Everyone, please refrain from directly posting your assessments of the 176.4 files to the thread until after the conclusion of the experiment, so as not to influence those still conducting the evaluation. Instead, please PM your preference report and related observations to Hans Polak. While I seen several assessments posted to the thread, I don't believe that Hans has received any preference reports by PM yet. All listening preference reports are appreciated, so please try to participate. 🙂
After Hans receives enough reports and posts their tabulated totals, your open discussion of what you heard is very much encouraged. Thanks.
Last edited:
It was meant as meta communication about the experiment. But ok, I get your point and will refrain from further posting until the experiment has ended... ;-)
Tell us what YOU heard
Everyone who has yet to conduct the 176.4 experiment, please do so at your earliest convenience and send Hans your preference for each of the four files by PM. We've seen some interesting initial reactions posted to the thread, so join in and let Hans know what YOU heard. Your preference reports are much appreciated.
Everyone who has yet to conduct the 176.4 experiment, please do so at your earliest convenience and send Hans your preference for each of the four files by PM. We've seen some interesting initial reactions posted to the thread, so join in and let Hans know what YOU heard. Your preference reports are much appreciated.
Without revealing files, has anyone experienced that the best files are the ones you can turn up the loudest.
That's an interesting observation. It seems like a useful comparative evaluation tip.
Without referring to our 176 experiment, I have also noticed such an effect occurring between OS and NOS playback. I hear the characteristic difference between them as becoming rapidly more obvious as the volume level is raised. I would think that this is due to a difference in various audible error artifacts. The files/tracks with fewer artifacts can be played more comfortably at a higher volume. Or, can be more comfortably listened to for a longer period at the same volume setting.
Last edited:
I would strongly fight the misconception by some that a NOS Dac is doing something wrong, is violating Nyquist or whatever.
A perfect Nos Dac does nothing but exactly translating digital input to analogue output, keeping the complete frequency content up to Fs/2 100% intact, where other PCM Dac principles may come very close but never achieve the mathematical 100%.
But because of the zero order hold, many images and mirror images are created at multiiples of Fs, directly starting as from Fs/2.
Either you don’t care because you can’t hear these ultrasonics, or you want a clean area above the primary audio frequency content for your dog, your speakers or whatever reason.
To get a clean area, you will have to upsample and filter the unwanted images because without upsampling it’s close to impossible to filter properly in the analogue domain after D/A conversion.
As we have seen so far in this thread, this filtering is not a step that should be underestimated, because filters might introduce anomalies in the passband that can change the nature of the sound to such a degree that it becomes audible.
When filtering, the usual bread and butter commercial filters are mostly not good enough.
That’s where ZB with his exquisite PGGB software came to assist us.
But going to a higher Fs using an exeptional good filter can also bring new problems.
When the settling during the ZOH time is flawed (because of slew rate limits, a low quality power supply or other non-linear problems), the flawed percentage of it’s contribution doubles for every doubling of Fs.
So when playing an upsampled and perfectly filtered 176.4/24 from an original 44.1/16 file, design flaws can make it sound worse compared to the original.
No way the Dac principle, either NOS or OS can be blamed for this.
Hans
P.s. I still forgot to mention the frequency droop, a linear process that comes with a ZOH filter, which can be compensated.
A perfect Nos Dac does nothing but exactly translating digital input to analogue output, keeping the complete frequency content up to Fs/2 100% intact, where other PCM Dac principles may come very close but never achieve the mathematical 100%.
But because of the zero order hold, many images and mirror images are created at multiiples of Fs, directly starting as from Fs/2.
Either you don’t care because you can’t hear these ultrasonics, or you want a clean area above the primary audio frequency content for your dog, your speakers or whatever reason.
To get a clean area, you will have to upsample and filter the unwanted images because without upsampling it’s close to impossible to filter properly in the analogue domain after D/A conversion.
As we have seen so far in this thread, this filtering is not a step that should be underestimated, because filters might introduce anomalies in the passband that can change the nature of the sound to such a degree that it becomes audible.
When filtering, the usual bread and butter commercial filters are mostly not good enough.
That’s where ZB with his exquisite PGGB software came to assist us.
But going to a higher Fs using an exeptional good filter can also bring new problems.
When the settling during the ZOH time is flawed (because of slew rate limits, a low quality power supply or other non-linear problems), the flawed percentage of it’s contribution doubles for every doubling of Fs.
So when playing an upsampled and perfectly filtered 176.4/24 from an original 44.1/16 file, design flaws can make it sound worse compared to the original.
No way the Dac principle, either NOS or OS can be blamed for this.
Hans
P.s. I still forgot to mention the frequency droop, a linear process that comes with a ZOH filter, which can be compensated.
Last edited:
It doesn't do a very good job at reconstructing the waveform at the anti-aliasing filter output, though.
Marcel,
Your comment is a bit cryptic.
Interesting to know what anti alias filter you are referring to and what can alias at the output of a Dac.
Hans
Your comment is a bit cryptic.
Interesting to know what anti alias filter you are referring to and what can alias at the output of a Dac.
Hans
Above is from your post of #1368
This is what I was getting at. An image that is overly sharp can be deliberately blurred (as you have shown) to cause our consciousness to refocus on the whole of the composition, this as opposed to leaving our consciousness focused on the sharpness of each individual pleat on the sail, being contrasted with each other pleat. The sharpness can destroy the foundation of what it was intended to be, in the variant contrast with other elements of the composition.
Deliberate blurring or obscuration, as what MQA seems intentionally doing and to much lesser degree as seemingly an unintentional artifact of oversampling, can be of advantage under circumstances of overly sharpened recordings or overly sharpened playback equipment. This suggests that under circumstances of perfect recordings and perfect playback equipment 44KHz NOS seems would be best.
I have considerably less incite into digital systems in comparison to most, including Marcel. Notwithstanding such limitations it is questioned if NOS 44KHz artifacts being suggested as potentially problematic by Marcel are correctable by 88KHz upsampling (or other mechanisms) and if so would this increase transparency or increase blurring?
Well, you know the sampling theorem: when a bandlimited signal is sampled at a rate greater than twice the highest frequency in the signal, one can in principle perfectly reconstruct the signal from the samples. In real life there are always imperfections, but still, you can get close to the original bandlimited signal. In practice one always uses a low-pass filter during recording to enforce a bandwidth limitation and in practice it is often split up into several stages, some analogue and some digital, but that doesn't change the principle.
An audio DAC normally tries to reconstruct the bandlimited continuous-time waveform that was sampled during recording. A NOS DAC with only zero order hold reconstruction does a particularly bad job at that. Of course you can argue about whether that matters and under what conditions it matters, but still.
When aliasing happens during the recording process, all relevant content is by defintion below Fs/2 including these eventual aliasing products. No Dac can repair this unwanted aliasing, either NOS or OS.
And that a NOS Dac does a bad job in reconstruction from a (technical) point of view has only to do with the images directly starting as from Fs/2 and upwards. That may be seen as unwanted.
But with restricted information gathered in this thread, NOS Dac users heard no difference between 44.1 and 88.2, indicating that images starting from 22.05Khz did not interfere with the sound perception in their situation.
So using terms like oversharpened sound by images we can’t hear is fully unconfirmed and is greatly confusing people.
But further in the reproduction chain, speakers could be driven by high levels of ultrasonic sound into non linear behaviour resulting in harsh sound because of IM in the audio range.
To even prevent this from happening, the only remedy is to use oversampling.
Hans
And that a NOS Dac does a bad job in reconstruction from a (technical) point of view has only to do with the images directly starting as from Fs/2 and upwards. That may be seen as unwanted.
But with restricted information gathered in this thread, NOS Dac users heard no difference between 44.1 and 88.2, indicating that images starting from 22.05Khz did not interfere with the sound perception in their situation.
So using terms like oversharpened sound by images we can’t hear is fully unconfirmed and is greatly confusing people.
But further in the reproduction chain, speakers could be driven by high levels of ultrasonic sound into non linear behaviour resulting in harsh sound because of IM in the audio range.
To even prevent this from happening, the only remedy is to use oversampling.
Hans
What seems wasn't made clear is that variant selections of recorded pieces, each have characteristics having been recorded and presented, are largely variant when played back at 44.1. The sharpening or over-sharpening of those pieces where helped by mechanisms of moderate blurring at 88.2 to increased blurring at 176. I
In the case of a sharpened image it is questioned as to how that occurs. As far as the image goes this can be considered an increase in dynamic contrast, whereupon the lines in the folds of the sail can appear with greater width and hence darker and more prominent than the space between folds.
In order to create a more definitive test some live recordings ought to be made from a microphone feeding a splitter that simultaneously records at 44KHz and at 88KHz, perhaps with the same brick wall filtering just below the Nyquist of each sampling frequency.
The comparison is then between a 2x OS 44KHz and the alternative NOS 88KHz. Hence the outcome is between an NOS 88KHz and an OS 88KHz DAC. It is questioned if these would cause the same outcome and if these would still be not as good as NOS 44KHz
In the case of a sharpened image it is questioned as to how that occurs. As far as the image goes this can be considered an increase in dynamic contrast, whereupon the lines in the folds of the sail can appear with greater width and hence darker and more prominent than the space between folds.
In order to create a more definitive test some live recordings ought to be made from a microphone feeding a splitter that simultaneously records at 44KHz and at 88KHz, perhaps with the same brick wall filtering just below the Nyquist of each sampling frequency.
The comparison is then between a 2x OS 44KHz and the alternative NOS 88KHz. Hence the outcome is between an NOS 88KHz and an OS 88KHz DAC. It is questioned if these would cause the same outcome and if these would still be not as good as NOS 44KHz
Last edited:
Can we please stop talking about non existing sharpening, over-sharpening and the even more far fetched non sense of blurring at 176.4
Things are getting mixed up and instead of blaming a non ideal FIR filter in an OS Dac as the reason for a sound difference, a new kid on the block was invented and called “NOS sharpening” instead.
As an extreme example of how silly this sharpening theory is, George had the guts to confirm that he heard no difference between two 44.1 files, one with no content above 5.5Khz and the other filled with images starting from 5.5Khz.
When it’s already hard to tell the difference with images starting at 5.5Khz for the majority of us probably being older then 40, it is extremely unlikely that anybody can tell the difference with images starting 4 times higher at 22Khz.
This of course assuming that speakers having to cope with ultrasonics are not interfering with IM, which is a whole different issue.
Hans
Things are getting mixed up and instead of blaming a non ideal FIR filter in an OS Dac as the reason for a sound difference, a new kid on the block was invented and called “NOS sharpening” instead.
As an extreme example of how silly this sharpening theory is, George had the guts to confirm that he heard no difference between two 44.1 files, one with no content above 5.5Khz and the other filled with images starting from 5.5Khz.
When it’s already hard to tell the difference with images starting at 5.5Khz for the majority of us probably being older then 40, it is extremely unlikely that anybody can tell the difference with images starting 4 times higher at 22Khz.
This of course assuming that speakers having to cope with ultrasonics are not interfering with IM, which is a whole different issue.
Hans
Last edited:
Some audio circuits can perceptually sound more or less like they are under-sharpened/under-focused, optimally-focused, or over-focused; such circuits can be completely analog. In addition, some dacs sound different at different sample rates; it may have nothing to do with possible perception of ultrasonics.
Moreover, making a claim that it takes 'guts' not to hear a difference when it hasn't been fully established that a difference can't be heard by anyone else on any other system seems more like an expression of frustration verses a statement of fact.
Moreover, making a claim that it takes 'guts' not to hear a difference when it hasn't been fully established that a difference can't be heard by anyone else on any other system seems more like an expression of frustration verses a statement of fact.
Last edited:
I think it take guts to confess that you can’t hear a difference when you were supposed to hear a difference in a test that was meant to illustrate a supposed sharpening effect from images after lowering Fs/2 from 22Khz to 5.5Khz.
I didn’t see any comment from your side confirming or denying any such effect 😀
Hans
I didn’t see any comment from your side confirming or denying any such effect 😀
Hans
32-bit@176.4kHz Files To Be Released
Let's all pause and take a breath for a minute or two. We're nearing the end of a long and productive thread, which all of you have very patiently supported and contributed to. You all have my thanks! 🙂
===============================================================================
I'm announcing an unplanned action regarding the 176 experiment. All of the 176 listening comments posted to the thread (I count, three) have not only been negative regarding the sound of the 176 files, but very negative. That unanimous reaction, coupled with the lack of test reports sent to Hans, leads me to suspect that something may have been inadvertently amiss during the 24-bit conversion. You may recall that I had mentioned that there was a delay in releasing the 176 files due to Hans and I having to select the re-quantized bit-depth of those files, and which was more involved than we had imagined. We settled on 24-bits, which is what you are listening to now. What I didn't mention is that ZB (the author of the PGGB), also produced a 32-bit conversions of those same files. So, what we're going to do is release that 32-bit conversion in a new Dropbox. Yes, we are aware that no DAC can accurately resolve a 32-bit sample, nor a 24-bit sample for that matter. We are simply checking for an audible difference in how the two conversion were performed which may have affected the higher bits.
Obviously, this interrupts our current 176 experiment, but as of now, something appears that it may be wrong with the 24-bit conversion, based on the highly negative reactions we've seen. At worst, the 176.4 PGGB upsampled files should only be no subjective improvement over the 88.2 PGGB upsampled files, but not sound worse. As Hans posted about, a possible explanation for a given negative 176 report versus the sound of the 88 files, is that it's due to non-linear errors stemming only from the increased sample-rate itself, and not having anything to do with the conversion process. However, the fact that we've seen multiple strongly negative 176 reports makes me wonder whether something is off about the 24-bit conversion.
As a result of that concern, we will be placing the 32-bit files in to a Dropbox in the next day or two. If you are having a negative reaction to the 24-bit files, please listen to the 32-bit files. If you don't notice an obvious improvement, then we will know that the 24-bit conversion is not the problem.
Let's all pause and take a breath for a minute or two. We're nearing the end of a long and productive thread, which all of you have very patiently supported and contributed to. You all have my thanks! 🙂
===============================================================================
I'm announcing an unplanned action regarding the 176 experiment. All of the 176 listening comments posted to the thread (I count, three) have not only been negative regarding the sound of the 176 files, but very negative. That unanimous reaction, coupled with the lack of test reports sent to Hans, leads me to suspect that something may have been inadvertently amiss during the 24-bit conversion. You may recall that I had mentioned that there was a delay in releasing the 176 files due to Hans and I having to select the re-quantized bit-depth of those files, and which was more involved than we had imagined. We settled on 24-bits, which is what you are listening to now. What I didn't mention is that ZB (the author of the PGGB), also produced a 32-bit conversions of those same files. So, what we're going to do is release that 32-bit conversion in a new Dropbox. Yes, we are aware that no DAC can accurately resolve a 32-bit sample, nor a 24-bit sample for that matter. We are simply checking for an audible difference in how the two conversion were performed which may have affected the higher bits.
Obviously, this interrupts our current 176 experiment, but as of now, something appears that it may be wrong with the 24-bit conversion, based on the highly negative reactions we've seen. At worst, the 176.4 PGGB upsampled files should only be no subjective improvement over the 88.2 PGGB upsampled files, but not sound worse. As Hans posted about, a possible explanation for a given negative 176 report versus the sound of the 88 files, is that it's due to non-linear errors stemming only from the increased sample-rate itself, and not having anything to do with the conversion process. However, the fact that we've seen multiple strongly negative 176 reports makes me wonder whether something is off about the 24-bit conversion.
As a result of that concern, we will be placing the 32-bit files in to a Dropbox in the next day or two. If you are having a negative reaction to the 24-bit files, please listen to the 32-bit files. If you don't notice an obvious improvement, then we will know that the 24-bit conversion is not the problem.
Last edited: