Hi all,
WSJT-X v1.6.1 now (SVN revision 6020) includes a first cut at using
Steve's 2-pass decoding scheme for JT65.
As you will have seen in previous posts by Steve and me -- based on use
of a stand-alone, single-purpose program -- the 2-pass scheme produces
significantly more decodes from crowded band segments. In both passes
it uses the "sfrsd" algorithm rather than the patented, closed-source
algorithm of kvasd.
Here's an update on the table I shared here on 10/2/2015, summarizing
JT65 decoding performance of JT65-HF-HB9HQX-Edition and several versions
of WSJT-X. All programs analyzed exactly the same data, recorded (by
WSJT-X r5912) on 20 meters.
Correct False
Program Decodes Decodes Decoder
------------------------------------------------------------
JT65-HF 2329 24 BM + kvasd
WSJT-X r5912 2249 0 BM + kvasd
WSJT-X r5955 2114 0 BM + sfrsd
WSJT-X r5955 1816 0 BM only
WSJT-X r6020 2874 0 BM + sfrsd (N=1000) + 2-pass
If you've been following SVN updates and relevant posts made here you
will know that I've also been experimenting with "experience-based"
decoding of JT65 signals. So far this has been done in SVN trunk, the
WSJT branch. I will soon start work on merging this code into WSJT-X.
This will require making some distinction between HF-style
"multi-decoding" of all JT65 signals found in the received passband, and
VHF/UHF/Microwave/EME-style decoding where one (or a small number)
signals are expected to be present in the passband.
The new JT65 decoder has some user-adjustable parameters (see attached
screen shot of the "Settings | Advanced" tab). At present the only one
activated is a spinner that sets the number of random erasure patterns
tried in the sfrsd algorithm. The default setting is "6", which sets
variable ntrials = 1000. Other settings as follows:
Setting ntrials
--------------------------
0 0 ==> BM only
1 3
2 10
3 30
4 100
5 300
6 1,000
7 3,000
8 10,000
9 30,000
10 100,000
11 300,000
12 1,000,000
Bigger numbers generally mean slower decodes.
As always, feedback is encouraged!
-- 73, Joe, K1JT
------------------------------------------------------------------------------
_______________________________________________
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel
