Turns out neither you nor another fellow responded to my demand for ABX DBT for 'proof' of audibility. However, you did appear to hear a difference and did not provide the 'proof' you demand from other people:
"...hearing the difference in Purifi files is not about phase perception but amplitude perception as the other file has amplitude modulation."
"There is nothing extraordinary in that the resulting AM modulation is audible."
https://www.diyaudio.com/community/threads/placement-of-resistors-in-signal-path.384534/post-6988187
https://www.diyaudio.com/community/threads/placement-of-resistors-in-signal-path.384534/post-6988168
Back in 1978 Quad set up a comparison test using the Quad II valved/tubed amp, the 303 transistor amp (which has an output capacitor) and the 405 current dumping amp.
The listening panel included Laurie Fincham (then at the technical helm of KEF), Jim Rogers (JR loudspeakers), John Crabbe and John Borwick.
Attached
The listening panel included Laurie Fincham (then at the technical helm of KEF), Jim Rogers (JR loudspeakers), John Crabbe and John Borwick.
Attached
The problem is making sense of the resulting data. You could just as easily ask, why don't you take an FFT of each dac playing the same song, then compare the spectra? Again the problem is in making practical sense out of the data.
Regarding using an ADC for two more or less SOA dacs, there is a principle in testing that the test instrument should be at least 10 times better than the DUT. Otherwise it gets complicated to disentangle the test instrument imperfections to those of the DUT. Given we measure ADC performance using PSS FFT analysis, we have the same issues of unspecified or underspecified dynamical behavior from its measurements as we do with the DUTs to be compared.
IOW, its probably not as simple in practice to make work as you seem to be suggesting.
The purpose is to look at the difference in the recordings as compared to the original. IOW which is more closer to the original. Any possible imperfections of the ADC should show up in both recordings.
Do you mean with synchronous clocking of the DACs and the ADC?
How would the difference file be analyzed?
What if the ADC and one DUT had a common problem due to using a common modulator architecture? What if both had some dynamic compression and or hump distortion that could cancel out?
I'm not sure why the clocking would be important. ADC is capturing the analog output of the DAC. To be able to record outside of audio range it might make sense to have the ADC record at high sampling rate (192k or 384k).
E.g. in Audacity it is possible to view individual samples if needed. Also subtraction (and other operations) is possible.
If you have fears about hump distortions you could use an ADC that does not have them (e.g. AK5394).
One problem I see with this scheme is that the synchronizing of the recordings for comparisons may be quite difficult unless there is some sync tone in the recordings.
E.g. in Audacity it is possible to view individual samples if needed. Also subtraction (and other operations) is possible.
If you have fears about hump distortions you could use an ADC that does not have them (e.g. AK5394).
One problem I see with this scheme is that the synchronizing of the recordings for comparisons may be quite difficult unless there is some sync tone in the recordings.
Agree subtraction is possible, but not sure how one would make a sensible analysis of the difference?
Also, I have viewed samples in Reaper and found that sample point values are different in recordings of the same source material. There are corresponding small variations in the displayed waveform connecting the samples as well, although its probably mostly an artifact the limited quality of the reconstruction filter? Thinking about that does raise a question in my mind about the dac reconstruction filter used. What about differences from that, do they count as distortion/noise during analysis?
Also, I would expect noise from sample point misalignment that might need to be integrated out. How to analyze the difference then? Maybe it would be possible to resample one file to align sample points?
Also, I have viewed samples in Reaper and found that sample point values are different in recordings of the same source material. There are corresponding small variations in the displayed waveform connecting the samples as well, although its probably mostly an artifact the limited quality of the reconstruction filter? Thinking about that does raise a question in my mind about the dac reconstruction filter used. What about differences from that, do they count as distortion/noise during analysis?
Also, I would expect noise from sample point misalignment that might need to be integrated out. How to analyze the difference then? Maybe it would be possible to resample one file to align sample points?
Last edited:
If the audible difference is as easy to detect as you have implied I'm quite sure there is no need to study the recordings at sample level. And as the recordings themselves are easy to make speculating about the sensibility of the analysis beforehand is pointless. Once there are some recordings it is possible to decide if the whole exercise is futile or how it could be improved.
Are you equipped to try this idea out with a couple of your dacs, maybe a sigma-delta, and IIRC you said you built a switched resistor dac?
If you can show its feasible/useful maybe I will need a more up to date ADC... Best one I have now is in an old Crane Song HEDD 192. Sounds good, but hardly SOA.
EDIT: I would be interested in this using DSD files, since one dac can only play DSD, and the other has that option. That could potentially take the PCM->DSD conversion software out of the analysis of differences. Not sure how to record Native DSD though, have to think about that...
If you can show its feasible/useful maybe I will need a more up to date ADC... Best one I have now is in an old Crane Song HEDD 192. Sounds good, but hardly SOA.
EDIT: I would be interested in this using DSD files, since one dac can only play DSD, and the other has that option. That could potentially take the PCM->DSD conversion software out of the analysis of differences. Not sure how to record Native DSD though, have to think about that...
Last edited:
Have lots of 16/44 too, but we can hear the difference in resolution here on the ESLs. Might be okay for initial test though.
What kind of music are you looking for, or would you like a variety? For dynamics and imaging there is one in particular I can think of.
For accurate reproduction of piano (hard to do well) it might be something else. Would probably go with the dynamics and imaging myself.
Also occurred to me, how about taking the recorded file from a dac, playing it back through the dac, and recording that. It should show how lossy the process is.
What kind of music are you looking for, or would you like a variety? For dynamics and imaging there is one in particular I can think of.
For accurate reproduction of piano (hard to do well) it might be something else. Would probably go with the dynamics and imaging myself.
Also occurred to me, how about taking the recorded file from a dac, playing it back through the dac, and recording that. It should show how lossy the process is.
Last edited:
This could be done and might reveal differences, though this is highly unpredictable under circumstances as subject to all the issues that Markw4 has mentioned. Even if this is done perfectly this has issues specific to revelation of hearing sonic differences.
I personally tested three highly spec'd DAC's (all with balanced output) as an SMSL Su9n, SMSL D-6 and SMSL D100. They all showed up sonic differences (not necessarily better) in my system (I used to have stacked quad ESL57's that likely would have revealed more). The point is that buying these for yourself and swapping them out is no guarantee that differences could be observed by others in their systems. Doing recordings of those differences is even more removed from supporting some form of reality.
I've had that experience (in the second paragraph) with MP3. I had to encode and hear a CD selection at 96k to be able to know what to listen for and hear the differences between the CD and 128k. I had been reading online where people with "good ears" were saying how awful MP3s sounded, but I had to consciously listen for it myself. That taught me, among other things, that my ability to hear wasn't nearly as good as I had thought or hoped.
It looks like we're already here. It's certainly a contentious topic, especially in one area Mark mentioned, speaker cables, that has many times resulted in locked threads.
Certainly there are many who want to buy/make good equipment and "just listen" and say "this make and model of X sounds better than that one" but I've always leaned strongly to the scientific investigation side of things and want to know why.
There are plenty of audio forums populated with "just listen" people, but diyaudio in one of the few places with a substantial population of "let's find out why" people.
Last edited:
Isn't that somewhat circular reasoning? Which way to ensure sufficient transparency on an ADC if the transparency of the DAC is already the EUT?
Wrt to other disputes in this round - a sensory test is considered as an objective measurement although the detector (aka listener) used is a human being.
ABX is a protocol only usable for discrimination tests, introduced - as already cited by rayma - around 1959. Harris wrote in a letter (1952) to the editors of the JASA about his findings that listeners got better results in A/B-tests compared to ABX (IIRC already quoted in other threads), so the experimenters at that time already knew about the fact that different protocols could lead to quite different results, hence the subsequent development of guards (usage of positive and negative controls) against such confounders.
Obviously, it depends on the EUT, usually a (potentially) multi-dimensional difference requires more care than a one-dimensional one.
