Maybe stupid question about Digital PCM Filter and oversampling

[theveterans]

Forgive me for my ignorance. I know next to nothing about sound science, but I do find it fascinating and try to learn what I can from reading. Is it possible that DAC chips and their sometimes multiple filters, which I've read multiple times as being transparent and therefore not responsible for changes in sound, interact differently with the analog stage of the DAC? In other words is the analog stage in any way responsible for the perceived change in sound? Again, I apologize for my ignorance here, just curious. Thanks again!

The analog output of a DAC is usually around 75 ohms and 600 ohms in Balanced XLR. As for analog stage in the DAC, discrete transistors in analog stage can certainly sound different than an op-amp based analog stage. Both will measure beyond the established audibility of an arbitrary ~ 90 dB SINAD as example yet depending on the headphones you use, you can possibly still hear differences with DACs

You think professionals at the manufacturing level care about infinitesimally small signals, the kind that we've said several times now are completely inaudible? You're terminally full of sht.

Professional (design audio engineers to be specific) do care lol. If they don't, I wouldn't get these wonderful sounding DACs but rather sterile sounding, harsh, digital fatigue inducing DACs instead

 

[bigshot]

Wrong again.

 

[castleofargh]

Forgive me for my ignorance. I know next to nothing about sound science, but I do find it fascinating and try to learn what I can from reading. Is it possible that DAC chips and their sometimes multiple filters, which I've read multiple times as being transparent and therefore not responsible for changes in sound, interact differently with the analog stage of the DAC? In other words is the analog stage in any way responsible for the perceived change in sound? Again, I apologize for my ignorance here, just curious. Thanks again!

Anybody can design something poorly enough that it sounds different. The amp section, the filters(if they don't filter enough or filter a lot within the audible range), and just about anything else can be poorly done and have big consequences for the signal. Then it's only a matter of how big they get and when that becomes audible. Making a DAC from scratch is a very complicated matter.

But nowadays, you can pretty much get 2 chips that do it all for you, the input is standardized(mostly USB nowadays), the output also follows standard or at least a rule to maximize voltage accuracy(output somewhere between 50 and 150ohm, going into the amp input that's at least several thousand ohms and nowadays often in the 10kohm).

And of course, a DAC is the most accurate part of our system after the source that ideally does nothing at all to the signal. The amplifier has the real work of operating with a wide variety of loads and voltages. The headphone/IEM has comparatively terrible fidelity, with most measures of fidelity being one or more magnitude worse than what a DAC does. And of course the frequency response will be whatever. You'll probably get more objective difference between 2 pairs of the same headphone, than between 2 DACs made to have high fidelity.

But very often one DAC will output higher voltage, and that could:
1/ make the listener feel like everything is different(bass, details, soundstage, you name it). And even when listeners notice that volume level difference because of how big it is, they still tend to A/B their DAC as is with the volume difference, so they still have those impressions of a bigger imaging or whatever. Nobody, who's ever learned anything about experimentation, would allow a non volume matched setting. But audiophiles and most reviewers "know better", they "have enough experience to not be affected", and all the lies you tell yourself when you try to avoid admitting that you're wrong or when you try to avoid doing something bothersome.
2/ the amplifier used will have a range of nominal input voltage. If it was designed to operate nominally with up to 2 or 2.5V, and you feed it with a DAC that outputs maybe 3V or even more(mostly high voltage is a trick to improve SNR in the DAC), then your amplifier might react very poorly to that high voltage input. It could distort much sooner(at lower power) into a given load, for example.

We can't list all the ways to misdesign and misuse gear. But that doesn't mean at all that it's hard to find very clean devices(except the transducers that have strong mechanical limitations, so even the best ones are much inferior to most cheap DACs in relative fidelity). And it's not that hard to pair products correctly, and to avoid the nonsense filter settings on a DAC.

You see in here some people talk as if any difference was audible. That is obviously pure nonsense! Humans have limited sensory abilities. And for basically all of those studied thresholds, in practice we stop detecting them even sooner with music because our focus is on different things and because the music itself masks a lot of cues that might on their own be noticeable.

 

[gregorio]

There are tons of discussions about if people can hear the difference in DAC filters or not.

It used to be true that no one can hear differences in DAC filters. Then around a decade ago some one decided to use a non-optimal filter (I believe it was in the Pono Player), a minimum phase anti-image filter with a relatively large transition band starting around 10kHz. Subsequently, the DAC chip manufacturers polled their customers looking for some new feature to add (as audible perfection with DAC chips had already been trivial ~20 years prior) and the idea of switchable oversampling (anti-image) filters was born. There are now some really silly filter options available and some of them are certainly audible. However, even the really silly ones are still fairly subtle. This is only really for the audiophile market though, other audio markets (like the pro-audio market) just use the same optimal filters that have been employed for decades (fast roll-off linear phase or an apodizing variation).

Wouldn't oversampling to >=88.2kHz just irrelevant that? Why not just resample everything to >=88.2kHz and don't care?

No. When you upsample you need to employ an anti-image filter to remove freqs above the Nyquist Point of the original file sample rate. In other words if you resample your 44.1kHz content to >=88.2kHz then the process will include an anti-image filter, just as it will if you don’t resample and leave it to the DAC. Hypothetically, your example would actually be worse because you are applying an anti-image filter at 22.05kHz (assuming a 44.1kHz original file sample rate) to upsample to say 88.2kHz and then the DAC will apply another anti-image filter at 44.1kHz to upsample to it’s internal sample rate (384 or 768 for example). Although, it’s not really “worse” per se and it’s certainly not audible.

If you don't filter them the aliases could still appear in the audible range.

No, that would be the case with the ADC process but not with the DAC process. The folding down of the freqs above Nyquist (sideband “images” or other freqs above Nyquist) into the range below Nyquist is called “aliasing” and only occurs during the ADC process. In the DAC process there are no freqs above Nyquist and the “images” do not fold down. However, the images still need to be removed because their high freq content is liable to cause IMD (in the audible range) in the amp and/or transducers. This is why the initial filter in the ADC is called an anti-alias filter, while the initial filter in a DAC is called an anti-image filter (although it’s sometimes also called an interpolation filter).

Or allow for more aggressive noise shapers too if you desire

You keep saying this and yet have demonstrated you have no idea what noise-shaping is. Neither the filter type nor it’s transition band has anything to do with noise shaping.

I use fast roll-off, the idea behind this question is/was, if i upsample to 88.2kHz, it should no longer matter which filter i use at all, as all filters will roll off after 20kHz so i am wondering why DAC maker but the effort in to provide several filters if they could just resample to FS*2 and be done.

As mentioned above they could just use a standard optimal anti-image filter as they did 20-30 years ago and typically resample to 384Fs/S. I’m not aware of any DAC that resamples to FS*2, even the earliest oversampling DACs back when CD was released were FS*4. But the reason “why DAC makers put in the effort to provide several filters” is marketing for the audiophile community, you don’t find these options in pro audio ADCs or DACs as they just use an optimal filter.

If you're peering at the analog signal beyond 20KHz in the analog domain, yes we won't hear it, but in digital domain, the differences in accuracy between 9.99 (bandwidth limited) and 9.9999 (still bandwidth limited) since 10 (infinite bandwidth is impossible to achieve) is audible to subjectivists

Why do you keep repeating this nonsense? Firstly, we won’t hear anything in the analogue domain regardless of the frequency, because we cannot hear electrical signals, we only hear sound (the acoustic domain). Secondly, your analogy is ridiculous and you’re contradicting yourself again! The “accuracy” below 20kHz (assuming a standard optimal filter) is 9.9999… with a 44.1kHz sample rate and an identical 9.9999… with any higher sample rate. Above 22.05kHz then the “accuracy” of 44.1kHz is zero, while the accuracy of higher sample rates is still 9.9999… BUT, as you correctly agreed, anything beyond 20kHz is inaudible (“yes, we won’t hear it”) and therefore how can something inaudible be “audible to subjectivists”? That is an obvious contradiction!

9.99 is not the same number as 10.00.
Your argument seems to be small signals do not matter.

You seem to be talking about something entirely different. @theveterans was talking about accuracy “analogously”, in terms of available bandwidth (IE. Sample rates) but “small signals” has nothing to do with bandwidth. You now seem to be talking about quantisation “accuracy” (IE. Bit depth rather than sample rates/bandwidth) and literally, not analogously. And of course, bit depth and sample rates are independent. Yes, “9.99 is not the same number as 10” and I too can just invent two different numbers (say 11 and 59,000,000) and correctly claim they are not the same number but what has that got to do with filters and oversampling or in fact with anything related to digital audio?

And yet those infinitesimally small signals in the digital domain still makes a difference on a mathematical standpoint

Hang on, you were talking about the difference in bandwidth and now you’ve switched to amplitude instead of bandwidth. But OK, on this new subject, then what “infinitesimally small signals” are you talking about? Just as there’s no such thing as “infinite bandwidth”, there’s also no such thing as “infinitesimally small signals”. In the acoustic domain the smallness of signals is limited by the Brownian motion of air molecules and in the analogue domain it’s limited by Johnson (thermal) noise and this is before we even consider the far lesser ability of microphones to respond to small signals. In the digital domain our limits are far smaller (lower) than in the analogue domain (let alone the acoustic domain) so it cannot make a difference if that difference cannot be realised in the analogue domain (or acoustic domain). Additionally, your “mathematical standpoint” is wrong anyway! Firstly, the difference isn’t between 9.99 and 9.9999, it’s orders of magnitude smaller than that and Secondly, the posted math is wrong because it did not include dither. So using the correct math (which does include dither) and continuing your false analogy, then your claimed “accuracy” of 9.99 and 9.9999 are in fact both 10, all the way down to the dither noise floor and finally, this is where noise-shaping can come into play. So in this version of your false analogy you’re claiming that the difference in accuracy, which is 10 in both cases and therefore is obviously zero, “is audible to subjectivists”! lol

However, this last nonsense point is off-topic because the question is about filters and oversampling, not bit depths and the quantisation accuracy of signals that do not exist!

G

 

[mjtbah98]

Anybody can design something poorly enough that it sounds different. The amp section, the filters(if they don't filter enough or filter a lot within the audible range), and just about anything else can be poorly done and have big consequences for the signal. Then it's only a matter of how big they get and when that becomes audible. Making a DAC from scratch is a very complicated matter.

But nowadays, you can pretty much get 2 chips that do it all for you, the input is standardized(mostly USB nowadays), the output also follows standard or at least a rule to maximize voltage accuracy(output somewhere between 50 and 150ohm, going into the amp input that's at least several thousand ohms and nowadays often in the 10kohm).

And of course, a DAC is the most accurate part of our system after the source that ideally does nothing at all to the signal. The amplifier has the real work of operating with a wide variety of loads and voltages. The headphone/IEM has comparatively terrible fidelity, with most measures of fidelity being one or more magnitude worse than what a DAC does. And of course the frequency response will be whatever. You'll probably get more objective difference between 2 pairs of the same headphone, than between 2 DACs made to have high fidelity.

But very often one DAC will output higher voltage, and that could:
1/ make the listener feel like everything is different(bass, details, soundstage, you name it). And even when listeners notice that volume level difference because of how big it is, they still tend to A/B their DAC as is with the volume difference, so they still have those impressions of a bigger imaging or whatever. Nobody, who's ever learned anything about experimentation, would allow a non volume matched setting. But audiophiles and most reviewers "know better", they "have enough experience to not be affected", and all the lies you tell yourself when you try to avoid admitting that you're wrong or when you try to avoid doing something bothersome.
2/ the amplifier used will have a range of nominal input voltage. If it was designed to operate nominally with up to 2 or 2.5V, and you feed it with a DAC that outputs maybe 3V or even more(mostly high voltage is a trick to improve SNR in the DAC), then your amplifier might react very poorly to that high voltage input. It could distort much sooner(at lower power) into a given load, for example.

We can't list all the ways to misdesign and misuse gear. But that doesn't mean at all that it's hard to find very clean devices(except the transducers that have strong mechanical limitations, so even the best ones are much inferior to most cheap DACs in relative fidelity). And it's not that hard to pair products correctly, and to avoid the nonsense filter settings on a DAC.

You see in here some people talk as if any difference was audible. That is obviously pure nonsense! Humans have limited sensory abilities. And for basically all of those studied thresholds, in practice we stop detecting them even sooner with music because our focus is on different things and because the music itself masks a lot of cues that might on their own be noticeable.

So with digital audio, pretty much any DAC would do just fine? How far back does this apply? Would a CD player DAC from 20-30 years ago (1 bit implementation) be as transparent as a more modern DAC? I have two CD players, one is a 1-bit 8x oversampling (JVC XL-Z431) from 1990 and the other is a 24 bit/192khz DAC (Teac cdp-650b) from 2018. Subjectively, I can hear a difference in the players' sound, I subjectively prefer the 1-bit DAC to the 24 bit one. But I'm starting to realize it isn't that simple. It could be that the older player's tray mechanism is much more reliable and that is affecting my perception of the sound quality of the newer player. I haven't done a DBT, so what I'm saying here is purely my perception. Once again, I apologize for my ignorance. I am just getting into digital audio after years of being an analog audio afficionado (vinyl and cassette in particular) where almost anything can affect how something sounds. So this is a little bit of a learning curve for me. Thank you everyone who gave me honest answers!

 

[Vamp898]

What @danadam said. If you oversample a file before sending it to the DAC, that operation also uses/needs a filter(or several) just above the initial sample rate/2. You don't solve the issue of the filter, you just apply the digital filter early.

I just tried it with the slow roll-off filter and when i upsample to 96kHz, the roll-off is in the 30kHz zone where it was at 15kHz before.

So not sure what i understand wrong but this looks like it solves exactly my issue. What was in the audible band before is now in the inaudible band so if i upsample to 96kHz, it no longer matters if i use fast roll-off or slow roll-off.

As the delay scales with the sample rate as said in the ESS documentation, the phase issues also should no longer matter.

So upsampling to >=88.2kHz seems to make all filters transparent while 44.1kHz can cause theoretically audible differences (by either early roll-off or phase issues).

And by the way, how did this thread escalate so quickly i went to sleep, came back and BAM dozens of posts

Again, this is just an thought experiment, i do not want to waste resources up sampling everything as i use fast roll-off anyway and so do not care as i have neither phase issues nor roll-off issues.

 

[gregorio]

So with digital audio, pretty much any DAC would do just fine? How far back does this apply?

1. Yes, although your “pretty much” should exclude NOS and “tube” DACs, as well as certain filter options on certain DACs.
2. How far back is a more tricky question.

Would a CD player DAC from 20-30 years ago (1 bit implementation) be as transparent as a more modern DAC? I have two CD players, one is a 1-bit 8x oversampling (JVC XL-Z431) from 1990 and the other is a 24 bit/192khz DAC (Teac cdp-650b) from 2018. Subjectively, I can hear a difference in the players' sound, I subjectively prefer the 1-bit DAC to the 24 bit one. But I'm starting to realize it isn't that simple.

We can’t say for certain without detailed measurements of the older DAC. Your 2018 DAC will be transparent but 1990 is 35 years ago. It may well also be audibly transparent but that’s not guaranteed, some cheaper ones probably were not back then but the price point for audibly transparent reduced during the 1990’s. There were probably very few if any DACs by the late 1990’s that weren’t audibly transparent.

Once again, I apologize for my ignorance. I am just getting into digital audio after years of being an analog audio afficionado (vinyl and cassette in particular) where almost anything can affect how something sounds.

Don’t apologise! When you start getting into the guts of it, digital audio can be very complex. Complex chip and filter designs that only specialist mathematicians and chip design teams fully understand. So don’t apologise, you’re just starting a road that we all had to start and that to some degree we’re all still on but maybe somewhat further along! Although it’s not terribly hard to get your head around the basic principles of how digital audio works, it’s more complicated in the audiophile world by the decades of misinformation put out by audiophile marketing. The main difference is that digital audio was specifically invented so that “almost anything” cannot affect how things sound. Certain things can affect the sound but a surprising number can’t/don’t. Feel free to ask any questions you may have, without apologising!

I just tried it with the slow roll-off filter and when i upsample to 96kHz, the roll-off is in the 30kHz zone where it was at 15kHz before.

No, it’s not. The anti-image filter must be at the Nyquist freq of sample rate of the file you’re upsampling, not the sample rate you’re upsampling to. So if you’re upsampling a 44.1kHz file, the stop band of the anti-image filter will be around 22kHz, regardless of whether you’re upsampling it to 88.2kHz, 96kHz, 192kHz or whatever.

G

 

[2leftears]

Forgive me for my ignorance. I know next to nothing about sound science, but I do find it fascinating and try to learn what I can from reading. Is it possible that DAC chips and their sometimes multiple filters, which I've read multiple times as being transparent and therefore not responsible for changes in sound, interact differently with the analog stage of the DAC? In other words is the analog stage in any way responsible for the perceived change in sound? Again, I apologize for my ignorance here, just curious. Thanks again!

Not if the DAC (the whole audio-component, not the DAC chip) is well-designed.

Unlikely. The analog output stage of a DAC will generally have high input impedance for the sole purpose of preventing that very interaction.

To expand on that for the benefit of @mjtbah98, that depends on whether the DAC chip has a voltage source output or current source output. If the output of the DAC chip is a current source with a low compliance, then it is designed to dump the current into a very low impedance I-V conversion analogue stage. Not sure if modern DAC chips with a current source output still exist, but back in the 80's and 90's there certainly were (TDA1541 e.g. which had a current source output with a compliance of only +/-25mV).

This is something to be aware of if you start making DIY mods to DACs, because you certainly can mess up the DAC chip performance if you are ignorant of the difference(s) between a current source output and voltage source output when feeding it into an analogue output stage. I remember seeing DIY analogue stage modifications where the hapless designer simply took the TDA1541 current output into a resistor to ground to generate a 2V line level, exceeding output compliance by a factor of 100

So with digital audio, pretty much any DAC would do just fine? How far back does this apply? Would a CD player DAC from 20-30 years ago (1 bit implementation) be as transparent as a more modern DAC? I have two CD players, one is a 1-bit 8x oversampling (JVC XL-Z431) from 1990 and the other is a 24 bit/192khz DAC (Teac cdp-650b) from 2018. Subjectively, I can hear a difference in the players' sound, I subjectively prefer the 1-bit DAC to the 24 bit one. But I'm starting to realize it isn't that simple. It could be that the older player's tray mechanism is much more reliable and that is affecting my perception of the sound quality of the newer player. I haven't done a DBT, so what I'm saying here is purely my perception. Once again, I apologize for my ignorance. I am just getting into digital audio after years of being an analog audio afficionado (vinyl and cassette in particular) where almost anything can affect how something sounds. So this is a little bit of a learning curve for me. Thank you everyone who gave me honest answers!

Well, I find the better CD players of the early 90's already to be transparent to my ears. I mean better in terms of well-engineered; not necessarily high-end. Some players from the early 90's that were sitting at the top-end of HiFi or bottom-end of High-End sounded transparent to my ears. Many well-designed TDA1541/SAA7220 combos from that era exist, although some purist will find something to complain about re. the SAA7220 digital oversampling filter, but personally I have no issues with that filter as far as transparency goes.

 

[colonelkernel8]

The analog output of a DAC is usually around 75 ohms and 600 ohms in Balanced XLR. As for analog stage in the DAC, discrete transistors in analog stage can certainly sound different than an op-amp based analog stage. Both will measure beyond the established audibility of an arbitrary ~ 90 dB SINAD as example yet depending on the headphones you use, you can possibly still hear differences with DACs



Professional (design audio engineers to be specific) do care lol. If they don't, I wouldn't get these wonderful sounding DACs but rather sterile sounding, harsh, digital fatigue inducing DACs instead

Do you know what a mastering engineer is? Do you know what an audio engineer is? Neither of these people have anything to do with the design of electronic components unless they are also electronics design engineers. Showing me the literal effing marketing material for the hack audiophool company isn't doing anyone any good. You're actually incapable of making a good argument.

 

[castleofargh]

I just tried it with the slow roll-off filter and when i upsample to 96kHz, the roll-off is in the 30kHz zone where it was at 15kHz before.

We're talking about the conversion you did before on your computer to turn 44.1 or 48 to 96kHz. That operation requires a filter somewhere around 22kHz and did apply it to remove the images created in the process. IDK what you use as a resampler, Some don't ask you anything other than the sample rate you want, some have a quality option, but some have the fanciest list of settings for the filter where you can play with just about anything so long as you have the CPU for it(like HQ player).

But in every scenario, some digital filter is applied to remove everything above the original range of the signal. There is no trick to completely avoid filtering after you changed the sample rate. You just move that from the DAC to the PC. And afterward of course, as you give the DAC a 96kHz track, it filters around half that, as it should.

Is that clearer?

We can get into how things are done and why, like zero stuffing(adding zeroes between samples until you get the sample rate you want, and what the interpolation filter(although you must not ask me because I lost that math a long time ago, so beside knowing that it gives the right value to the previous zeroes and attenuates the signal that the resampler will just compensate with digital gain so we're none the wiser, I really can't tell you anything else).


@mjtbah98 Sorry, I don't know. Delta sigma DAC chips certainly have improved over the years, so we should expect the newer ones to at least be objectively better. But when did certain products become transparent? I couldn't say. Even today I'm sure you can find some non-transparent devices if you really go look for them. If only some niche audiophile brand that does it on purpose while saying in the least suable way, that it sounds different because it's better.
The CD players are also perhaps a special case because it's more involved than just a DAC, there is the state of the CD itself and whatever error compensation/correction scheme the CD player is using. I know nothing about that, so I'll let others discuss it or tell if I'm bringing up irrelevant ideas.

 

[colonelkernel8]

To expand on that for the benefit of @mjtbah98, that depends on whether the DAC chip has a voltage source output or current source output. If the output of the DAC chip is a current source with a low compliance, then it is designed to dump the current into a very low impedance I-V conversion analogue stage. Not sure if modern DAC chips with a current source output still exist, but back in the 80's and 90's there certainly were (TDA1541 e.g. which had a current source output with a compliance of only +/-25mV).

This is something to be aware of if you start making DIY mods to DACs, because you certainly can mess up the DAC chip performance if you are ignorant of the difference(s) between a current source output and voltage source output when feeding it into an analogue output stage. I remember seeing DIY analogue stage modifications where the hapless designer simply took the TDA1541 current output into a resistor to ground to generate a 2V line level, exceeding output compliance by a factor of 100

Fair point and my mistake to not mention it. I always forget the Analog Devices and Burr Brown stuff because it's so old now.

 

[Slaphead]

Sorry, I don't know. Delta sigma DAC chips certainly have improved over the years, so we should expect the newer ones to at least be objectively better. But when did certain products become transparent? I couldn't say. Even today I'm sure you can find some non-transparent devices if you really go look for them. If only some niche audiophile brand that does it on purpose while saying in the least suable way, that it sounds different because it's better.

Nail on the head - these $5'000 plus DACs need to sound different in order to differentiate themselves from the lower price brackets where DACs are already perfectly transparent and also have for the most part a perfectly good analogue section.

It's there we have a problem in that if you spend big bucks you may end up with a worse performing DAC than if you'd spent sensible money.

IMO there are a number of boutique brands that exploit this. One, that shall remain nameless for fear of invoking the wrath of their disciples, recently released a headphone/speaker amp (no DAC) for "professional use" at a cost in excess of $3500 that measured so badly, and I mean really badly, it's been considered a total joke. Honestly any cheap HiFi amp would measure better.

OK, measurements are not the whole story. but when they're that bad then it's a big red flag for anybody except for the most ardent of the "converted".

 

[bigshot]

Slaphead, if it isn’t too much trouble can you please pm me a link to info about that bad audiophile amp?

 

[theveterans]

Slaphead, if it isn’t too much trouble can you please pm me a link to info about that bad audiophile amp?

I guess it's the Chord Alto I believe

 

[JamoBroGuy]

Slaphead, if it isn’t too much trouble can you please pm me a link to info about that bad audiophile amp?

Me too, I gotta see this.

@theveterans I doubt it. I just checked the ASR review, it doesn’t measure that badly. Not bad enough to be considered a joke.