Let's develop an ASR inter-sample test procedure for DACs!

[John_Siau]

In the paper, you write:

How did you determine that in most cases the volume control is located after the interpolator?

The ESS ES9039 chips have the volume control before the interpolator as well as the AKM AK4499. I mentioned this in this thread here.

Golden Sound uses a fs/4 sine at 45 degree up to +3.01 dBFS at 44.1 kHz sample rate such as in the test of your DAC3.

Archimago tests using white noise with 0 dBFS as well as -4 dBFS peaks.

Are those not interpolator overload tests?

Many popular Analog Devices DSP chips have internal dedicated fixed-point ASRC resources on all of the digital inputs. These do not have to be turned on, but many developers choose to use them because they allow a constant sample rate in the DSP core. Filter coefficients can be set up once and then they will work for all incoming sample rates. Since the volume control would also be implemented in the DSP, this ends up following the ASRC by default. Consequently, the most likely implementation is to have a fixed-point ASRC in front of the digital volume control!

miniDSP has fixed this problem in their SHD devices, but according to them, the fix was not trivial.

Yes, Golden Sound and Archimago are publishing 0dBFS+ tests, but ASR, Stereophile (and most others) are not publishing these tests. It is also rare to see anything in the manufacturer's specifications. Finally, we don't even have appropriate 0 dBFS+ test tones built into the AP test stations. There are many digital playback devices that have 0 dB digital headroom.

We have work to do. We have tests to run, equipment to fix, and music to enjoy!

We need to start testing.

 

[pjug]

Dealing with this issue playing CDs can also be handled by using a CD transport and sending the digital signal to something that can create headroom. For example WiiM devices can do this if you use the volume limit (this is applied before it does any other DSP). What they call 95% is -3dB. 90% is -6dB.

I just wanted to follow up on this because I now see that when I say it applies the headroom before additional DSP this means Wiim doesn't even increase bit depth first. So you lose LSB if you do 95% volume limit. I don't see any way of adding hedroom with increased bit depth before applying their EQ either. It seems their DSP is a work in progress and not really very good at this point.

 

[Scytales]

I have updated the measurements I published in post #580 with a new device : a Sony SCD-XA9000ES.

Please refer to post #580 for comments about the other players I had tested.

The Sony SCD-XA9000ES is a multichannel SA-CD/CD player. It uses DAC chips with a built-in interpolation filter. The DAC offers 2 different types of digital filter for CD replay. Obviously, the DACs have great if not perfect headroom with any of the 3 ISP test files on NTTY's test CD.

The DAC chip used in this player has a Sony designation : CXD9657N. But it has been pointed out by many keen observers for many years that this DAC is actually a Burr Brown PCM1738. In fact, the service manual of another Sony SA-CD player, the SCD-C555ES 5 discs changer, do mention the two designations in some places : either CXD9657 or PCM1738. Now that is interresting, because the Pioneer DV-868AVi (DV-59AVi in the US and DV-S969 in Japan) uses Burr Brown PCM1738, but, contrary to the Sony, the Pioneer have no headroom for ISP as shown in the above table.

So either the DAC with a Sony designation is not entirely identical to the Burr Brown DAC or Sony managed to drive the input of the DACs in a way more suitable to handle ISP, for instance by digitally attenuating the signal sent to the DACs.

Edition 12/14/2024 :
Typo corrected in the table : the SCD-XA9000ES THD+N with optional filter at 11,025 kHz is 0.0029 % and not 0.0028 %.

 

[NTTY]

Interesting, thanks for sharing!
The inter sample overs happen mainly in the interpolator (and the Sample Rate Converter, if any is used). That means Sony ensured sufficient headroom in their DSP as opposed to BB in the internal interpolator of the PCM1738.

 

[peng]

Many popular Analog Devices DSP chips have internal dedicated fixed-point ASRC resources on all of the digital inputs. These do not have to be turned on, but many developers choose to use them because they allow a constant sample rate in the DSP core. Filter coefficients can be set up once and then they will work for all incoming sample rates. Since the volume control would also be implemented in the DSP, this ends up following the ASRC by default. Consequently, the most likely implementation is to have a fixed-point ASRC in front of the digital volume control!

miniDSP has fixed this problem in their SHD devices, but according to them, the fix was not trivial.

Yes, Golden Sound and Archimago are publishing 0dBFS+ tests, but ASR, Stereophile (and most others) are not publishing these tests. It is also rare to see anything in the manufacturer's specifications. Finally, we don't even have appropriate 0 dBFS+ test tones built into the AP test stations. There are many digital playback devices that have 0 dB digital headroom.

We have work to do. We have tests to run, equipment to fix, and music to enjoy!

We need to start testing.

Thanks for that video with Gene, I watched it but may have to re-watch as I don't understand why you seem to think recording/mastering is not the main issue, iirc even the word "debunk" was used on that. It seems logical to me, it is very much the recording/mastering that could make or break this ISP distortions thing (edit: because while the recording/mastering process does not create the distortions, it does allow it to happen later in the playback process, in the sense that if they had allowed for adequate headroom/not exceeding 0dBFS under any circumstances, better still if there is some sort of standard that all can follow, then the playback devices and/or processes would not have to deal with the issue.

I would think that if that part is not done well, such as by hitting true peaks above 0 dBFS during recording/mastering, the resulting distortions (edit: introduced in the playback process would be hard to fix down stream in the audio chain where those DSP devices such as AVRs/AVPs, minidsp devices have to deal with the input signal that has such inter sample peaks exceeding 0 dBFS.) Or I misunderstood you and Gene in that video?

Further edit: Now that I have started to read up on this thread from the beginning, I probably don't need a response to this post any more.

 

[JIW]

Thanks for that video with Gene, I watched it but may have to re-watch as I don't understand why you seem to think recording/mastering is not the main issue, iirc even the word "debunk" was used on that. It seems logical to me, it is very much the recording/mastering that could make or break this ISP distortions thing.

I would think that if that part is not done well, such as by hitting true peaks above 0 dBFS during recording/mastering, the resulting distortions would be hard to fix down stream in the audio chain where those DSP devices such as AVRs/AVPs, minidsp devices have to deal with the input signal. Or I misunderstood you and Gene in that video?

There is no distortion inherent to ISPs. It is the lack of headroom above the digital maximum value in interpolators and converters that is the source of distortion.

 

[peng]

There is no distortion inherent to ISPs. It is the lack of headroom above the digital maximum value in interpolators and converters that is the source of distortion.

Appreciate your response but that's not my intended question, though it made me realize immediately I poorly worded my point, as I really meant to say the distortions resulted by the downstream devices (not the ISP itself) such as dsp, dac etc., would be hard to fix in terms of having to deal with multiple variables, whereas if the recording/mastering simply adhere to allowing for adequate headroom. As an example, minidsp has just fixed their console and FW build for their SHD, and now the is before ASRC so that could fix the ISPD issue, but that's only for the minidsp SHD. For those who have the other models such as the Flex models, I don't know if they would get the fix eventually, or ever..

Thanks again, I need to edit my post so as not to cause further confusion.

 

[NTTY]

There is a recommendation to prevent mastering too close to 0dBFS, but nobody cares as the market still asks for loud masters which are indeed the real cause. And a LOT of them are out there since end of the ‘80s.
So you can either select the masters you listen to (I do), or ask the downstream devices to handle the stupidly loud ones, or both.
Either way, I’m interested to know those devices that handle ISPs better than others.

You can refer to the Nielsen / Lund AES paper, which tried to reinforce good mastering practices suggesting a "Dynamic Range Approval".

 

[JIW]

Appreciate your response but that's not my intended question, though it made me realize immediately I poorly worded my point, as I really meant to say the distortions resulted by the downstream devices (not the ISP itself) such as dsp, dac etc., would be hard to fix in terms of having to deal with multiple variables, whereas if the recording/mastering simply adhere to allowing for adequate headroom. As an example, minidsp has just fixed their console and FW build for their SHD, and now the is before ASRC so that could fix the ISPD issue, but that's only for the minidsp SHD. For those who have the other models such as the Flex models, I don't know if they would get the fix eventually, or ever..

Thanks again, I need to edit my post so as not to cause further confusion.

How could it even be fixed? The original signal is no longer available after processing and distortion could be part of the original signal as well. The exact distortion characteristics for any kind of signal at any degree of overload would have to be known to remove the distortion.

For devices without headroom in the SRC, using a DA/AD step before can introduce sufficient headroom if the DAC can deal well with ISOs. Ultimately, those devices have a flawed design if no headroom can be introduced before the SRC.

How much is adequate headroom? Benchmark used Steely Dan's Two against Nature to illustrate ISOs being far from rare nor a consequence of loud masters. Of course, true peak measurements can be used nowadays to avoid ISOs but that does not remove the ISOs from older recordings.

 

[peng]

How could it even be fixed? The original signal is no longer available after processing and distortion could be part of the original signal as well. The exact distortion characteristics for any kind of signal at any degree of overload would have to be known to remove the distortion.

For devices without headroom in the SRC, using a DA/AD step before can introduce sufficient headroom if the DAC can deal well with ISOs. Ultimately, those devices have a flawed design if no headroom can be introduced before the SRC.

How much is adequate headroom? Benchmark used Steely Dan's Two against Nature to illustrate ISOs being far from rare nor a consequence of loud masters. Of course, true peak measurements can be used nowadays to avoid ISOs but that does not remove the ISOs from older recordings.

I am sort of quoting John, he did say minidsp got it fixed in the SHD. I don't remember if it was in the video or a post on some forums, and say "sort of" because I might not got is exact wording right. My understanding, from that vides is that such devices, or on the software side, the likes of JRiver, may be Room, Foobar as well, allows one to lower the input signal "... before ASRC.." by a fraction of dB to a few dB. I starting using JRiver's internal volume to lower it from the default 100% to 60%, not sure how many dB that would be as I don't know much about the scale they used, but probably at least 3 dB so Steely Dan's or not, I felt a little safer from such distortions.

Admittedly I am not one who could hear the distortions 100% of the time but I probably did sometimes so I'll apply whatever "fix" I know of, regardless. Again, why don't people talk more about what the recording/mastering industry should do, instead of making it a big thing like Gene and John did in that video? With due respect though, I understand may be they both would rather talk about things they know best, like in the hardware/software domain but things not on the recoding/mastering that the so called sound engineers are more familiar with.

Found John’s post about the minidsp’s fix, but apparently only the SHD got the fix.

Post in thread 'Let's develop an ASR inter-sample test procedure for DACs!'

https://audiosciencereview.com/foru...le-test-procedure-for-dacs.49050/post-2149010

 

[JIW]

I am sort of quoting John, he did say minidsp got it fixed in the SHD. I don't remember if it was in the video or a post on some forums, and say "sort of" because I might not got is exact wording right. My understanding, from that vides is that such devices, or on the software side, the likes of JRiver, may be Room, Foobar as well, allows one to lower the input signal "... before ASRC.." by a fraction of dB to a few dB. I starting using JRiver's internal volume to lower it from the default 100% to 60%, not sure how many dB that would be as I don't know much about the scale they used, but probably at least 3 dB so Steely Dan's or not, I felt a little safer from such distortions.

Admittedly I am not one who could hear the distortions 100% of the time but I probably did sometimes so I'll apply whatever "fix" I know of, regardless. Again, why don't people talk more about what the recording/mastering industry should do, instead of making it a big thing like Gene and John did in that video? With due respect though, I understand may be they both would rather talk about things they know best, like in the hardware/software domain but things not on the recoding/mastering that the so called sound engineers are more familiar with.

Implementing adequate practices in production only helps going forward. There are decades worth of possibly problematic recordings for which only the consumer can remedy the issue. Digital attenuation without sample rate conversion is the only way with oversampling DACs.

You can use a sound meter app on your phone or a sound meter if you have one and pink noise to determine which volume setting in JRiver has your desired attenuation. 60% seems a lot to me since it probably non-linear.

 

[peng]

Implementing adequate practices in production only helps going forward. There are decades worth of possibly problematic recordings for which only the consumer can remedy the issue. Digital attenuation without sample rate conversion is the only way with oversampling DACs.

You can use a sound meter app on your phone or a sound meter if you have one and pink noise to determine which volume setting in JRiver has your desired attenuation. 60% seems a lot to me since it probably non-linear.

Agreed, but it would at least stop adding to the potential problem going forward.

 

[pjug]

Agreed, but it would at least stop adding to the potential problem going forward.

Personally I agree with you that it is stupid to release tracks that require creating headroom to be converted properly, and I think this is poor engineering. Others disagree and without any rules about it there is no way to say who is right on that. But as a practical matter doing what you say above (if somehow you were able to make it happen) would only leave us with a slightly less messy situation. So we are stuck having to solve it on the playback side and I don't see much point in resisting that.

 

[peng]

Personally I agree with you that it is stupid to release tracks that require creating headroom to be converted properly, and I think this is poor engineering. Others disagree and without any rules about it there is no way to say who is right on that. But as a practical matter doing what you say above (if somehow you were able to make it happen) would only leave us with a slightly less messy situation. So we are stuck having to solve it on the playback side and I don't see much point in resisting that.

I started reading this thread from the beginning and realized something similar have already been discussed (or even debated) before.

For clarity, let me say that I do like the fact that people like JS and others are focussing on solutions provided on the hardware/software side, but I also feel people on the recording and mastering side should also do what they can going forward, that's all.

 

[Bow_Wazoo]

I would like to find out whether and to what extent the Neutron Player produces IS peaks when oversampling is activated.

How can I check this?

The background is that I use EQ.
@dmitrykos , recommends activating oversampling, as this increases the precision of the EQ calculation.
Since my preamp is already at -5.3dB due to the EQ profile, and with the combination of Arya Stealth and SMSL DL100, I sometimes no longer achieve the desired volume (so I use the player's normalize function). Unfortunately there is no option to reduce this by another 3dB.

I actually discovered a thread about it.

Automatic Gain Protection (AGP) vs Intersample Overs - Neutron Forum

Post by dmitrykos » Mon Jul 11, 2022 11:39 am

> add an optional intersample over detection

It would be excessive load on CPU because core would need to check amplitude of every and the benefit of that is miserable. To my view strategy of manual decrease of Preamp to the negative value of the loudest low-frequency band would be the best approach. Otherwise it is better to Normalize (Peak type) album with such EQ preset to guarantee 0 clipping.

I will consider adjusting RMS's window length to make AGP more reactive to a short bursts of exceeding sample values."

If I understood correctly, intersample peaks are not possible when using the normalize function?!

 

[KSTR]

I would like to find out whether and to what extent the Neutron Player produces IS peaks when oversampling is activated.

How can I check this?

I'm not familiar with that player but can give some general information.
If the original sample stream has "baked-in" ISOs then any upsampling will reproduce it as it must be, leading to sample values above unity. The question is what happens when the upsampled stream is converted to 24-bit integer which what a DAC expects. In the best case it clips the values, and might even be using soft-clipping.
The only way to avoid this static clipping is reducing the volume before conversion into the final 24-bit stream. If this is offline processing, one can find the peak value and reduce volume (normalization). For real-time processing, you can only use an estimate like 6dB or so and reduce the volume by that and hope the best.

 

[Bow_Wazoo]

I use the normalize option obligatorily.
If I understood correctly, I should be on the safe side with it

 

[KSTR]

If I understood correctly, I should be on the safe side with it

Yes. There is mimimal chance, though, that the upsampled and normalized stream can still provoke ISOs (at the new sample rate). This will be very rare and and only very small ISO level, therefore likely irrelevant in practice... dependent on oversampling rate. With only 2x it might actually be a problem (assumed that the DAC cannot tolerate even the smallest ISOs).

 

[Bow_Wazoo]

I'm aiming for 352/384kHz.

 

[Sean_S]

Since these peaks are due to oversampling, I wonder if it plays into (is part of the reason) why some people like the sound of these boutique R2R DACs, which all seem inevitably to be NOS?