putting it Together: Packet Mixing R2 sends Pkt 1 to AP R4 sends Pkt 2 to AP 2 1 4 R4 R3 R1 1 2 R2 3 2 1 Routing link Overhearing link Yale University

Putting it Together: Packet Mixing May 21, 2007 Putting it Together: Packet Mixing 3 R2 sends Pkt 1 to AP R4 sends Pkt 2 to AP R1 sends Pkt 3 to AP 2 1 4 R4 R3 R1 1 2 R2 3 2 1 3 Routing link Overhearing link Yale University

Putting it Together: Packet Mixing May 21, 2007 Putting it Together: Packet Mixing 3 4 R2 sends Pkt 1 to AP R4 sends Pkt 2 to AP R1 sends Pkt 3 to AP R3 sends Pkt 4 to AP 2 1 R4 R3 R1 1 2 R2 3 4 2 1 3 4 Routing link Overhearing link Yale University

Putting it Together: Packet Mixing May 21, 2007 Putting it Together: Packet Mixing 3 4 R2 sends Pkt 1 to AP R4 sends Pkt 2 to AP R1 sends Pkt 3 to AP R3 sends Pkt 4 to AP AP sends mixed packet using SC: Layer 1: Layer 2: 2 1 4 3 R4 R3 R1 R2 1 2 1 2 2 1 3 4 3 4 Routing link Overhearing link Yale University

Scheduling under Packet Mixing May 21, 2007 Scheduling under Packet Mixing Per-neighbor FIFO packet queues Qd is queue for neighbor d – first denoted by head(Qd) Total order on packets in all queues Ordered by arrival time – first denoted by head(Q) Rule: always transmit head(Q) Prevents starvation d2 d3 d1 head(Q) head(Q3) Yale University

Superposition Coding Scheduling May 21, 2007 Superposition Coding Scheduling Overview Layer 1: Select head(Q) with dest d1 at rate r1 Layer 2: Select floor(r2 / r1) packets for dest d2 ≠ d1 at rate r2 Allows different rates for each layer Selects best rates given current channels Ensures sufficient decoding probabilities head(Q2) 3 packets from Q4 Layer 1 Layer 2 Destination 2, rate 12 Mbps Rate : 12 Mbps Packets: 4 Throughput: 48 Mbps Destination 4, rate 36 Mbps Yale University

Superposition and Network Coding Scheduling May 21, 2007 Superposition and Network Coding Scheduling Algorithm Iterate over discrete rates for each layer, r1 and r2 Layer 1: Select network-coded packet, N1 packets encoded Layer 2: Selects floor(r2 / r1) network-coded packets, N2 packets encoded Only consider neighbors that support r2 in second layer Effective throughput is r1 • (N1 + N2 ) head(Q1) head(Q2) head(Q3) head(Q4) , head(Q5) head(Q6) , head(Q7) head(Q8) Layer 1 Layer 2 Yale University

Evaluations: Setup Algorithms implemented in ns-2 version 2.31 May 21, 2007 Evaluations: Setup Algorithms implemented in ns-2 version 2.31 Careful attention to physical layer model Standard ns-2 physical layer model does not suffice Use packet error rate curves from actual 802.11a measurements [Doo et al. 2004] Packet error rates used for physical layer decoding and rate calculations Realistic simulation parameters Parameters produce similar transmission ranges as Cisco Aironet 802.11g card in outdoor environment Yale University

Evaluations: Network Demand May 21, 2007 Evaluations: Network Demand Setup 1 AP 10 clients 8 flows Vary client sending rate Packet mixing gains are sensitive to network demand Queues are usually empty with low demand Few mixing opportunites Network Coding shows ~3% gain with TCP [Katti et al. 2006] Yale University

Evaluations: Internet → Client Flows May 21, 2007 Evaluations: Internet → Client Flows Setup 1 AP 20 clients 16 flows Backlogged flows Vary % of flows originating at AP Superposition Coding superior when Internet → client flows are common Throughput gains as high as 4.24 Yale University

GNU Radio Implementation May 21, 2007 GNU Radio Implementation Open Source software radio RF frontend hardware (USRP) Signal processing in software Components Implementation of Superposition Coding in GNU Radio environment 802.11 MAC implemented with Network Coding support Measurement Results Yale University

May 21, 2007 Backup Slides Yale University

Example Coding Techniques May 21, 2007 Example Coding Techniques Transmitter-side Downlink superposition coding [Cover 1972, Bergmans and Cover 1974] XOR-style network coding [Katti et al. 2006] Receiver-side Uplink superposition coding Analog [Katti et al. 2007] and physical-layer [Zhang et al. 2006] network coding Yale University

Multirate NC: mnetcode May 21, 2007 Multirate NC: mnetcode SC requires multirate for better gains, so extend NC as well Algorithm Run single-rate COPE algorithm snetcode(r) for each rate r and select best Skip rate if not supported by neighbor Only consider head(Qd) for each neighbor d N packets XOR'd at rate r Effective throughput is N • r Yale University

Simple Cross-layer Mixing May 21, 2007 Simple Cross-layer Mixing Utilize physical- and network-layer coding Algorithm SC Layer 1: Select NC packet with mnetcode Must include head(Q) SC Layer 2: Select packet with Gopp Problems No NC used in Layer 2 packets Limited rate combinations head(Q1) head(Q2) 3 packets from Q4 Layer 1 Layer 2 Yale University

Evaluations: Client → Client Flows May 21, 2007 Evaluations: Client → Client Flows Setup 1 AP 20 clients Backlogged flows Vary # of flows Both SC and NC mixing alone improve with # of flows More opportunities Gains each SC and NC exploited successfully by SC1 and SCJ schedulers Yale University