On Sep 25, 2008, at 1:46 PM, funofnet Funofnet wrote: > > > >> Hi all, > >> > >> I noticed that sometimes when I generate two link gain model files > >> ("linkgain.out") > >> by the "LinkLayerModel" java class for the same configuration, then > >> I ran a simulation to compute for example the packet delivery rate > >> of my application, I obtained incomparable results. > >> > >> Let me give you an example for best understanding: > >> A simulation with the first linkgain file lets me get an > PDR=0,6..... > >> in contrast, a simulation with a second linkgain file (I repeat: > for > >> the same configuration) lets me get an PDR=0,05.... > >> > >> > >> > >> It is a strange thing which lets Tossim simulation results to be > not > >> RELIABLE!!!!! > >> > > >Do you mean "deterministic?" > > >Of course the link model generator is not deterministic; it considers > >hardware variations. The point is that two sets of nodes, placed in > >identical physical layouts, might have different packet reception > >ratios due to sensitivity variation across the nodes. You're testing > >two different networks, and so the results are, well, different. If > >you want to test the same network, use the same link gain file. > > >Phil > > Thanks professor, > But the amazing thing is the huge difference between the two results > (0.6 and 0.05) > Normally, results should be comparable (0.6 , 0.5, 0.7 in the worst > case 0.3).
Why? For a CC2420, a 1.5dB chance in the SNR is the difference between a PRR of 90% and 10%. Noise floor variations are much larger than 1.5dB. Not to mention the fact that the signal itself varies by at least that much over time. Kannan has a paper appearing in SenSys on this, you can take a look at the tech report version here: http://sing.stanford.edu/pubs/sing-07-01.pdf Phil _______________________________________________ Tinyos-help mailing list Tinyos-help@millennium.berkeley.edu https://www.millennium.berkeley.edu/cgi-bin/mailman/listinfo/tinyos-help