I was one who downloaded. I'm sorry I haven't responded yet, I intend to, just not enough time so far.
Is the download still active, and if so, where? Couldn't find it skimming through the last pages...
Just noticed something about bohrok2610 files. Are they by chance phase reversed from the original file? Or maybe one of the FPGA versions inverts the phase as opposed to the other FPGA version? Not sure what's going on yet, but something sounds possibly a bit suspicious.
EDIT: Just checked in a wave editor, and they all look to be in the same phase. Maybe a phase shifts at certain frequencies accounts for the difference in transient sounds.
EDIT: Just checked in a wave editor, and they all look to be in the same phase. Maybe a phase shifts at certain frequencies accounts for the difference in transient sounds.
Last edited:
I now had the time to analyze @bohrok2610 's files (the *nodelay.wav's), left channels. After manual trimming and DC offset adjust it already looked like they are extremely close (except for a sub-sample shift) and clock drift is close to nonexistent (despite being not clock-synced?). DW says the same, as does REW. Only microscopic differences within the last kHz or so.
We have the usual trimable gain&DC and a sub-sample offset, but required drift correction is zero.
To my surprise residual waveform is dominated by gain drifts, causing the residual to pump up and down in volume, with peaks reaching -80dBFS. That indicates DC drift and 1/f noise of the reference voltages used for DAC and ADC. This is not uncommon IME.
Otherwise the residual sounds natural and completely ok (and the spectrum does look good as well), no significant spectral differences or other artifacts, as far as noticable within the broadband noise of course.
Now the point is, I would expect to see the same level and kind of differences when just doing two takes of one and the same setup -- actually I think this kind of base measurement is always required to see what's a reasonable resolution limit. Like an A/B/A/B/A/B/A/B sequence (4 takes of each version alternated), after a 2hrs thermal settling period, playing B looped.
To me at least, aged ears and all, the data suggests there is hardly enough audible difference in these recordings of the DAC output. I didn't ever bother to try an ABX.
We have the usual trimable gain&DC and a sub-sample offset, but required drift correction is zero.
To my surprise residual waveform is dominated by gain drifts, causing the residual to pump up and down in volume, with peaks reaching -80dBFS. That indicates DC drift and 1/f noise of the reference voltages used for DAC and ADC. This is not uncommon IME.
Otherwise the residual sounds natural and completely ok (and the spectrum does look good as well), no significant spectral differences or other artifacts, as far as noticable within the broadband noise of course.
Now the point is, I would expect to see the same level and kind of differences when just doing two takes of one and the same setup -- actually I think this kind of base measurement is always required to see what's a reasonable resolution limit. Like an A/B/A/B/A/B/A/B sequence (4 takes of each version alternated), after a 2hrs thermal settling period, playing B looped.
To me at least, aged ears and all, the data suggests there is hardly enough audible difference in these recordings of the DAC output. I didn't ever bother to try an ABX.
So far, three different people are in strong agreement about some basic differences between the FPGA versions. If you can't hear those differences when played direct from the FPGA then you can't hear them.
Making degraded copies and comparing them makes the differences more masked by the sound of the ADC. Just goes to show some analysis program relying on copies doesn't necessarily work like the human ear/brain.
Once again, I will refer to Paul Frindle's list of what was found to be audible, which IMHO is similar to what Bruno Putzeys found about hysteresis distortion: only a few people are going to hear some real differences. That's because there are thresholds of audibility for the AVERAGE ear, not for all ears. 50% of the population will be more sensitive than average, and 50% will be less sensitive.
Seems like some of the less sensitive fraction of the population is always looking for some justification such that if they can't hear it then nobody can. Then they demand ABX for every single reported difference. They expect 10-days of cooperation for each observation they can't hear. Their expectations of what they can demand from other people are not realistic.
Making degraded copies and comparing them makes the differences more masked by the sound of the ADC. Just goes to show some analysis program relying on copies doesn't necessarily work like the human ear/brain.
Once again, I will refer to Paul Frindle's list of what was found to be audible, which IMHO is similar to what Bruno Putzeys found about hysteresis distortion: only a few people are going to hear some real differences. That's because there are thresholds of audibility for the AVERAGE ear, not for all ears. 50% of the population will be more sensitive than average, and 50% will be less sensitive.
Seems like some of the less sensitive fraction of the population is always looking for some justification such that if they can't hear it then nobody can. Then they demand ABX for every single reported difference. They expect 10-days of cooperation for each observation they can't hear. Their expectations of what they can demand from other people are not realistic.
Last edited:
At what rate? Is it related to the tempo of the music?
Reason I ask is that all the people who heard differences found dynamics to be a significant difference between FPGA versions. Isn't pumping up and down on drum hits, etc., "dynamics?"
Last edited:
Clocks were synced (DAC using ADC clock).
I normally have my ADC inputs AC coupled but for these recordings I shorted the ADC input caps as DC coupling improves nulling. The side effect may well have been DC drift.
I noticed pumping of harmonics when I tested v03 and v04. It is similar in both versions. I reported about the pumping in post #2698. It was even included in chapter 3.6 of Marcel's report regarding low level distortions in post #2696. As this pumping of harmonics is not signal related it will appear at different spots with v03 and v04 recordings.
Similar pumping is clearly visible on the 2 otherwise very well matched recordings as shown in the waveform difference (see below).
Similar pumping is clearly visible on the 2 otherwise very well matched recordings as shown in the waveform difference (see below).
Attachments
No, zero correlation to the music.
Delta file attached (with 80dB gain applied).
I would have preferred 2x sample rate for the recordings.
Ahm, is the original file to be found somewhere in this thread?
Exactly.
Last edited:
What exactly do you mean here?
Also, may I kindly suggest to record two or more takes of the same modulator, but with the same procedure you would use when actually replacing modulators, to check for the baseline delta? At the moment this seems to be a big unknown.
Could you record at double rate?
That's new information, actually, at least to me. So far I understood that you had seen it with v4 and had hypothesized (not noticed) that it might also have been there with v3.
Understood. This is not easy to track down...
Do we know whether the process is deterministic (always starting from a defined and clean state)? Like when starting a playback of a (window-enveloped) sine track, will the "piston movement" in the spectrum always be exactly the same?
To check, the (same-to-same) residual then should be as close to noise as possible (besides the drift), and the relations of fundamental vs harmonics should stay intact
Last edited:
BTW, I have a DSD256 file of same song made from the original wav file but its almost one GB in size. If you just want to see if taking the FPGA out of the loop changes anything, here is the file:
https://www.dropbox.com/scl/fi/e79j...lues.dsf?rlkey=ry8qgcssrzjfsokdil1dtro5d&dl=0
https://www.dropbox.com/scl/fi/e79j...lues.dsf?rlkey=ry8qgcssrzjfsokdil1dtro5d&dl=0
It is there with v03 but due to the splitted peaks of v03 it is not so apparent in REW as the peaks remain in place but the HD number indicators move up and down. It is of course possible that pumping in v04 is greater.
BTW PJotr25 also noticed pumping with other modulators with RTZ but on a much lower scale.