--------------------------- SIGCOMM 2009 Session 1: Wireless Networking 1 Scribe: banchs@it.uc3m.es --------------------------- *Cross-Layer Wireless Bit Rate Adaptation* _Mythili Vutukuru_, Hari Balakrishnan, Kyle Jamieson --------------------------- Rate adaptation: needed because of channel variability (large-scale attenuation, small-scale fading, interference) Requirements: estimate channel accurately, responsive, do not react to interference Existing algorithms: frame-based (slow), snr-based (specific to environment) Key ideas: - SoftPHY hints to provide confidence information - SoftRate estimates BER based on this information SofPHY hints: - demodulator computes error vectors at the physical layer - out of this internal state, there is information on the confidence - these are the softPHY hints - in Viterbi decoder: log likelihood ratio SoftRate: - interference-free BER - interference is detected based on a threshold over the drop - periods affected by interference are not used to compute BER Rate selection at the sender: - assumes that a higher and lower rates have at least an order of magnitude in BER - throughputs are computed from rate and BER - the ranges of BER for which a given rate is optimal are precomputed - when the BER moves out of the optimality range the rate is changed - thresholds can be computed to move more than one rate at a time Evaluation: - SoftPHY hints predicts BER accurately - SoftRate performs close to the Static Best - It performs 2x throughput of frame-based schemes - It performs similarly to carefully trained snr-based schemes and better than these schemes (4x) when the operating point deviates from the training    Future work: other cross-layer protocols Q: Which are the assumptions on BER values upon which the selection of optimality range depends? A: There are no assumptions other than the separation of one order of magnitude. The exact BER values do not matter. Q: As the rates go higher, does the drop due to interference also decrease? How does this affect the threshold choice? A: The gaps do not depend on the rate, only on the distance from the interferer. Q: SoftPHY hints are obtained across all the bits of a packet, while SNR data is only measured at the beginning of a packet. Why isn’t SNR also measured for all the packet? A: To measure SNR, you need the symbols in the preamble. Q: Any idea on an improvement over your work? A: Less complex information than SoftPHY hints. Q: You have conducted the evaluation for TCP traffic. If you have VoIP packets, you have plenty of channel time and should choose the lowest data rate. A: Yes, rate adaptation is only needed for data intensive applications. --------------------------- *SMACK – A Smart ACKnowledgment Scheme for Broadcast Messages in Wireless Networks* _Aveek Dutta_, Dola Saha, Dirk Grunwald, Douglas Sicker --------------------------- Traditional ack mechanism for broadcast: send one ack after the other Proposal: send all replies at the send time thus reducing ack time OFDM implementation: each user transmits a tone in a different subcarrier to ack reception Built on 802.11 physical layer, acks can be received between 10 microsecs for 52 users (number of subcarriers available) Threshold based detection from experiments: although there are fluctuations, a constant threshold can be used Ensuring reliability: - varying signal power: closed loop transmit power control - narrowband interference: multiple subcarriers - wideband interference: FFT before and after - wideband short-lived: two unassigned subcarriers spaced Higher layer fallback: if the scheme fails, higher layer mechanisms should be used to recover Evaluation shows not many false positives nor false negatives (around 99% correct) Q: What about alternative solutions such as piggybacking? A: It may be an option in some cases although in other cases when you need a quicker feedback the proposed solution may work better. In any case, the solution proposed will always be combined with a higher layer protocols. Q: May strong receivers mask the weaker ones? A: Experiments show that the solution works for varying SNRs between 15 to 27 dBs. In case of bigger variations, you can use power control. In addition, there will always be a fallback mechanism. Q: How many receivers can the system support? What if the number goes higher? A: 52 responders in 802.11. For more receivers, you need to use multiple slots, in this way you can scale. Q: What is the timescale of the closed-loop power control? A: This is ongoing work. Q: In case of a multi-AP network with multiple collision domains, will other data traffic cause false positives? A: No, since data traffic would back off and not interfere. Q: How can synchronization in the order of microseconds be achieved? A: This is not part of this work. Q: What about links with quality lower than 15 dB? A: The threshold should be low enough for distant nodes. If you cannot have 802.11 communication, you do not need the Acks. Q: Would it be possible to use more bits of information with multiple bits in a subcarrier? A: This would be hard. It would be preferable to multiple slots instead. --------------------------- *White Space Networking with Wi-Fi like Connectivity* Paramvir Bahl, Ranveer Chandra, Thomas Moscibroda, _Rohan Murty_, Matt Welsh --------------------------- FCC allows the use of white spaces by unlicensed devices as long as the owners (TV and microphones) are not using them Goal: deploy wireless infrastructure that provides good throughput while avoiding interfering with incumbents Spectrum is heavily fragmented, we need to adapt to variable channel widths and also to special and temporal variation of available channels Discovery algorithm - trying all possible central frequencies and widths is inefficient - SIFT improves discovery time - It infers width from packet duration which allows quicker converge - Two algorithms proposed: linear SIFT and jump SIFT - Evaluation: 2x reduction for 5 adjacent channels Spectrum assignment: the MCham algorithm measures of the expected throughput at each possible channel and chooses the best one Evaluation with a prototype shows that each time background traffic varies, the system moves to the best part of the spectrum and provides near optimal performance Q: How does incumbent detection work? A: There has been a lot of work on this, including a geo-location data base. We take it as a black box. Q: Why have you chosen to use contiguous spaces? A: Because we were using off-the-shelf components. A lot of the principles of our work, though, would also work in case of subcarrier suppression. Q: What about interference to receivers nearby that use adjacent channels? A: Yes, this is an issue. Q: Is csma/ca is the right way to go for white spaces? A: We are actively looking at it. Q: How difficult is it to modulate signal to multiple carrier with noncontiguous part of the spectrum? It seems more difficult to look for big chunks of space. A: Yes, this is a possibility and we are looking at it. As I said before, many of the principles hold in this case. Q: Why doe it takes 20 to 25 sec to figure out channels available? A: This is due to implementation specific reasons.