Auto mac rateless wireless concurrent medium access

AutoMAC(Rateless Wireless Concurrent Medium Access)

Aditya Gudipati, Stephanie Pereira, Sachin KattiStandford University

Wireless Network Presentation

Presented on :

Mobile Computing and Networking 2012 (MobiCom'12), August 22-26, 2012, Istanbul,Turkey.

Current Wireless Design

• Current Design : one Tx, one frequency, one at time (uplink and downlink) >> Avoid the interference

• RTS/CTS/Backoff Mechanism (protocols overhead)>> Client need to sense the carrier, to avoid collision

Limiting network throughput

Proposed Wireless Design

• Opposite approach : Exploit the interference

• Multiple clients concurently Tx to AP (uplink)

• AP concurently Tx to multiple clients (downlink)

• Decode all constituent packetsfrom interference

Increasing network throughput

-y

h2x2 + N

h2x2

Exploiting Interference (Uplink)

A

B

AP

x1

x2

y = h1x1+ h2x2 + N SIC(Successive Interference

Cancelation)

h1x1

Exploiting Interference (Uplink)• No Interference Troughput

• Throughput for Client A

• Throughput for Client B

• Throughput system

Rsic > Rnint

gained extra throughput, without hurting ongoing Tx

Exploiting Interference (Downlink)

y = h1x1 /√2 + h2x2 /√2 + N

A

B

AP

y = h1x1 /√2 + h2x2 /√2 + N

(x1 + x2) /√2

SNR Low(Weaker Node)

SNR High(Stronger Node)

SIC(Successive Interference

Cancelation)

h1x1 /√2

-y

h2x2/√2+ N

h2x2/√2

h1x1 /√2

Exploiting Interference (Downlink)

• number Tx required by stronger client < weaker client to decode

• dividing Tx power, instead of increasing• higher throughput when SNR different• throughput at the equal SNR ∼ no interference

schemes

Exploiting Interference (Downlink)

Throughput A (Alice) and B (Bob) can be achieved by exploiting interference compare to interference avoidance scheme

AutoMAC

• Sender didn't know actual SINR at receiver for decoding ->Rateless Property>> Rateless Codes (send rateless Tx, until receiver can decode)

• Wastefull Tx because packet can be decoded less than rateless encoding requirements >> Speculative ACKing (number rateless Tx depends on the SINR value)

AutoMAC : MAC Design

• Short Contention advertisement• announce the number of users send

concurrently Tx in the next slot• not sent, if AP want to transmit or concurrent

Tx in previous slot continue

• Frequency domain backoff technique• client pickup random subcarrier

AutoMAC : Exploit Uplink Interference

• Packet Encoding• Orthogonal choice of rateless code (Strider or

Spinal codes)• Decoding Constituent Packet with SIC• Per-frame channel estimation• Decoding Packet (Strider)• Substracting Node'1 collission contribution

AutoMAC : Speculative ACKing

• Diferrent encoding time -> encoding order and relative SNR between AP-Client

• Number of Tx is less for last decoded packet than the first packet -> better SNR

• "Key" of ACKing technique : • Estimate number of Tx for encoding packet ->

estimate effective SNR value• SINR vs number Tx graph -> after M Tx, receiver

start ACK packet p

AutoMAC : Exploiting Broadcast on Downlink

• same encoding technique for all client– rateless encoding (strider's) -> rateless frame– n frames add in complex domain -> one frame– prepend header, pass frame to OFDM PHY for Tx

• decoding the own packet at last– every client assumed best channel– ensure SINR to decode and send the ACK -> AP

receive know the received signal strength– apply the SIC in the order Tx from AP

Evaluation

• outperform by 35% (include downlink)• decode packet in interference upto 3 packets• eleminates hidden terminal• multiplex upto 3 packets (downlink)• accurately estimates channel, frequency, and

sampling offsets• 60% throughput gain over 802.11 MAC (uplink)

and 50% over an 802.11 style MAC

Evaluation

The Gains will increase as relative with SNR increase, and AutoMAC gain 50 - 60 % over Conventional or Time Fair Wifi.

Evaluation

Time Spent and throughput fairness comparison

Gain Analysis

• Collision minimum, because no lack of coordination and inability to decode -> Centralized MAC

• Lower contention time -> frequency domain backoff technique

• No overheads time for ACK waiting (SIFS,DIFS) -> ACK one for every batch

• fraction time spent is higher 20% -> efficient MAC protocol

Conslusion and References