On 5/9/06, Prabal Dutta <[EMAIL PROTECTED]> wrote:
Actually, Rob, I think we're both right (and wrong). If you look at the Telos data, RSSI shows much lower variance and much greater range for short distances whereas LQI shows comparable variance but greater range for medium and longer distances. If distance approximates PRR, then RSSI would be a good estimate for decent links whereas LQI would be a good estimator for worse links. I have some other data that I'm not quite ready to release yet, so I'll agree with you that that I can't argue based on it without sharing first. ;-)
I completely agree with this: if good links exist (PRR is close to 1), you can find them based on RSSI. When no such links exist, LQI seems to have a better correspondence to the weak links. I will be on a lookout when your paper appears in print. Cheers, Rob
You're right that RSSI will pick up narrowband interferences (especially since its sampled over only eight symbols) which is why I wrote "...RSSI to interference and noise largely determine whether packets are received...", the key word being largely (but not always, as in the case of narrowband interference). - Prabal On 5/9/06, Robert Szewczyk <[EMAIL PROTECTED]> wrote: > You're correct that the distance is just introducing confusion into > this discussion, it would have been much better to do plots of PPR v > LQi and PPR v. RSSI. > > I would disagree however with your characterization that RSSI has a > better correlation with PPR (and you have not presented any actual > data to back up your conjecture). The graphs in the telos paper, > while noisy, point at least somewhat towards a decent correlation > between PPR and LQI; the flat region of the RSSI curve with large > error bars would imply a distribution that is quite different than the > PPR. LQI does have a close correspondence with the PPR -- it measures > number of chip errors in the encoding of the first few bytes of the > transmission; once that number goes beyond what's correctable, then > your packet is going to get lost. Effectively this boils down to > sampling the PPR on a very short time window. In contrast, RSSI will > pick up narrowband interferences and count them toward the > measurement, even though they may end up being irrelevant for the > actual reception. > > Rob > > On 5/9/06, Prabal Dutta <[EMAIL PROTECTED]> wrote: > > The Telos graphs are nice in that they show the relationship between > > some of these variables but some mental gymnastics are required to > > make the leap from what's shown to PRR vs RSSI and PRR vs LQI. And, > > given the variances involved, I'm not sure that everyone would come to > > the same conclusion. > > > > I would argue that from a PRR, RSSI, and LQI perspective, distance is > > a largely unnecessary nuisance variable. Distance (and position) > > ultimately affect RSSI and LQI, and perhaps knowing the distance helps > > in some way to determine PRR, but at the physical level, RSSI to > > interference and noise largely determine whether packets are received. > > > > So, if you can directly measure and correlate PRR, RSSI, and LQI, then > > it would make sense to do so and ignore any intermediate > > parameterization (like distance). > > > > - Prabal > > > > On 5/9/06, Robert Szewczyk <[EMAIL PROTECTED]> wrote: > > > There is a graph that shows a dependence between distance and PRR, > > > LQI, and RSSI in the paper about Telos design (figure 5 in > > > http://www.polastre.com/papers/spots05-telos.pdf) > > > While it does not actually show the regression, and distance is made > > > explicit, it should at least give you a flavor of what to expect out > > > of each type of measurement > > > > > > Rob > > > > > > On 5/9/06, Prabal Dutta <[EMAIL PROTECTED]> wrote: > > > > Yes, there is *some* relation: they're positively correlated. But on > > > > the CC2420 radio, LQI shows significant variance for a given packet > > > > reception rate (PRR). You may be much better off using RSSI, which > > > > shows far less variance for a given PRR. > > > > > > > > You can use (logistic) regression to determine the relationship the variables: > > > > - send a lot of packets between a lot of motes > > > > - compute the PRR as the fraction of packets received over the number > > > > sent for each tx, rx link pair > > > > - note the RSSI and LQI values > > > > - Plot (and regress) PRR vs RSSI and PRR vs LQI > > > > > > > > Hope that helps. > > > > > > > > - Prabal > > > > > > > > On 5/9/06, Venkat Manoj <[EMAIL PROTECTED]> wrote: > > > > > > > > > > Hi all, > > > > > > > > > > Can anybody tell me if there is some relation between the link quality > > > > > indiactor (LQI) and the probability of packet loss, between two motes? Or is > > > > > there some method to find out the relation? > > > > > > > > > > Thanks, > > > > > Venkat. > > > > > _______________________________________________ > > > > > Tinyos-help mailing list > > > > > Tinyos-help@Millennium.Berkeley.EDU > > > > > https://mail.millennium.berkeley.edu/cgi-bin/mailman/listinfo/tinyos-help > > > > > > > > > > > > > > > > > > > > > > > _______________________________________________ > > > > Tinyos-help mailing list > > > > Tinyos-help@Millennium.Berkeley.EDU > > > > https://mail.millennium.berkeley.edu/cgi-bin/mailman/listinfo/tinyos-help > > > > > > > > > >
_______________________________________________ Tinyos-help mailing list Tinyos-help@Millennium.Berkeley.EDU https://mail.millennium.berkeley.edu/cgi-bin/mailman/listinfo/tinyos-help
