WSPR sample rates - killing two birds with one stone...

Recapping on Frequency Accuracy
As you can see from previous entries in my blog I have been delving into why WSPR and ARGO report different audio frequencies with an error of 17Hz at 1500Hz.

This has been tracked down to WSPR using a non-standard sample rate of 12000sps. ARGO uses 5512sps and 11025sps which are standard rates and harmonically related to the HARDWARE sampling rate of 44100sps I had set on my UA-1EX USB audio interface. When using WSPR the OS-supplied driver was forced to do a tough non-integer downsampling from 44100sps to 12000sps. This gave the frequency error when using WSPR and the accurate reading when using ARGO.

My audio interface is probably a little unusual so far as I can switch the HARDWARE sampling rate. When I switched the UA-1EX to 48000sps, WSPR then reported the correct audio frequency as the downsampling was a simple integer divide down to 12000sps. This can be achieved simply by the driver by only retaining and using every fourth sample (after adjusting the anti-aliasing filter) or by block averaging four samples (no need to adjust anti-aliasing filter). The second case is the most likely as it allows different derived sampling rates at the same time (and the anti-aliasing filter set for the native sampling rate).

Unexpected Side-Benefit...
Once I had switched to the 'WSPR-friendly' sampling rate of 48000sps, not only did it fix the audio frequency error, but I have noticed a dramatic increase in decoding sensitivity. Now virtually any station (there are still a few consistent exceptions...) I see on the screen, no matter how faint (lots of stations below -20dB SNR; right down to -29dB SNR), is being decoded.

Perhaps the non-integer downsampling process introduces an artificial 'flutter' on the signal which stresses the WSPR decoder.

Also, and this might be my imagination, I seem to be decoded by others significantly more often. So much so I will probably drop my signal by 10dB back to 500mW today or tomorrow.

I see from the documentation that WSPR QSO running in WSJT uses a different sampling rate to WSPR standalone. I also note that people are reporting that WSPR QSO is less sensitive than WSPR standalone.

I reckon these people are experiencing the reverse of my original WSPR standalone problems. That is, their soundcards are defaulting to 48000sps and so the WSPR standalone sample rate of 12000sps is an easy integer downsampling process resulting in good decoding. But when they use WSPR QSO in WSJT that same 48000sps is causing a difficult NON-integer downsampling process causing less sensitive decoding.

In the days to come I will get WSPR QSO in WSJT installed and runnning and compare the decoding sensitivities to WSPR standalone to determine whether it is linked to the native sampling rate of the audio interface.

Because I have an audio component grade soundcard (UA-1EX) I can do this in HARDWARE.

I will report my findings when done.