> On Sun, 18 Apr 2010 09:06:46 +0200 (CEST), Rein Couperus <[email protected]> > said:
>> On Sat, 17 Apr 2010 21:43:54 +0100, Stelios Bounanos said: >>> On Sat, 17 Apr 2010 10:37:31 -0700, Brian Lloyd said: >>> (I forgot to reply to all.) >> (I removed one level of > characters below) >>> It is really Really *REALLY* important to understand what is going on with >>> these protocols in the presence of propagation anomalies in order to be able >>> to make decisions about which modes work best under what conditions. My own >>> recent experiments in monitoring and graphing the phase/frequency changes of >>> the WWV carrier over time (20-30 minute samples typically) have convinced me >>> that sequential testing of protocols, i.e. transmitting a message and then >>> immediately transmitting the same message using the other protocol, is not >>> likely to produce valid results unless repeated numerous times and then the >>> results averaged with outliers discarded. >> True, and I seriously doubt that anyone has bothered to do that. > The pskmail client - server communication does this all the time. It looks at > the > result of the last frames (both S/N and arq result) and decides on the mode > to use for the next frame (separately for client and server mode). > That way we always use the most effective mode for the circumstances. > We don't need modes wider than 500 Hz to be fast, and change between raw PSK, > robust PSK, MFSK and THOR modes. > This way we are gathering lots of information on this subject automatically :) > The contestia and olivia modes do not transmit the right character set so we > cannot use them. I fixed up the quoted text above so it's clearer to see who was writing what. This thread was posted to the fldigi-alpha mailing list and can be read here (start with first "Contestia" email): http://www.mail-archive.com/[email protected]/index.html http://lists.berlios.de/pipermail/fldigi-alpha/2010-April/thread.html#1099 Brian was saying that non-repeated, non-averaged tests are likely to be flawed. I said I doubted that anyone tested the right way (I still do :), since on-air test reports do not mention this stuff at all. At best, the testing is said to have been done over a period of time deemed long enough, but no graphs, no averaging, no apparent effort to discard invalid data. I know you've said that pskmail collects data to decide which modes to use. But other than switching modes, and perhaps logging that data locally, it's not telling anyone anything :-) If it could collect the S/N, loss % etc. for a period of time and set of modes, and make it available, that could be very interesting! Pskmail servers might export such statistics (even to non-participants) over HTTP. > 73, > Rein PA0R -- 73, Stelios, M0GLD.
