Frequency Calibration - the answer...

If you look at my blogs below you will see I have been puzzled by a discrepancy between what ARGO and Spectran reports as the frequency of an audio tone supplied and what WSPR reports.

I have FINALLY figured out what is happening (I am sure others wiser than me already know this - but I didn't).

Basically, there are 'standard' sample rates used in soundcards. The most common 'standard' rates are the harmonically related sample rates of 5512sps, 11025sps, 22050sps and 44100sps. The last one is the sample rate used for .WAV files destined to be put on audio CDs, and is very common. However, for the latest soundcards the 'standard' sample rates are 48000sps and 96000sps (harmonically related to each other).

The designer of a soundcard has to make a decision then as to what hardware rate he will use, one which is harmonically related to the 44100sps 'crowd' or the 'new crowd' based around 48000sps. Once he chooses one then the other crowd must be implemented by some software interpolation as the two 'crowds' are not harmonically related.

Say he picks 44100sps as his sample rate of choice - now 5512sps (more correctly 5512.5sps), 11025sps and 22050sps are available by downsampling by an integer factor of 8, 4 and 2 respectively. But 48000sps and 96000sps (if made available has to be interpolated (or more correctly extrapolated) by NON-integer factor. And vice-versa if the designer chose 96000sps as the base sample rate.

WSPR has chosen a sample rate of 12000sps which is harmonically related to 48000sps or 96000sps. Therefore if you have a soundcard which is based around 48000sps then it will read the correct audio frequency in WSPR. When you use ARGO (which uses 11025sps in bandview and 5512sps in horizontal scroll mode) then you are into the software interpolation mode and all bets are off as far as consistency between soundcards is concerned.

On the other hand, if your soundcard is based around the 44100sps rate, then ARGO is happy because the sample rates it uses are integrally harmonically related and will report the correct audio frequency. But if you use WSPR (with a NON-integer harmonic relationship of 12000sps) then you are at the mercy of the software interpolation and now its all bets off here.

Because I use an audio studio grade soundcard (albeit at the bottom of the range), there is a switch on the interface which (I presume) selects a native sample rate of either of 441000sps, 32000sps, 48000sps or 96000sps.

When I was doing the measurements and found the 17Hz discrepancy between ARGO (which reported the correct frequency) and WSPR the UA-1eXs were switched to 44100sps (i.e., 12000sps for WSPR would have to be derive using a non-integral division factor).

After finding Joe's post to Patrick and Joe's comment about sample rates defaulting to 48000sps as the reason he chose 12000sps, I remembered about the hardware sample rate switch on the UA-1eXs.

So I restarted one of the UA-1eX in 48000sps mode and re-ran the calibration tests...

Guess what ???

WSPR now reports the correct 1478Hz (pretty close to the 1480Hz reference input) as it is doing an 'easy' divide by 4 factor, and ARGO reports an incorrect 1510Hz as it is forced to do a divide by 8.707482993 and obviously can't do it to that accuracy.

Spectran in this 48000sps mode reports an incorrect 1510Hz @ 5512sps, 1461Hz @ 8000sps, 1469Hz @ 11025sps, 1469Hz @ 22050sps, but correct frequencies 1479Hz @ 44100sps and 48000sps.

So the answer is ????

Well, for me - hardware switch the UA-1eX to 48000sps when I am using WSPR and switch to 44100sps when using ARGO or Spectran.

General Principle

Find out the sample rate of the software you are using and (if your soundcard will do it) switch the hardware so that the sample rate is harmonically related to that sample rate.

I am not puzzled any more.