Wondering about the WTF and Foobar 192 khz files with clock locked some more.
Decided to look at sample values at the peak of the waveform. With clocks locked you would expect the same exact place to be sampled at the peak for each peak of a repetitive waveform like the one used. Which should mean the same exact value for that sample. In the case of 44.1 khz sample rates the point of sampling will repeat exactly every 441 samples. Analog noise will prevent exactness every time, but should be low enough the variance at the peak samples should be quite small.
So with Foobar locked files I checked the peak sample and every 441 samples thereafter. The following is in -dbfs. The results look like what I would expect with locked clocks.
.42813
.42813
.42786
.42813
.42841
.42813
.42841
.42869
.42841
WTF locked file is about .1 db lower in level, but otherwise very similar.
.52642
.52756
.52784
.52784
.52756
.52784
.52756
.52727
.52727
What variance you see is consistent with somewhere near -80 db of noise interference.
Now looking at a 192 khz file with locked clocks I only need to find a peak and skip 192 samples to find what should be the same exact sampling point of the waveform. When I do that with one of the WTF files I saw:
.59906
.54863
.48092
.40294
.43883
.50931
.56931
.61197
.63190
.62768
Much more variance. Now that can be caused by at least two things. One is a high noise level, and the other would be clocks that aren’t well locked or are unstable in the locking. The FFT shows that there is not a general high noise level across the board. I therefore believe the sample time between the ADC and DAC is varying because the lock is not stable. Each sample is not occurring at the same point in time on the waveform. Looking at unlocked clock files shows even more variance, but that is expected as the exact peak sample point is drifting on and off peak over time.
I don’t know what connection they used to lock the clocks, but I have seen such behaviour when using TOSLINK cable. It locks fine at up to 96 khz, and while it locks and records at 192 khz you see such instability in the timing sometimes with marginal connections. Again it would be nice to see the 192 khz signals done at a quarter sample rate.
This is further evidence I think to distrust any of the results with the locked 192 khz files.