I have read through the whole thread, and I must say I am puzzled by the way this whole issue is handeled by the industry, but more so how it is handled here in this thread. So I really want to weigh in here.
I have made some tests on my setup, which is currently a Wiim Ultra feeding a Topping E70 DAC feeding an Apollon Nilai Amp. I also posted some of my findings concerning this issue on the Wiim user forum:
https://forum.wiimhome.com/threads/resampling-broken-for-high-loudness-tracks.6464/
For convenience I will post the relevant plots here as well.
My test files:
Intersample_7350_+1.25_dBFS_44k_PCM24_LR.flac
Intersample_7350_0_dBFS_44k_PCM24_LR.flac (reduced volume so that the intersample peaks are at 0dBFS, for reference measurements)
Intersample_11025_+3.01_dBFS_44k_PCM24_LR.flac
Intersample_11025_0_dBFS_44k_PCM24_LR.flac (reduced volume so that the intersample peaks are at 0dBFS, for reference measurements)
I measured the digital SPDIF output of the Wiim.
Starting out without resampling we see that the device has no problems transmitting the stream:
However when resampling is enabled, things go badly astray:
The 500Hz peak is at -50dB and the 14700Hz peak at -45dB! Thats not just hard digital clipping, this is much worse!
Now if that was not bad enough, let's try the 11025Hz file with +3dBFS peaks:
The 3.2kHz peak is at -40db and the 4.5kHz peak is at -50db! This is really horribly bad! Again that is much worse than hard digital clipping, becasue that at least would not produce peaks below the original tone. But here we are with peaks at 3.2kHz and 4.5kHz, just perfectly in region of the highest sensivity of the human ear.
While the Wiim Ultra can be used as a DAC, I have not used it as one here, and the shown measurements are all on the digital output. So this might not be a perfect demonstration of a resampling DAC, but I wanted to point out that there is an issue with intersample peaks, and nobody in the industry gives a damn. Yes, the resampling here might be unnecessary. However, the device has this option, and if so it should behave properly doing that. And no, no change of volume, pre-gain, EQ or whatever will heal this issue if the user enables this option!
But here in this thread arguments have been brought foreward to not even test for this issue. The aruments are basically the following:
You appear to make solving this issue your problem to solve, but solving the issue is actually none of your business, you're only supposed to be measuring how well the industry is in solving the issue. And somebody needs to do that! Because especially with DAC's which are resampling, or do have DSP's, somebody needs to find out if these are actually working as expected.
Also, I want to point out that you start with an assumption of what a solution to the problem might be without any objective tests whatsoever. You write: "If the solution is to reduce dynamic range, it means..". Lets first determine the gravity of our problem before jumping to conclusions, by actually measureing what we're having.
And online databases, used by all..? The very industry which brought us the loudness war, should now to that? When is this going to happen? When the hell freezes over? You're overthinking things by a huge margin. Let's first start out by finding out what the status quo is, by starting to actually mearsure.
And that other argument is even weirder:
And if its the manufacturors pushing you, telling you to use less then maximum valued tests, well then address them with my very next point.
Let me deconstruct the other arguments:
only few tracks actually have intersample peaks: For any given year starting with the mid 80's I can give you a very widespread selling CD album of that time which has intersample peaks. But here the screenshot of just the very first album I could think of, I don't know if its worst or just average for todays music, but it is a very strongly seelling album and it gets streamed a lot:
That 1.172368 is at +1.38dBFS! So that's more than the 7350Hz test tone does!
files which do have intersample peaks, have only few of those distributed throughout the file, so nobody will hear that: Here I have taken a track from that album above. I have converted to 32bits and resampled to 192kHz using sox, and then I loaded that resulting 32bits file in audacity to mark the clippings:
let me zoom in at around 2:15
In the best case a DAC will have enough analog headroom. If resampling is used however, then the best case is that these are all hard digital clippings. If the resampler does funny things due to being overwelmed like my above measurement of the Wiim shows, well, then things are much worse than just hard clippings! Does anyone really want to argue that this is inaudible? Have we come that far to now start arguing that as long as we have just a couple of hard clippings per second things are good enough? Really?
only resampling DAC's have these issues: are we sure that those non-resampling DAC's have enough analog headroom? Perhaps they have not, perhaps they output full 2V or 4V already at 0dBFS. Just as a reminder, +3dB are 40% on top of the max value! So instead of 2V thats now 2.8V and instead of 4V that would be 5.6V! The album I showed would need 17%, thats 2.34V or 4.69V. And what about the amplifier, how will it deal with that? These are actually really high levels, and since the industry is in complete denail and nobody tests for this, I must presume we are currently operating our whole equipment in basically unspecified territory!! Yes, you read correctly, that's exactly how I mean it! How should we know that's fine, if we don't test for it?
And what about those resampling DAC's? Perhaps they have a volume knob which does help, but perhaps they don't. Wouldn't it be nice to know?
the issue is only in specifically constructed test files: well, there it's most obvious and best visible, of course. But that's the case with all test files which are used today. Pretty sure in the past there have been people saying the same about the test files we now routinely use. Or does anyone really like that 1kHz tone so much that they have it in their Tidal playlist? And the multitone with those 20+ tones from 20Hz up to 20kHz, who's really a fan of listening to that one? No, these all are tests produced specifically to learn about the behaviour of our equipment in different edge cases. And what this thread asks for is to add one test track for this specific edge case under discussion.
if we increase the headroom, the industry will fill that by increasing the volume further: obviously thats not possible, since the industry is already at the digial maximum of 0dBFS
I have made some tests on my setup, which is currently a Wiim Ultra feeding a Topping E70 DAC feeding an Apollon Nilai Amp. I also posted some of my findings concerning this issue on the Wiim user forum:
https://forum.wiimhome.com/threads/resampling-broken-for-high-loudness-tracks.6464/
For convenience I will post the relevant plots here as well.
My test files:
Intersample_7350_+1.25_dBFS_44k_PCM24_LR.flac
Intersample_7350_0_dBFS_44k_PCM24_LR.flac (reduced volume so that the intersample peaks are at 0dBFS, for reference measurements)
Intersample_11025_+3.01_dBFS_44k_PCM24_LR.flac
Intersample_11025_0_dBFS_44k_PCM24_LR.flac (reduced volume so that the intersample peaks are at 0dBFS, for reference measurements)
I measured the digital SPDIF output of the Wiim.
Starting out without resampling we see that the device has no problems transmitting the stream:
However when resampling is enabled, things go badly astray:
The 500Hz peak is at -50dB and the 14700Hz peak at -45dB! Thats not just hard digital clipping, this is much worse!
Now if that was not bad enough, let's try the 11025Hz file with +3dBFS peaks:
The 3.2kHz peak is at -40db and the 4.5kHz peak is at -50db! This is really horribly bad! Again that is much worse than hard digital clipping, becasue that at least would not produce peaks below the original tone. But here we are with peaks at 3.2kHz and 4.5kHz, just perfectly in region of the highest sensivity of the human ear.
While the Wiim Ultra can be used as a DAC, I have not used it as one here, and the shown measurements are all on the digital output. So this might not be a perfect demonstration of a resampling DAC, but I wanted to point out that there is an issue with intersample peaks, and nobody in the industry gives a damn. Yes, the resampling here might be unnecessary. However, the device has this option, and if so it should behave properly doing that. And no, no change of volume, pre-gain, EQ or whatever will heal this issue if the user enables this option!
But here in this thread arguments have been brought foreward to not even test for this issue. The aruments are basically the following:
- only few tracks actually have intersample peaks
- files which do have these, have only few of those distributed throughout the file, so nobody will hear that
- only resampling DAC's have these issues
- the issue is only in specifically constructed test files
- if we increase the headroom, the industry will fill that by increasing the volume further
What are we comparing here? Loss of 3db dynamic range for DAC's which nowadays all have easily 120+db of it? Who cares if a 120db DAC will only have 117db of dynamic range? It would still be in the Excellent region, perhaps not "top of the charts" but isn't the idea of ASR to differentiate, why must everything be captured inside this one number? Anyhow, in return of the 3dB loss we are gaining the possibility of not having these horrible issues shown above. And nobody is arguming that all DAC's should just blindly use a 3db headroom in all circumstances. If the DAC has a volume control which solves the issue, that is perfectly fine as well, if it has enough analog headroom, also fine. But if it has some option to do something (like my Wiim above), and this is then breaks during intersample peaks, then in this case, better decrease the dynamic range. And if 117 is not enough because of DSP and other stuff needing also some headroom, well too bad, than the industry must do better and increase the 120db value such that there is enough headroom.
You appear to make solving this issue your problem to solve, but solving the issue is actually none of your business, you're only supposed to be measuring how well the industry is in solving the issue. And somebody needs to do that! Because especially with DAC's which are resampling, or do have DSP's, somebody needs to find out if these are actually working as expected.
Also, I want to point out that you start with an assumption of what a solution to the problem might be without any objective tests whatsoever. You write: "If the solution is to reduce dynamic range, it means..". Lets first determine the gravity of our problem before jumping to conclusions, by actually measureing what we're having.
And online databases, used by all..? The very industry which brought us the loudness war, should now to that? When is this going to happen? When the hell freezes over? You're overthinking things by a huge margin. Let's first start out by finding out what the status quo is, by starting to actually mearsure.
And that other argument is even weirder:
Why do we have to be careful? We are talking here about a real issue, with audible consequences! Who the heck cares how the audiophiles test their stuff? That's basically the whole idea behind ASR: use test tracks which ARE unnatural in order be able to objectivley measure the limits of our equipment. If the audiophiles are fine with not testing their equipment with edge casey but valid tracks, good for them. They then can continue discribing what they hear on their vinlys of Beethoven as 'thick' and 'thin' and 'natural' and what not. But here in the real world people stream music which is mastered today and that music is unfortunaltely as it is after 30 years of loudness war. We cannot change what we're getting.
And if its the manufacturors pushing you, telling you to use less then maximum valued tests, well then address them with my very next point.
Let me deconstruct the other arguments:
only few tracks actually have intersample peaks: For any given year starting with the mid 80's I can give you a very widespread selling CD album of that time which has intersample peaks. But here the screenshot of just the very first album I could think of, I don't know if its worst or just average for todays music, but it is a very strongly seelling album and it gets streamed a lot:
That 1.172368 is at +1.38dBFS! So that's more than the 7350Hz test tone does!
files which do have intersample peaks, have only few of those distributed throughout the file, so nobody will hear that: Here I have taken a track from that album above. I have converted to 32bits and resampled to 192kHz using sox, and then I loaded that resulting 32bits file in audacity to mark the clippings:
let me zoom in at around 2:15
In the best case a DAC will have enough analog headroom. If resampling is used however, then the best case is that these are all hard digital clippings. If the resampler does funny things due to being overwelmed like my above measurement of the Wiim shows, well, then things are much worse than just hard clippings! Does anyone really want to argue that this is inaudible? Have we come that far to now start arguing that as long as we have just a couple of hard clippings per second things are good enough? Really?
only resampling DAC's have these issues: are we sure that those non-resampling DAC's have enough analog headroom? Perhaps they have not, perhaps they output full 2V or 4V already at 0dBFS. Just as a reminder, +3dB are 40% on top of the max value! So instead of 2V thats now 2.8V and instead of 4V that would be 5.6V! The album I showed would need 17%, thats 2.34V or 4.69V. And what about the amplifier, how will it deal with that? These are actually really high levels, and since the industry is in complete denail and nobody tests for this, I must presume we are currently operating our whole equipment in basically unspecified territory!! Yes, you read correctly, that's exactly how I mean it! How should we know that's fine, if we don't test for it?
And what about those resampling DAC's? Perhaps they have a volume knob which does help, but perhaps they don't. Wouldn't it be nice to know?
the issue is only in specifically constructed test files: well, there it's most obvious and best visible, of course. But that's the case with all test files which are used today. Pretty sure in the past there have been people saying the same about the test files we now routinely use. Or does anyone really like that 1kHz tone so much that they have it in their Tidal playlist? And the multitone with those 20+ tones from 20Hz up to 20kHz, who's really a fan of listening to that one? No, these all are tests produced specifically to learn about the behaviour of our equipment in different edge cases. And what this thread asks for is to add one test track for this specific edge case under discussion.
if we increase the headroom, the industry will fill that by increasing the volume further: obviously thats not possible, since the industry is already at the digial maximum of 0dBFS
Last edited:
