Mythili Vutukuru

1
Cross-layer Wireless Bit Rate Adaptation

• Mythili Vutukuru
• MIT CSAIL
• Joint work with Hari Balakrishnan and Kyle
  Jamieson

2
Time-varying wireless channel

• Large-scale attenuation
• Due to changing distance
• Small-scale fading
• Due to multipath
• Interference
• Unpredictable

Bit errors, frame losses
3
Online Bit Rate Adaptation

• Varying modulation coding
  (redundancy)
• Better channel ? Higher rate
• Huge gains possible (802.11g has 1-54 Mbps)

Requirements
Estimate channel accurately
Responsive
Do not react to interference
4
Existing Algorithms
SNR-based
Frame-based

• RRAA, Wong et. al., 2006.
• SampleRate, Bicket, 2005.

• RBAR, Holland et. al., 2001.
• CHARM, Judd et. al., 2008.

Data
Data
ACK
SNR using preamble
Estimate frame loss rate at each bit rate
Lookup table SNR ? best rate
5
Problems With Existing Algorithms
Indoor
Channel Quality
Outdoor
SNR-based
SNR ? bit rate table specific to environment
Which table to use when?
X
Frame-based
Reacts to interference
Too slow
6
SoftRate Key Insight
Channel Quality
SoftPHY Hints
Per-bit Confidences
Estimate BER
Detect Interference
SoftRate
Interference-free BER
SoftPHY design more general compared to JB07,
KKB08
7
SoftRate Contributions

• Adapts to channel accurately and quickly
• Robust to collision losses
• Feedback interference-free BER from SoftPHY
  hints
• 2X gain over frame-based and SNR-based

8
Rest of the talk

• Computing SoftPHY hints
• Interference-free BER from SoftPHY hints
• SoftRate algorithm
• Evaluation

9
Computing SoftPHY Hints
TX
Modulator
Encoder
Symbols
Bits
Coded bits
Soft Output Decoder
RX
Demodulator
Decoder
Symbols
Coded Bits
Bits
SoftPHY Hints
Error vectors
(rcvd symbol closest valid symbol)
10
BER from SoftPHY Hints
decoded bits
Soft Output Viterbi (or) BCJR decoder
For linear block or convolutional code
Log Likelihood Ratio
s
SoftPHY hint of a bit

p

Probability of bit error
BER Average p over all bits in the packet
11
Computing Interference-free BER
Compute Interference-free BER
Interference
12
SoftPHY Hints With Weak Signal
13
The SoftRate Protocol
Data
Receiver
Interference-free BER
BER
Interference Detection
Sender
Pick rate with max throughput
SoftPHY Hints
?
14
Rate Selection at the Sender
R . Frame Delivery Rate f(BER)
BER
Throughput
24 Mbps
18 Mbps
12 Mbps
Adjacent rates have an order of magnitude
difference in BER
15
When is the current rate optimal?
BER
Throughput
24 Mbps
Optimality range for 18 Mbps
18 Mbps
12 Mbps
16
When to lower the rate?
BER
Throughput
24 Mbps
18 Mbps
12 Mbps
17
When to increase the rate?
BER
Throughput
24 Mbps
18 Mbps
12 Mbps
18
The SoftRate Protocol
Data
Receiver
Interference-free BER
BER
Interference Detection
Sender

• Precompute optimality ranges
• If BER below optimality range,
• increase rate.
• If above range, decrease rate.
• Otherwise, continue at current rate.

SoftPHY Hints
19
Evaluation Method
TCP
ns-3 simulations
Rate Adaptation (SoftRate, SNR-based,
Frame-based)
SoftPHY Traces
PHY GNU Radio USRP
Experiments
Channel Simulator
20
Evaluation Questions

• SoftPHY
• Can SoftPHY hints estimate channel BER?
• Can SoftPHY hints identify interference?
• SoftRate
• Gains of SoftRate in mobile channels?
• SoftRate robust to interference?

80 of the time
Almost always
21
SoftPHY Hints Predict BER
22
SoftRate Evaluation in Mobile Channels

• Compare with
• StaticBest omniscientbest for each pkt
• SNR-based RBAR and CHARM
• Frame-based RRAA and SampleRate

TCP
AP
Clients
Wired LAN
Traces
23
Is SoftRate close to optimal? (Walking Speed)
Within 10 of the optimal
24
SoftRate vs. Frame-based Walking speed
Up to 2X over best frame-based algorithm
25
SoftRate vs. SNR-based Varying Mobility
(Walking speed)
1 10 100 1000
(Walking speed)
(Train speed)
Approx speed (mph)
4X over untrained SNR-based algorithm
26
SNR vs. BER Varying Mobility
SNR?BER and SNR?best rate specific to operating
environment
27
Evaluation Answers

• SoftPHY
• Can SoftPHY hints estimate channel BER?
• Can SoftPHY hints identify interference?
• SoftRate
• Gains of SoftRate in mobile channels?
• SoftRate robust to interference?

YES
80 of the time
2X - 4X
Almost always
28
SoftRate Summary

• Accurate, responsive, robust to collision losses
• Feedback interference-free BER from SoftPHY
  hints
• 2X over frame-based, 4X over untrained SNR-based
• Looking ahead
• BER computation from SoftPHY hints useful for
  other cross-layer protocols