Understanding the Real-World Performance of Carrier Sense
MIT Computer Science and Artificial Intelligence LaboratoryNetworks and Mobile Systems
http://nms.csail.mit.edu
Kyle Jamieson, Bret Hull, Allen Miu, Hari Balakrishnan
Introduction
• Carrier sense is a crucial building block for many radio networks– Wireless sensor networks– Wireless local area
networks• Performance depends
on carrier sense
MAC layer
Physical layer
Application layer
Carrier sense
A research direction
• Let’s quantify how well carrier sense performs in real-world radio networks
• Let’s study diverse radio networks and draw high-level conclusions– Modulation type– Network size (number of nodes)– Data rates
Experimental setup
Experimental testbedSensor network 802.11b/g LAN
Nodes 60 3
Radio Chipcon CC1000 Atheros 5212
Data rate 38.4 Kbps 1 to 54 Mbps
Modulation FM narrowband OFDM/DSSS
MAC B-MAC (software) 802.11 (hardware)
Sensor network testbed• 60-node Mica2
sensor network• Six radio hops in
diameter• Ethernet
backchannel to log packet receptions
100 ft.
16,076 sq. ft.
http://mistlab.csail.mit.edu
Outline
• IntroductionImplementing carrier sense• Benefits of carrier sense• Drawbacks of carrier sense• Conclusion
How carrier sense works:energy detectionS
igna
l stre
ngth
(dB
m)
Time
Squelch (“noise floor”)Instantaneous signal strength
Energy detect clearEnergy detect busy
How carrier sense works: other mechanisms
• Preamble detection• Decorrelation amplitude
– Unique to spread-spectrum radios
• AGC unlock– True when AGC adjusts rapidly
Spreading code
×× Received data
Spreading code
Transmit data
PacketPreamble
Outline
• Introduction• Implementing carrier senseBenefits of carrier sense• Drawbacks of carrier sense• Conclusion
Aggregate load lowers link delivery rate
WSN experiment with all nodes sending, carrier sense on
~360 links > 70% at 4 pps
Carrier sense improves link delivery rates
Carrier sense avoids collisions under high load
Only 80 links in the network
are > 70% without CS
Carrier sense improves throughput
Large-scale experiment with an offered load of 1 pps/node
Outline
• Introduction• Implementing carrier sense• Benefits of carrier senseDrawbacks of carrier sense• Conclusion
Sender-side decision;receiver-side collision
R
S
Will any transmissions
collide with mine?
Carrier sense is at best a heuristic for predicting transmissions’ success
Exposed terminals fool carrier sense
R S S΄ R΄
Carrier sense indicates busy, yet the transmission would have succeeded (S, S’ are exposed terminals)
Missed transmission opportunity
Carrier sense misses transmit opportunities
Large-scale experiment with CS energy detect, 0.25 pps per node
Carrier sense misses transmit opportunities
Large-scale experiment with carrier sense off, 0.25 pps per node
Capture fools carrier sense
R captures B’s transmission despite A’s concurrent transmission
R
A
B
Missed transmission opportunity
Capture prevalent at low bit rates
At some low 802.11 bit rates, node B should disable carrier sense
Collision
Capture
Hidden terminals fool carrier sense
R S’S
Carrier sense is
free!
Carrier sense indicates free, yet both transmissions fail (S, S’ are hidden terminals)
Related work
• Capture-aware MAC– Whitehouse et al., Em-Nets ’05– Priyantha, PhD thesis ‘05
• Channel sampling to infer congestion– CODA, Wan et al., SenSys ’04
• Models to pick carrier sense sensitivity– Yang and Vaidya, INFOCOM ’05
Conclusion and future research
• An experimental evaluation of the benefits and drawbacks of carrier sense
• Algorithm to track correlation between signal strengths and packet reception
• Use a congestion control algorithm: CODA or Fusion [SenSys] and turn off or reduce carrier sense