1
Link-level Measurements from an 802.11b Mesh
Network

• Daniel Aguayo (MIT), John Bicket (MIT), Sanjit
  Biswas (MIT), Glenn Judd (CMU), Robert Morris
  (MIT)

Summary presented by Gary Woo
2
Outline

• Assumptions in neighbor abstraction
• What is Roofnet?
• Experiment/Results
• Hypotheses
• Cause
• Conclusion
• Additional experiments

3
Neighbor abstraction

• Nodes are divided into pairs that can communicate
  and pairs that can not
• Examples of usage
• Wired networks
• Obvious physical connection
• MAC protocol in 802.11
• Supported based on S/N and BER relationship

4
Roofnet

• Located in Cambridge, MA next to MIT
• 38 nodes with roof top antennas

5
Experiment setup

• Use 802.11b cards with Prism 2.5 chip set
• Set to channel 3 (2.422Ghz)
• Pseudo-IBSS (ad hoc) mode
• Run early Sunday morning June 6th, 2004
• 1500 byte 802.11 broadcast for 90 seconds
• Sequence Number, Arrival Time, Receive Signal
  Strength Indication (RSSI), Silence values are
  saved

6
Results
Red expected

• Results show that nodes dont adhere to
  neighbor abstraction

7
Delivery probability burstyness

• Average loss rate are all about 50

8
Loss due to long interference?

• Allan Deviation of loss rate measures
  fluctuations (high means very bursty, low means
  more smooth)

Yellow box shows that around 80 of all pairs
have low Allan Deviation of loss rate (around
0.06)
9
Loss due to distance?

• Yellow circles are the same size (approximately
  centered around Sender)
• There is relation of distance/receiving signal,
  however not consistent

10
Loss due to noise?

• High signal and low noise should result in
  delivery
• Roofnet doesnt show strong correlation

Result for1 Mbit/s High signal packets still
lost
11
Loss due to 802.11 interference?
Few interference packets before experiment

• If cause is this interference, then the more
  interference packets the more packets would be
  drop

12
So what is the cause?

• when you have eliminated the impossible,
  whatever remains, however improbable, must be the
  truth Sherlock Holmes in The Adventure of the
  Beryl Coronet (1892)

13
Reflected signals

• Buildings and other physical objects reflects
  signals
• Weaker duplicated signal are received by nodes
  after a delay
• RAKE receiver can suppress reflected copies with
  delays up to 250 nanoseconds

14
Experiment for reflected signals

• Signal sent to the receiver, and combined with a
  duplicated version of the original signal (with
  added delay, and reduce strength)

15
Results for reflected signals experiment

• Yellow shows RAKE working
• Green shows loss if reflected signal is almost as
  powerful as original the receiver is confused
• Blue shows intersection with the modulations
  symbol boundaries

16
Results for reflected signals experiment
(continued)
17
Distance between links

• Most links have greater than 500 nanosecond
  delays, RAKE cant handle

18
Conclusion

• No clear distinction of links that drop a lot of
  packets verses ones that receive most
• Link distance and Signal to Noise ratio has
  relation but not strong
• Most likely cause is multi-path fading (reflected
  signals)

19
Varying signal strength
1 Mbit/s

• The stronger the signal, the higher the
  probability of being delivered

20
Throughput compared based on bit rate
Shaded portion shows that X Mbit/s can
successfully send more packets than other rates

• Algorithm should wait until delivering half the
  packets before reducing bit rate

21
References and Acknowledgements

• Author's slides for SIGCOMM 2004 (figures in
  color were taken from these slides)
• http//www.pdos.lcs.mit.edu/roofnet/sigcomm-talk.p
  pt
• Authors electronic version of paper (other
  figures and tables were taken from this
  document)
• http//www.pdos.lcs.mit.edu/rtm/papers/p442-aguay
  o.pdf
• Roofnets group website
• http//www.pdos.lcs.mit.edu/roofnet/index.php