By the way, I very much concur with the above notion. That we have a personal auditory memory, or template, of what live instruments sound like. Which automatically judges that we are hearing playback, rather than live music. Conversely, it's also what fools our brains, on those rare magical occasions, to judge that some given playback could actually be live instruments playing. I don't believe that it's necessary for that memory or template to be objectively accurate, only that it's what our brains use to judge auditory artificiality, from auditory reality. Just my current thinking.
The only thing that's clear to me is that (at least) one of us two must be very confused. The other people on the thread can figure out who by comparing posts #1126 and #1131.
I'm also confused but in a different way... I'm participant #5 and the pdf shows accurately my report. What puzzles me is that found "B" being the most involving -to use an alternative to non fatigue term that works better for me. I mean if I had chose "C" and rank down to "A" it would made more sense. Furthermore, when I switched "B" from NOS to OS it came closer to original files. I need to spend more time with this.
Regarding the types of DACs used for the echo experiment, this is the information I have:
#1: Traktor Audio 2, very probably oversampling
#2: Asus Xonar U5 soundcard with Cmedia CM9882A audio codec, signal upsampled by pulseaudio to 96 kHz using soxr-vhq, several IIR filters in the chain for equalizing
#3: S.M.S.L. M100 with an AK4452. It isn't clear what filtering is set up in the AK4452 by the M100.
#4: oversampling DAC of unspecified type
#5: PCM1794 that can be switched between bypassed and not bypassed digital filter
#6: oversampling DAC of unspecified type
Now that I wrote this list, I notice that the one who had a significant result is the only one who uses software for the first interpolation step.
#1: Traktor Audio 2, very probably oversampling
#2: Asus Xonar U5 soundcard with Cmedia CM9882A audio codec, signal upsampled by pulseaudio to 96 kHz using soxr-vhq, several IIR filters in the chain for equalizing
#3: S.M.S.L. M100 with an AK4452. It isn't clear what filtering is set up in the AK4452 by the M100.
#4: oversampling DAC of unspecified type
#5: PCM1794 that can be switched between bypassed and not bypassed digital filter
#6: oversampling DAC of unspecified type
Now that I wrote this list, I notice that the one who had a significant result is the only one who uses software for the first interpolation step.
Last edited:
Suspect List Status
I think that this would be a good time for another review of our suspect list. Here it is, with my comments in italics:
B) RECONSTRUCTION/IMAGE-BAND HANDLING
===============================================
2) Lack of an FIR interpolation-filter, freeing the DAC from certain processing 'artifacts' , such as:
a) time-domain signal echoes produced within Equiripple on-chip FIR filters.
b) impulse response ringing (pre or post)
c) half-band filters plainly violating Nyquist
d) are prone to clip on peak sample normalized recordings - the intersample overshoot issue.
e) purely analog image-band suppression inherently sounds different than digital suppression?
B2a, B2b and B2c - The investigation in to suspect Category group B2 in general has seemingly resulted in identification of the guilty sub-system ('typical' OS interpolation-filters) within an DAC which causes NOS and OS to sound characteristically different, even though not the specific mechanism responsible. Therefore, these three items will remain unverified possible root cause mechanisms. The PGGB 88.2KHz upsampling experiment indicates that the closer the oversampling comes to a theoretically ideal SINC-function, the more the subjective sound of OS becomes like NOS. Until it sounds close enough to eliminate the subjective difference between the two.
B2d can certainly be a problematic issue if prevalent enough. It's magnitude of impact when only minimally present is not clear to me. To truly isolate it's potential subjective impact would seem to require an specific experiment which offers a choice between a file with inter-sample-overs, and that same file having those removed via level reduction, prior to the interpolator block. Then equalizing the volume level of the second file so that it matches that of the file with the inter-sample-overs. We will not, however, be performing such an experiment in this thread. Although it would be interesting to see it performed in some other thread.
B2e - Since we've shown that high-performance digital OS image-band suppression can remove the subjective difference between NOS and OS, we've necessarily then also shown that purely analog image-band suppression is not necessary for doing so.
3) Phase-modulation of the baseband signal due to insufficiently suppressed image-bands. In other words, because the signal waveform is not fully reconstructed according to the sampling theorem requirements.
B3 Seems something of a puzzlement. While pedestrian OS interpolation would seem to be sufficient to, subjectively, perfectly reconstruct an audio signal waveform, perhaps more might yet be subjectively extracted as the filter performance is increased in some parameter or two. Therefore, our final experiment will be to test for whether even higher-performance OS image-band handling can produce sound which is subjectively superior to either NOS, or to it's PGGB 88.2 upsampled subjective equivalent. This final experiment will be conducted after Hans returns from holiday. It will subjectively evaluate 176.4KHz PGGB upsampling versus 44.1KHz NOS/88.2KHz upsampled.
4) The unsuppressed image-bands are, somehow, producing audible IM products directly within the ear.
Whether B4 is an audible mechanism or not (the research seems very sketchy), IMD seems highly unlikely to produce the relaxed sound character associated with NOS. Just the opposite. In addition, analog filtered image-band DACs appear to retain the classic NOS character while suppressing potential image-band related IM products. I propose it's removal.
C) ALTERED JITTER IMPACT
=============================================
5) Different jitter impact due to fewer D/A conversion cycles per second.
6) Reduced supply and ground noise due to slower clock rates.
It seems that all of Category C can also be eliminated from significant suspicion. The results of the PGGB, 88.2KHz upsampling experiment indicate, that the OS interpolation-filter design, and not sample-rate related jitter is primarily responsible for the subjective difference between NOS and OS. I propose it's removal.
D) SAMPLE-PERIOD RELATED QUANTIZATION ERRORS
==============================================
7) Converter settling-time becomes a smaller percentage of each conversion period as the conversion rate is made slower.
8) Harmonic-distortion may be sample-rate dependent.
It seems that all of Category D can also be eliminated from significant suspicion for the same reason that Catagory C could be eliminated. The results of the PGGB, 88.2KHz upsampling experiment indicating that the OS interpolation-filter design, and not sample-period related quantization errors are responsible for the subjective difference between NOS and OS. I propose it's removal.
Which is not to suggest that there isn't opportunity for improvement in this area, whether for NOS, or for OS. Such as, circuit settling-time error mitigation. There are two primary means to mitigate this error, RC filter it, or to utilize a Sample-and-Hold circuit.
Please feel free, to post your thoughts and comments.
I think that this would be a good time for another review of our suspect list. Here it is, with my comments in italics:
B) RECONSTRUCTION/IMAGE-BAND HANDLING
===============================================
2) Lack of an FIR interpolation-filter, freeing the DAC from certain processing 'artifacts' , such as:
a) time-domain signal echoes produced within Equiripple on-chip FIR filters.
b) impulse response ringing (pre or post)
c) half-band filters plainly violating Nyquist
d) are prone to clip on peak sample normalized recordings - the intersample overshoot issue.
e) purely analog image-band suppression inherently sounds different than digital suppression?
B2a, B2b and B2c - The investigation in to suspect Category group B2 in general has seemingly resulted in identification of the guilty sub-system ('typical' OS interpolation-filters) within an DAC which causes NOS and OS to sound characteristically different, even though not the specific mechanism responsible. Therefore, these three items will remain unverified possible root cause mechanisms. The PGGB 88.2KHz upsampling experiment indicates that the closer the oversampling comes to a theoretically ideal SINC-function, the more the subjective sound of OS becomes like NOS. Until it sounds close enough to eliminate the subjective difference between the two.
B2d can certainly be a problematic issue if prevalent enough. It's magnitude of impact when only minimally present is not clear to me. To truly isolate it's potential subjective impact would seem to require an specific experiment which offers a choice between a file with inter-sample-overs, and that same file having those removed via level reduction, prior to the interpolator block. Then equalizing the volume level of the second file so that it matches that of the file with the inter-sample-overs. We will not, however, be performing such an experiment in this thread. Although it would be interesting to see it performed in some other thread.
B2e - Since we've shown that high-performance digital OS image-band suppression can remove the subjective difference between NOS and OS, we've necessarily then also shown that purely analog image-band suppression is not necessary for doing so.
3) Phase-modulation of the baseband signal due to insufficiently suppressed image-bands. In other words, because the signal waveform is not fully reconstructed according to the sampling theorem requirements.
B3 Seems something of a puzzlement. While pedestrian OS interpolation would seem to be sufficient to, subjectively, perfectly reconstruct an audio signal waveform, perhaps more might yet be subjectively extracted as the filter performance is increased in some parameter or two. Therefore, our final experiment will be to test for whether even higher-performance OS image-band handling can produce sound which is subjectively superior to either NOS, or to it's PGGB 88.2 upsampled subjective equivalent. This final experiment will be conducted after Hans returns from holiday. It will subjectively evaluate 176.4KHz PGGB upsampling versus 44.1KHz NOS/88.2KHz upsampled.
4) The unsuppressed image-bands are, somehow, producing audible IM products directly within the ear.
Whether B4 is an audible mechanism or not (the research seems very sketchy), IMD seems highly unlikely to produce the relaxed sound character associated with NOS. Just the opposite. In addition, analog filtered image-band DACs appear to retain the classic NOS character while suppressing potential image-band related IM products. I propose it's removal.
C) ALTERED JITTER IMPACT
=============================================
5) Different jitter impact due to fewer D/A conversion cycles per second.
6) Reduced supply and ground noise due to slower clock rates.
It seems that all of Category C can also be eliminated from significant suspicion. The results of the PGGB, 88.2KHz upsampling experiment indicate, that the OS interpolation-filter design, and not sample-rate related jitter is primarily responsible for the subjective difference between NOS and OS. I propose it's removal.
D) SAMPLE-PERIOD RELATED QUANTIZATION ERRORS
==============================================
7) Converter settling-time becomes a smaller percentage of each conversion period as the conversion rate is made slower.
8) Harmonic-distortion may be sample-rate dependent.
It seems that all of Category D can also be eliminated from significant suspicion for the same reason that Catagory C could be eliminated. The results of the PGGB, 88.2KHz upsampling experiment indicating that the OS interpolation-filter design, and not sample-period related quantization errors are responsible for the subjective difference between NOS and OS. I propose it's removal.
Which is not to suggest that there isn't opportunity for improvement in this area, whether for NOS, or for OS. Such as, circuit settling-time error mitigation. There are two primary means to mitigate this error, RC filter it, or to utilize a Sample-and-Hold circuit.
Please feel free, to post your thoughts and comments.
Last edited:
Although I prefer being right I don't necessarily mind being wrong, nor do I necessarily care the manner that files are manipulated to promote subjectively superior sound. In examining my entries as participant #3 it turns out that none of my entries in comparison to the most correct responses were correct, and also that none of my entries as least correct were correct. Oddly if all selections were reversed they would have been dominantly correct.
Certainly it could be argued that I chose poorly as "he chose poorly" in picking the wrong challis in Raiders of the Lost Ark, though this also begs the question as to what constitutes a "correct" choice. To support a declaration there is a correct choice indicates a prerequisite conclusion of subjective correctness as related to pre and post echoes. Given there is no commentary by participants specific to selections there is no evidence to support selections being made on the basis of pre and post echoes. This makes it unclear were we are at.
Certainly it could be argued that I chose poorly as "he chose poorly" in picking the wrong challis in Raiders of the Lost Ark, though this also begs the question as to what constitutes a "correct" choice. To support a declaration there is a correct choice indicates a prerequisite conclusion of subjective correctness as related to pre and post echoes. Given there is no commentary by participants specific to selections there is no evidence to support selections being made on the basis of pre and post echoes. This makes it unclear were we are at.
Gerrit,
Have no worries. All of us, except #2, were all over the place with our rankings.
It's possible that the listening set-up of participant #2 just happened to be particularly good at revealing the sort of low-level FIR filter echo issues which the experiment was intended to test. In addition, the fact that you utilized an OS DAC, rather than NOS to conduct the test added another layer of echo to that otherwise already injected by Marcel - Including adding echo to the one file in each set with no Marcel injected echo. I suspect that the overwhelming dominance of OS DAC responses was a significant factor in the inconclusive results of the experiment. By the way, I believe that the possibility of unknowingly having reverse ordered your rankings is covered under Hypothesis-1, within Marcel's excellent final report.
Last edited:
In consideration of the above (as to develop some perspective specific to pre and post echo phenomenon) can you describe the nature of files in that respect as considered most incorrect Ken or Marcel? It is considered that if the reverse is true this questions the direction of further investigations.
The echo in the experiment, for which we attempting to detect the threshold of, was artificially injected in to two files of each three file set. Meaning, that one of the three files had no injected echo. Since the injected echo is not part of the original signal, it's presence is undesired. So, a perfect file ranking would have been:
1) The file with no injected echo.
2) The file with the lowest degree of injected echo.
3) The file with the highest degree of injected echo.
Think of the ranking as if it were THD. The lowest degree of added distortion is best, the highest degree the worst.
You preferred the largest echoes on Scarlatti, the medium echoes on Williams, the smallest on Jarret and the largest on Saint-Saens.
We have considered adding a version _0 of each file that would have no injected echo for comparison, but that would have made the number of files to be compared even larger. In any case, if people have a preference for NOS DACs because they don't produce these artificial pre- and post-echoes, we would also expect them to prefer the files without artificial pre- and post-echoes.
We have considered adding a version _0 of each file that would have no injected echo for comparison, but that would have made the number of files to be compared even larger. In any case, if people have a preference for NOS DACs because they don't produce these artificial pre- and post-echoes, we would also expect them to prefer the files without artificial pre- and post-echoes.
What Marcel seems stating is in response to my question, whereupon my preferences where dominantly toward largest echoes in reflection of the selection data I provided. Although the file names might reflect types of music they are considered only as names applied to identify specific files, hence not intended to reflect of any preference for a type of music.
My intent is to go over the files again to verify that my dominant preference is still toward files having largest echoes and from there to determine the cause of my mistake, or of my preferences being potentially modified by the type of DAC used, or of alternative interactions with other electronics generating down the line, or that pre and/or post echoes inherently cause preferential experiences. This could require using NOS DAC's as originally intended in the experimentation.
My intent is to go over the files again to verify that my dominant preference is still toward files having largest echoes and from there to determine the cause of my mistake, or of my preferences being potentially modified by the type of DAC used, or of alternative interactions with other electronics generating down the line, or that pre and/or post echoes inherently cause preferential experiences. This could require using NOS DAC's as originally intended in the experimentation.
@Ken: I have one more proposed suspected suspect:
What about de-emphasis?
This is mostly done in the digital domain combined with the digital filter I think?
So if this is bypassed in a NOS DAC this could lead to clear audible differences.
I have no idea how many CD's have pre-emphasis though....
While Marcel’s test was very well prepared to find out what possible effect echos may have on perceived sound reproduction, one of the problems is that we don’t know the provenance of the original files.
It is for sure that these files have been downsampled to 44.1/16, so already polluted with echos to some degree.
Adding new echos may result in all sorts of unexpected effects.
Added to that the is fact that testers where using OS Dacs causing additional echos, invalidating the outcome of the test to some extra degree, as compared to a situation where the effect of echos can be tested in isolation.
Why I am saying this is because with the first Audacity test everybody was right with my own recorded LP file, which was not the case with the other 3 files.
So I could make another LP recording in 88.2 or even 176.4 and ask ZB to downsample this file to an echofree 44.1/16 with his excellent PGGB software.
Then ask Marcel to add the echos and let these files be played by NOS Dac users and we have isolated echos from all other (partly unknown) side effects.
Is this a proper proposal ?
Hans
It is for sure that these files have been downsampled to 44.1/16, so already polluted with echos to some degree.
Adding new echos may result in all sorts of unexpected effects.
Added to that the is fact that testers where using OS Dacs causing additional echos, invalidating the outcome of the test to some extra degree, as compared to a situation where the effect of echos can be tested in isolation.
Why I am saying this is because with the first Audacity test everybody was right with my own recorded LP file, which was not the case with the other 3 files.
So I could make another LP recording in 88.2 or even 176.4 and ask ZB to downsample this file to an echofree 44.1/16 with his excellent PGGB software.
Then ask Marcel to add the echos and let these files be played by NOS Dac users and we have isolated echos from all other (partly unknown) side effects.
Is this a proper proposal ?
Hans
Last edited: