Hi Bill, I’m kind of stuck at the moment. Hoping that you will see something that I’m missing. In r6773, I am getting consistent decoding results using either msk144d or wsjt-x on linux, but on OS X I am getting different results between msk144d and wsjt-x, and also different results between either of these programs on OS X and linux.
If you get a chance, perhaps you can run on Windows and see what happens. Maybe we’ll find that the problem is specific to OS X? Dunno. In any case, here’s a link to a zipped batch of wav files. These are recordings of Joe sending MSK144: https://dl.dropboxusercontent.com/u/33211132/msk144_test2_0604_k9an.zip For the record, here’s what I get when I decode these files on linux using the following command (from within the wsjtx/build directory: $ ./msk144d ~/Desktop/msk_test\ data/msk144_test2_0604_k9an/*.wav 115345 1 4.7 1497 & K9AN K1JT FN20 115445 0 2.8 1492 & K9AN K1JT FN20 115515 0 6.5 1491 & K9AN K1JT FN20 115615 0 0.8 1493 & K9AN K1JT FN20 115715 0 10.4 1493 & K9AN K1JT FN20 115915 0 2.0 1492 & K9AN K1JT FN20 120045 0 14.0 1491 & K9AN K1JT FN20 120115 -3 1.5 1495 & K9AN K1JT FN20 120245 4 10.3 1491 & K9AN K1JT FN20 120315 0 2.1 1492 & K9AN K1JT FN20 120345 3 9.2 1491 & K9AN K1JT FN20 120415 2 2.6 1499 & K9AN K1JT FN20 120945 1 12.7 1497 & K9AN K1JT FN20 121045 1 2.4 1486 & K9AN K1JT FN20 121115 -1 11.6 1493 & K9AN K1JT FN20 121145 7 2.9 1491 & K9AN K1JT FN20 121345 0 5.8 1495 & K9AN K1JT FN20 121445 1 0.4 1497 & K9AN K1JT FN20 121515 4 12.4 1496 & K9AN K1JT FN20 121715 1 12.2 1488 & K9AN K1JT FN20 121815 2 10.7 1494 & K9AN K1JT FN20 121845 6 8.1 1492 & K9AN K1JT FN20 I get the same results if I decode from wsjt-x on linux. The results from running msk144d on OS X are almost the same, but there are a couple of differences. But if I run from within wsjt-x on OS X, I get the following: 115515 0 6.5 1491 & K9AN K1JT FN20 115615 0 5.9 1492 & K9AN K1JT FN20 115715 0 10.4 1493 & K9AN K1JT FN20 115915 0 2.0 1492 & K9AN K1JT FN20 120045 1 14.0 1493 & K9AN K1JT FN20 120245 4 10.3 1490 & K9AN K1JT FN20 120345 3 9.2 1490 & K9AN K1JT FN20 120415 2 2.7 1499 & K9AN K1JT FN20 120945 1 12.7 1496 & K9AN K1JT FN20 120945 -1 13.3 1492 & 9BVTCXFI4N5R 121115 -1 11.6 1492 & K9AN K1JT FN20 121145 7 2.9 1490 & K9AN K1JT FN20 121345 1 5.8 1496 & K9AN K1JT FN20 121515 4 12.4 1495 & K9AN K1JT FN20 121715 1 12.2 1488 & K9AN K1JT FN20 121815 2 10.7 1493 & K9AN K1JT FN20 121845 6 8.1 1492 & K9AN K1JT FN20 Instead of 22 decodes, I’m getting only 17 decodes, and one of them is a false decode. And some of the files that did decode on both platforms provide different results, i.e. different candidates decoded successfully, as can be seen for example in the different time stamps for the 115615 results given above. This suggests that there is some data corruption somewhere. A while back, Joe added “srand(-1)” to the ldpc_decode routine located in /lib/ldpc/dec.c to ensure that the random number generator that is used to dither the soft-symbols would be initialized the same way across runs and platforms. This is the only “randomization” that this algorithm is doing - everything else should be deterministic. Comments/suggestions would be most welcome! Steve k9an ------------------------------------------------------------------------------ What NetFlow Analyzer can do for you? Monitors network bandwidth and traffic patterns at an interface-level. Reveals which users, apps, and protocols are consuming the most bandwidth. Provides multi-vendor support for NetFlow, J-Flow, sFlow and other flows. Make informed decisions using capacity planning reports. https://ad.doubleclick.net/ddm/clk/305295220;132659582;e _______________________________________________ wsjt-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/wsjt-devel
