> On Aug 8, 2021, at 11:37 PM, Phil Karn via wsjt-devel > <[email protected]> wrote: > On 8/8/21 6:41 PM, H.Shrikumar KA6Q wrote: > >> Comments inline, old-school (the way it was meant to be) ... > Right!!
Ah, just like the good old days! I assume it's OK to trim as well? Yay! > OTOH, we're hams; do we want our stuff to rely on the Internet? > Personally I want my stuff to be _able_ to operate independently of the Internet, though I'm happy to have it around for normal operation, and to detect (for a recent instance) things like my external GPS antenna on my local Pi4-based NTP server failing. > WSPR uses a r=1/2 k=32 convolutional code. At the end of each message it > flushes the convolutional encoder with 31 0-bits to return it to all-0s (same > as the starting state). These "tail bits" provide the error *detection* you > need here, because it is very unlikely for my Fano decoder to make it all the > way through the tail with corrupt data unless you set the decoding time limit > *very* high. While it's not a CRC, the tail serves a very similar function. > When you set a decoding limit, you are making a trade off between error > *correction* and error *detection*. Too small a limit, and you erase (throw > away) frames that might have decoded correctly. Too high a limit and you risk > decoding a frame incorrectly (getting garbage marked as correct). EVERY > coding scheme has this trade off, whether or not you actually tweak the knob. > Even a CRC can be used for error *correction* at the expense of decreased > error *detection* Would it make any sense to try the decoding over several settings for the decoding limit, or is that too meta? Or is it actually time spent running over the dataset, and there's no way to tell how 'well' the decode worked? Obviously, getting a result sooner rather than later gives you a better 'quality metric' (or whatever it would be called), so is it worth keeping the data and this score and using a running average of the score to set your threshold? [Or am I, more likely, talking through my hat and just muddying the waters?] Thanks! 73, Willie N1JBJ
_______________________________________________ wsjt-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/wsjt-devel
