• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

High Resolution Audio?

andreasmaaan

Major Contributor
Patreon Donor
Joined
Jun 19, 2018
Messages
3,295
Likes
2,416
Mqa claims that in old digital records adc had different filters. Then mqa say to the dac wich filter is adapted so that the dac use the same as the adc.
I agree that ideally dac should decode pcm to analog the same way that the adc encode analog in pcm.
Using different filter in adc and dac can lead to hearable differences you think?
Apodising filter claims that it adapts to the filter used on the adc.
Isn’t it clear based on everything that’s been discussed that this claim must be:
  1. bogus in cases where multiple ADCs have been used in a recording, and/or where a single ADC has been used multiple times (which is in fact very common, as in many cases digital recordings are run through analogue effects units in the process of mixing and/or mastering and then re-digitised)
  2. unsubstantiated (MQA has provided no evidence that it actually does this, and there is plenty of evidence that in fact it does not)
  3. even if true, irrelevant in almost all cases, since all competent ADCs use anti-aliasing filters that are transparent in the first place
https://audiophilestyle.com/ca/reviews/mqa-a-review-of-controversies-concerns-and-cautions-r701/
 

mansr

Senior Member
Joined
Oct 5, 2018
Messages
393
Likes
856
Location
Hampshire
Did you know that PCM was made to have the same filter on both ends too.
There is neither a requirement for, nor any benefit from, using the same filter for anti-aliasing (ADC) and anti-imaging (DAC). Although the basic requirement is the same, low-pass at no higher than half the chosen sample rate, the similarities end there. Of course, the ideal would be a perfect low-pass filter in both places, but that's not possible, so approximations are required. The impact of these approximations are actually quite different. At the ADC end, an anti-aliasing filter with poor stopband attenuation will result in frequencies above Nyquist being captured as potentially audible alias frequencies. This is obviously bad, and the damage cannot be undone in any subsequent processing. At the DAC end, the purpose of the filter is to remove images above Nyquist created by the basic D/A conversion stage. Now these images fall in the ultrasonic range. While they are inaudible, we still want to remove them to avoid the possibility of intermodulation distortion in downstream circuitry causing audible artefacts. If the downstream components are well-behaved, we can actually get away without any filter whatsoever (and some even insist on it).

Long story short, any talk of matching filters between DAC and ADC is nothing but bullshit.
 

Blumlein 88

Major Contributor
Joined
Feb 23, 2016
Messages
5,692
Likes
4,693
I'm sorry but I don't think you can kill a DAC with cymbals - only @RayDunzl demonstrated he can kill a DAC with a spoon. :D
You've not yet seen my pizza pan rapped with a spoon recordings. It has small holes so it breaks up into many multiple frequencies extending to many kilohertz. The extra surface area gives it more energy like a cymbal only at higher frequencies. :)
 

Blumlein 88

Major Contributor
Joined
Feb 23, 2016
Messages
5,692
Likes
4,693
There is neither a requirement for, nor any benefit from, using the same filter for anti-aliasing (ADC) and anti-imaging (DAC). Although the basic requirement is the same, low-pass at no higher than half the chosen sample rate, the similarities end there. Of course, the ideal would be a perfect low-pass filter in both places, but that's not possible, so approximations are required. The impact of these approximations are actually quite different. At the ADC end, an anti-aliasing filter with poor stopband attenuation will result in frequencies above Nyquist being captured as potentially audible alias frequencies. This is obviously bad, and the damage cannot be undone in any subsequent processing. At the DAC end, the purpose of the filter is to remove images above Nyquist created by the basic D/A conversion stage. Now these images fall in the ultrasonic range. While they are inaudible, we still want to remove them to avoid the possibility of intermodulation distortion in downstream circuitry causing audible artefacts. If the downstream components are well-behaved, we can actually get away without any filter whatsoever (and some even insist on it).

Long story short, any talk of matching filters between DAC and ADC is nothing but bullshit.
The same on both ends in that it works best if they are steep enough. Half band filters work for instance, but they usually result in some low level artifacts. You get different filtering at the DAC end causing response differences in the reconstructed signal. That doesn't seem optimum or desriable to me. Neither does running the DAC without a filter and hoping everything upstream filters out the imaging well enough. Speakers and your ears usually will, but I see no benefit to doing it that way.
 

Krunok

Major Contributor
Joined
Mar 25, 2018
Messages
2,818
Likes
1,304
Location
Zg, Cro
You've not yet seen my pizza pan rapped with a spoon recordings. It has small holes so it breaks up into many multiple frequencies extending to many kilohertz. The extra surface area gives it more energy like a cymbal only at higher frequencies. :)
Oh, nice, I see we will not get a winner any time soon so I'll just fetch popcorns and enjoy.. :)
 
Last edited:

mansr

Senior Member
Joined
Oct 5, 2018
Messages
393
Likes
856
Location
Hampshire
The same on both ends in that it works best if they are steep enough.
Only in the sense that the ideal for both is a perfect low-pass filter, but that can never exist. There is nothing to be gained from having the unavoidable imperfections be equal at both ends. Regardless of the filter used at the ADC end, the final output will be better the closer the reconstruction filter is to the ideal, and conversely.
 
Joined
May 29, 2019
Messages
63
Likes
11
Location
Bavaria, near lake Ammersee
What happens if you go without filter can be studied with the help of foobar converting DSD (or SACD ISO) with, or without the 30KHz filter engaged in the SACD plug-in. From what I learned, the unfiltered sounds maybe a nuance "brighter" or "harsher", but not "cleaner", so the choice is obvious for me (even though the filter´s ringing at the track begin and end needs to be cured - manually!?). My take from that experiment: Amplified signals with high noise outside the hearing band do have an audible influence and should be filtered properly, here I think the 30KHz LP could be good choice
 
Last edited:

maty

Addicted to Fun and Learning
Joined
Dec 12, 2017
Messages
602
Likes
257
Location
Tarragona (Spain)
Yes, better sound with Direct (64fp, 30 kHz lowpass) if you want PCM. My conf:

foobar2000-Tools-SACD-plugin.png


Months ago I finally managed to make SACD ISO sound better with JRiver MC v24. SACD plugin is a great add-on.
 
Joined
May 29, 2019
Messages
63
Likes
11
Location
Bavaria, near lake Ammersee
Yes, better sound with Direct (64fp, 30 kHz lowpass) if you want PCM. My conf:

View attachment 27497

Months ago I finally managed to make SACD ISO sound better with JRiver MC v24. SACD plugin is a great add-on.
Thanks and good you made this screenshot, it reminded me to check for updates. I was still running on a ´15 dll.

Unfortunately nobody cared so far to tame this filter ringing at the begin and the end of a converted track, pls. see below.
Probably caused from some unadapted boundary conditions?
I remove these clicks manually, but I am really interested, if this was or is still state of the art???

1560249206344.png
 
Joined
May 29, 2019
Messages
63
Likes
11
Location
Bavaria, near lake Ammersee
Aha!...after reading some of the above linked "Recommendation ITU-R BS.1770-4", especially the latter part "Appendix 1 to Annex 2 (informative) ", starting from page 20pdf. This is well eleborated and easy to understand.
I noticed that this really concentrates on intersample peaking.
This was not in my primary focus in my intended analog application, so I was purely focused on the above picture. I think this has accidently conveyed a view with several samples at max. level to me, because I understood that regular peak meters are capable on counting these consecutive samples, and based on threshold number, they will issue a peak alarm.
This is exceeded by this TP-meter´s capabilities (plus other nice measurement features), which is here connected behind the DAC, on the analog line, driving the analog amp, because I use it similarly for the analog Vinyl.
(Why?...For me this is somehow a similar urge, like checking before what I eat...here of course during consumption...)

In general, the applied use-case for the TP-value only, is to judge whether heavy digital clipping or analog distortion is relevant on the analog line. From what I have seen, there are many digital tracks present causing above 0dBfs readings, but in the range below +1dBfs (In terms of distortion to expect, Vinyl is more difficult to judge, but TP gives quite a reasonable view on the level, when stylus tracking loss becomes critical).
This kind of TP reading is OK for me, since I can do (with limited accuracy) cherry picking on different tracks or album versions and have a general view on track properties...

- This raises finally the question, whether intersampling peaks are measureable in such a setup at all, since the DAC did the resampling signal restoration before, my ADC driving the TP-meter running at 96K (well, the DAC-filter roll-off is somewhere around 85KHz) - does anybody have an opinion on that???


I have to try the "killer" 1/4 sampling rate tone....
One addendum to the above from the author:
I think here we can observe the interesting case, when a DAC (ESS9018s) actually generates an output greater than 0dBfs.
Fed by a digitally clipped track, it is probably the DAC´s inherent oversampling implementation, which causes e. g. a +0.7dB overshoot case, as I described it in Post #304
( https://www.audiosciencereview.com/forum/index.php?threads/high-resolution-audio.6525/post-184411 ).
The connected TP-meter would act in this case as a regular peak meter (it has still +4dB analog headroom at 0dB DAC output) and reads the analog +0.7dB overshoot from the DAC.
What do you think, is it the DAC alone, or does it originate from the ADC in front of the TP-meter also?

As a 2nd addendum:
The max. TP-level of the original digital signal (without having entered a DAC, right out of the CM6631a USB/SPDIF converter) is +0.6dBfs. Since my analog calibration accurary is <0.3dB, I would regard the level of +0.7dBfs of the double converted signal as the same...

some impression from a clipped area:
1560260680680.png
 
Last edited:

mansr

Senior Member
Joined
Oct 5, 2018
Messages
393
Likes
856
Location
Hampshire
What happens if you go without filter can be studied with the help of foobar converting DSD (or SACD ISO) with, or without the 30KHz filter engaged in the SACD plug-in. From what I learned, the unfiltered sounds maybe a nuance "brighter" or "harsher", but not "cleaner", so the choice is obvious for me (even though the filter´s ringing at the track begin and end needs to be cured - manually!?). My take from that experiment: Amplified signals with high noise outside the hearing band do have an audible influence and should be filtered properly, here I think the 30KHz LP could be good choice
With DSD64, the modulator noise exceeds the level of typical music content from around 30 kHz and up. You really don't want to be sending that crud to amps and speakers, hence the low-pass filter. Your test is not at all comparable to using a filterless DAC.

To be clear, I am not suggesting that filterless DACs are in any way a good idea. They are not. However, the fact that they nonetheless exist shows that the audible distortions are not too severe.

The point I'm trying to make is that a poor anti-aliasing filter at the ADC causes errors below Nyquist whereas a poor anti-imaging filter causes errors above. Only one of these directly affects the audible range. Unfiltered images can still be harmful through indirect effects like IMD.
 
Joined
May 29, 2019
Messages
63
Likes
11
Location
Bavaria, near lake Ammersee
With DSD64, the modulator noise exceeds the level of typical music content from around 30 kHz and up. You really don't want to be sending that crud to amps and speakers, hence the low-pass filter. Your test is not at all comparable to using a filterless DAC.

To be clear, I am not suggesting that filterless DACs are in any way a good idea. They are not. However, the fact that they nonetheless exist shows that the audible distortions are not too severe.

The point I'm trying to make is that a poor anti-aliasing filter at the ADC causes errors below Nyquist whereas a poor anti-imaging filter causes errors above. Only one of these directly affects the audible range. Unfiltered images can still be harmful through indirect effects like IMD.
Sure, fully understood, it was just meant as an example in what direction it goes, without changing any HW.

The interesting thing though is that the difference in sound perception in this case may come from the ear and not so much through equipment influence (at decent loudness).
I cannot prove it, but I would postulate it´s from nonlinear processing in the human hearing system (sort of frequency downmix from ultrasonic to the hearing range)...anybody have an opinion on that?
 

Calexico

Active Member
Joined
May 21, 2019
Messages
266
Likes
16
Bob Stuart is an authority, no doubt. When he speaks one would normally listen. The thing about MQA is that it seems to be a solution in search of a problem.
I have been in the IT business for a long time. I have seen 64 Kb/s CSU/DSU labelled as "High-Speed" ... SIXTY FOUR K-I-L-O_B-i-t-s per SECOND. Those were the days when something called a T-1 which at the torrid pace of 1.544 Mb/s was reserved to corporations at over $1,000.00/month... There was also a T-3 which was reserved for the very largest enterprises in the world at 45 Mb/s... well over $3,000.oo a month 20 years ago... We are not there now, no one would dare claim 3 Mb/s as "fast"... So what's the use for MQA when and where PCM Redbook is transparent and available? ... What does it bring that PCM doesn't? Less storage? Not really an issue in this world of 10 TB HDD @ $200... And when we factor that most people, cannot reliably distinguish 320 mp3 or whatever codec you care to name to CD? Now let's move to the audiophile who wants his files bit perfect ? He has Redbook or if he (mostly HE) fancies it ... He'd move to the so-called Hirez, those that do not require special hardware and software to process their files which I believe MQA does ... So the more one looks a MQA the less you understand its value proposition ... Perhaps the same fate as DVD-A?
Same with dsd format.
 
Joined
May 21, 2019
Messages
266
Likes
16
Good morning. I've got a question.
@amirm @Blumlein 88
lot of people doing oversampling with pc.
That mean the filter cut off freq of the dac will be at a higher frequency than around 20khz.
Why not making thd noise test @30khz to see if the dac is good for high res file?
And to see if some ultrasonic noise is more present than with the cut off frequency at 44.1 khz samplerate which is around 20khz.
Maybe with high res there is more ultrasonic noise than with cd format? Can it interract with amp or with frequencies in the audible range?
That would mean that it's better to leave the dac at 44.1khz to keep cut of freq around 20khz.
I hope i'm clear it's hard to explain my question.
I felt you miss my questions
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
1,416
Likes
2,197
Location
The Neverlands
When you oversample a 44/16 recording the oversampling algorithm applies a sharp filter so the content will still have nothing but some left over artifacts of that filter in it above 20kHz till about Nyquist of the upsampled frequency + whatever garbage higher than that.
When someone is using a DAC that has poor post filtering, or one that starts to drop off audibly before 15kHz when it is fed a 44.1 kHz file will have the benefit of the filter in the DAC being at a higher frequency and even when the slope is not steep that won't affect the audible band any more.
In this case the difference between upsampled and native 44/16 reproduction will be audible and thus beneficial to upsample.
Some, however, prefer the early roll-off in which case it isn't beneficial.

During tests 44.1 is used because that's what people listen to most and is where the DAC performs the worst as the filters are near the audible limits.

It is quite easy to test at 192kHz which would show similar 'effects' as when testing at a higher bitrate but at a higher frequency.
Reproducing 40kHz is easy for a DAC that can convert 192kHz files and ... really does not matter as we can't hear it so 'accuracy' of a waveform in that freq. range is not important at all. It is important to be accurate from 40Hz to 10kHz.
 
Last edited:

Blumlein 88

Major Contributor
Joined
Feb 23, 2016
Messages
5,692
Likes
4,693
Good morning. I've got a question.
@amirm @Blumlein 88
lot of people doing oversampling with pc.
That mean the filter cut off freq of the dac will be at a higher frequency than around 20khz.
Why not making thd noise test @30khz to see if the dac is good for high res file?
And to see if some ultrasonic noise is more present than with the cut off frequency at 44.1 khz samplerate which is around 20khz.
Maybe with high res there is more ultrasonic noise than with cd format? Can it interract with amp or with frequencies in the audible range?
That would mean that it's better to leave the dac at 44.1khz to keep cut of freq around 20khz.
I hope i'm clear it's hard to explain my question.
I felt you miss my questions
That would vary from DAC to DAC. The distortion at the higher frequency is sometimes higher. Usually the difference is small. IMD a little more so with some DACs. That is usually from the analog output circuitry rather than being a property of the DAC chip.

What varies more is the noise floor. Some DACs at 44khz will let the noise floor rise not too far above that. Switch to 96 khz and the noise floor doesn't rise until a higher frequency than at 44 khz. Yet others have about the same noise floor output at both sample rates. Some keep the noise floor low at all the sample rates.

You also might want to remember at 30 khz, the 2nd harmonic is 60 khz and the 3rd harmonic is 90 khz and rarely will you have high levels at 30 khz.

Yes some DACs have a noise floor that would prevent full resolution at higher frequencies in terms of even 16 bits. Others have a fairly flat noise floor.

The worst recent one I've seen has a noise floor of -50 db above 50 khz. I don't think that will intermodulate down audibly and that level is unlikely to be a problem for following gear. In the past some DACs had some moderately high level idle tones up higher in frequency, but those are very uncommon to my knowledge since the early 2000's.

So I don't know if I answered your question. Here are a couple of sweep measurements I had on file.


1560418096690.png


With the settings I'm using the background goes to gray at - 100db. In the upper pane, the light blue line is 3rd harmonic distortion which is around -91 db. This was with a -1 db output. All other distortions are less than -100 db and don't show up. The red line is the slow sweep. The left hand side of the pane is with the DAC running at 48 khz and the right hand side running at 96 khz. The second sweep goes to 40 khz, but the ADC was recording at 192 khz so any signal up to 96 khz would be shown. You see the blue haze in the top of that upper pane. That is noise at around -85 db, but that was actually in the ADC not in the DAC. So nothing much at all is above -100 db. You might not see it so easy in my graph, but distortion does drop below -100 db around 6 khz and less on the sweep. It does this at both 48 and 96 khz sample rates.

The lower pane I accidentally recorded at - 8db on another DAC. The horizontal blue line is an idle tone at 31 khz in that DAC which is at something like -95 db. All harmonic distortion is well below -100db. Some would show up around -100 db if that DAC was operating at max output. In this case it shows that until you get very high ultrasonic levels there is no distortion above the -100 db floor of this graph. Even up to 40 khz.
 

Attachments

Last edited:
Joined
May 21, 2019
Messages
266
Likes
16
When you oversample a 44/16 recording the oversampling algorithm applies a sharp filter so the content will still have nothing but some left over artifacts of that filter in it above 20kHz till about Nyquist of the upsampled frequency + whatever garbage higher than that.
When someone is using a DAC that has poor post filtering, or one that starts to drop off audibly before 15kHz when it is fed a 44.1 kHz file will have the benefit of the filter in the DAC being at a higher frequency and even when the slope is not steep that won't affect the audible band any more.
In this case the difference between upsampled and native 44/16 reproduction will be audible and thus beneficial to upsample.
Some, however, prefer the early roll-off in which case it isn't beneficial.

During tests 44.1 is used because that's what people listen to most and is where the DAC performs the worst as the filters are near the audible limits.

It is quite easy to test at 192kHz which would show similar 'effects' as when testing at a higher bitrate but at a higher frequency.
Reproducing 40kHz is easy for a DAC that can convert 192kHz files and ... really does not matter as we can't hear it so 'accuracy' of a waveform in that freq. range is not important at all. It is important to be accurate from 40Hz to 10kHz.
For exemple for those who convert pcm to dsd the filter has cut off frequency at 20khz? Because the dac will have different cut off frequency.
Indeed when you oversample the cut off frequrncy is always the native fs/2?
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
1,416
Likes
2,197
Location
The Neverlands
For exemple for those who convert pcm to dsd the filter has cut off frequency at 20khz? Because the dac will have different cut off frequency.
Indeed when you oversample the cut off frequency is always the native fs/2?
Indeed. That's how oversampling works. New values between the already present samples are calculated. No higher frequencies in between the samples are generated/recovered/invented.
The only benefit it can have is mentioned in my earlier post. When poorer or no filters are used the filter frequency of the DAC itself is higher and doesn't affect the audible range any more.
Also DAC's with a gentle filter that would show massive amounts of energy above 22kHz would 'shift' that garbage (as the upsampling algo usually has a sharp filter) moves all the garbage that DAC produces to above the nyquist of the upsampled frequency or in the case of DSD into the rising noise.
 
Top Bottom