25 Dec 2015

WSJT-X V1.6 decode sensitivity

For many months I had been using V4 of the dedicated WSPR software After about a day of use I do not think WSJT-X V1.6 has demonstrably better sensitivity.  On spots of EA5DOM and PA3ABK/2 the S/N of decodes seems very similar in S/N terms. It is possible this not widely release version decoded at a similar level to WSJT-X V1.6. Certainly the QSB makes making a clear picture far from easy. What I mean is the signals vary far more with fades than the difference in sensitivity between packages. At the moment I am not using the preamp.

10 comments:

David (G0LRD) said...

Hi Roger,
An improvement in decoding ability should not change the reported S/N ratios, so I'm not surprised that these numbers look similar. The release notes state there is "a new two-pass decoder that can decode overlapping signals".
73 David G0LRD

Roger G3XBM said...

I had been told the 472kHz sensitivity was about 3dB better. I thought this would have meant weaker signals became spottable and the S/N of most (all) stations would be better. Maybe I misunderstood?

Anonymous said...

It's a very complex situation, the signal levels change at different rates or are simply not present at all whilst the QRN and QRM levels change with time as well in ways that may smooth out in the very long term but which are not totally predictable short term.

Improving your own reception often means reducing QRM levels, a lot of which comes from your own household equipment. As I have improved my own station the number of decodes has increased and the snr has improved. Having a near-by reference station is very revealing, when I started LF wspr'ing I had a local that had many snr numbers that were positive, while I always had negative snr's! That was totally down to picking up my own noise field. As you improve your setup the snr's improve and you log more stations although short term happenings can always mislead!.

Whilst Roger has proved that even a short earth electrode antenna works on transmit, it may not be the best RX antenna because of the way that noise from different devices is dumped into local grounds. A small tuned loop can be independant of ground, putting it in a quiet location with a well screened feeder will make a VAST difference as will common mode chokes at both ends of feeders..

There is a very effective check on noise control using a domestic MW ferrite rod radio during the middle of the day, tune to a quiet spot then place the radio over the feeder so the feeder is close to the rod and at right angles to the rod. If you can hear 50/100Hz noise increase then you need to choke the cable or use an isolating transformer. It is a very easy but revealing test!

Ultimately good LF results are more about NOT receiving local noise! the signals are there to be heard, it's just that we need to UNCOVER them!

Better DX for all in the NEW YEAR...

Alan
G8LCO

Jan Lustrup LA3EQ said...

I have been running the new 1.6.0 version parallel with the old WSPR-X on 474,2kHz receive. It shows a few dB better signal on some, but not all spots. The bad part is it does does not report all heard. Quiet a lot of stations do not decode at all on the 1.6.0 but they decode nicely on the old version, even though they are very strong. So there must be a bug in the decoding software?
Have you noticed this?
Jan LA3EQ

David (G0LRD) said...

The point I was trying to make is that the SNR reports are a straight measurement of the the wanted signal energy relative to the received noise energy in a 2.5kHz equivalent bandwidth. This should not be influenced by anything other than the end-to-end signal path as defined by transmit power, path loss, receiver performance, local noise sources, etc. If this new decoder performs better, I would expect the numbers to be the same but more decodes seen in the -27dB to -33dB range. Looking at the source code, the new C code for SNR calculation is not quite the same as the older Fortran one, so this may account for some minor differences.

73 David G0LRD

Jan Lustrup LA3EQ said...

But why is the WSJT-X 1.6.0 missing some decodes that the old WSPR-x is spoting ok??? It has nothing to do with s/n or local noise or even signal level!

Anonymous said...

Very interesting observations. Of late many decodes have been lost because the software tries once to send a spot then gives up when the server is busy. Before a week ago there were bad times every 10 mins from the hour, that then became every 10+2 mins as well then 10+2+4 as well! When mentioned on the Forum some while back I was told that if you save everything you resend when re-starting! That happens SOMETIMES provided that you re-start when the server is not too busy !!!!!!!!!!!!

I am taken aback that recently written s/ware does not store and forward ( and then catch up when you do connect). That means that all of the server logs become unreliable as a data record because around 20% of the spots are missing from some observers at some times. If people beacon at a fixed hour then hour +10 they will be excluded from many logs when the server was busy. Since the sysadmin appeared recently for a few hours matters seemed to become better, now they have slid back.

Even more remarkable is the sheer volume of spots, if you look at the Stats the daily volume is now vast. Every spot is uploaded around the same time for all bands!!!!!!!

Yet even more remarkable is the admission that server patches and updates are very rare events.

Jan's comments suggest an experiment, run 1 Rx on 1 band, then take the audio to two PC's each running the two programs but using two differing callsigns. I wonder if anyone has done this although there are a few /1 and /2 callsigns on WSPR. I would have thought that with the 100 or so testers trying out new versions that the decode performance was well known to the developers. That way the relative decode performance would be recorded on the server ( when or if you can connect !!!!!!!!!)

The snr readings are different for the variants, the 1.60 goes up to +18 so far, way more than +13, although high levels compress, it might be a linear log scale but it has a definite upper limit where it flat tops.

David's point is well made but there are many complexities to decoding real signals when you are dealing with signals around the threshold. I was compairing notes with a local on some WSPR decodes made a halfwave away from me with the same decoder on MF. We had variations of up to 7dB at times although normally around 1-2dB. Although early days the spot snr distribution with 1.60 seems biased, there seem to be too few spots at the lowest levels, that should be easy to graph.

Best regards to all,
Alan G8LCO

Jan Lustrup LA3EQ said...

I will do that experiment now. The call LA3EQ/1 will be the old WSPR-X program and LA3EQ/2 will be WSJT-X 1.6.0. I will conduct this test with a 3 meter high "T2FD" antenna as my LF/MF antenna mast just blew down in tonights storm...The wind is >27 m/s now.... :(

Jan Lustrup LA3EQ said...

I have started both programs at 23:00 UTC and will let them run all night until morning....Already WSJT 1.6.0 has not shown more then 1 decode while WSPR-X has 6 decodes..

Jan Lustrup LA3EQ said...

Good morning.....37 m/s storm last night...After 9 hours on 474.2kHz the old WSPR-X program decodes 4 times more stations then the new WSJT-X 1.6.0. version.