I want to share my data from a while back when I was looking at PRR/LQI/RSSI on MicaZ: http://enl.usc.edu/~om_p/etxlqirss/
For this particular discussion, I wish I had a plot of LQI/PRR and RSS/PRR (instead of LQI/PRR and LQI/RSS) to compare with Phil's and Prabal's data. Nevertheless, my data seems to suggest that RSS has less variance than LQI especially when the links are not that great. ------------------- On May 10, 2006, at 3:08 PM, Joe Polastre wrote: > What is preventing the release of your data? Levis and his student > have a paper already available in emnets, my paper was out last year, > so why can't you release your data? they're called technical reports > for a reason. > > bottom line--you need to show compelling data to be taken seriously. Here's a link to the camera-ready of Kannan's EmNets paper, which I think has reasonably convincing data. It shows (most of) my evidence that leads me to the same conclusion as Prabal: LQI is not a very good link estimator. Full stop. http://csl.stanford.edu/~pal/pubs/emnets06.pdf High level points: o LQI is correlated with packet loss o individual LQI readings have high variance o in our experiments, it takes a window of >50 packets to obtain an average LQI within 10 units of the observed mean o 10 units of LQI error is the difference between a 20% and 50% PRR, or a 50% and 80%. Note that if you send >50 packets, you can estimate (with DSN) PRR within 2-4%. The low variance of RSSI makes it a much better indicator of environmental changes. The *one* advantage that LQI has over RSSI is that it is much more chip-independent, in that hardware variations such as receive sensitivity (CC2420 advertises within 3dBm but we've observed much greater variations) are factored out. But it's not clear whether that's very useful in the end. The only way to know is to build and compare link estimators, which no-one has done. The long and the short of it is that the data in the Telos paper is very misleading. As the X-axis is distance there is no way to know the correlation between the values for individual directed node pairs. The variance across readings in individual pairs makes LQI a very inaccurate estimator. As far as I know, there has not been any evaluation of whether LQI is effective, and by that I mean a comparative evaluation against other techniques. The plots in the Telos paper have led many network protocols to mistakenly use LQI when it is not very good at all. Phil _______________________________________________ Tinyos-help mailing list [email protected] https://mail.millennium.berkeley.edu/cgi-bin/mailman/listinfo/tinyos-help
