Worlds Greatest DAC and what it does differently

[beerandmusic]

I noticed a few more have responded.  Thank you. 

I will take the time to read through carefully.  Again, I am a layman, and most of this stuff is way over my head.

I am "extremely" left brained, which is likely why I am socially dysfunctional.  My logic differs than most peoples.

I am also ADD...which is why I am going to go here now:::

 

Someone commented in another posting, that I made a rather offensive remark, where in my thinking, it was not offensive at all.  Quite the opposite.  In my obscure way of thinking, honesty conquers all other forms of communication, and anything that is less than honest is a "disservice" to humanity.  I believe everyone should be "candid" with their true feelings and emotions.  I believe that if everyone did this, that at first there would be a lot of chaos and death, but once it stabilized, and universally accepted, there would be more harmony, understanding and acceptance.  I am sure someone will give me an "off topic" for these thoughts, and that is fine.  This is my thread, and where someone may feel it is off topic, I do not.  It tells you a little bit about my character, and where I am going with this thread.  I do not seek conflict, I am not biased, and I only seek understanding and harmony.  Of course, I realize it will be a rocky road getting where I am going.

 

Anyway, enough of my ADD ...switching gears....I will read through carefully, and respond soon, in my layman's manner.  Time for lunch. 

 

What i currently believe, but am open to anything, is that whereas I do believe the source matters, that the better the source is, the less upsampling matters....and the most important part of this adventure, will to be keeping in mind, that if a DAC is doing it's job properly, then all DACS should sound the exact same knowing they have a finite input of a previously recorded signal and their only function is to be accurate in recreating that signal.

 

Edit to add::  knowing the original signal has been digitized, engineers should be able to know if their DAC is functioning properly by comparing the input signal (prior to being digitized) to the output signal.  If there is any differences between dacs, then at least one of them is faulty (keep that in mind).  When troubleshooting a DAC design, engineers should compare the input to the output gradually adding more things in to see when there is a difference on the output.  Initially, testing simple analog signals, and then continue to be more complex. 

 

Lunch time...

 

 

[barrows]

In a "perfect" world, where electronics all acted "perfectly" with no errors or losses' and where digital filters were "perfect", with infinite stop band attenuation and no artifacts, then 16 bit 44.1 kHz sample rate would result in perfect output from a DAC.

 

But nothing in the real world works "perfectly", everything is subject to errors in one way or another:  digital filters produce artifacts (alias products and or ringing), analog circuitry is subject to interferences from high speed processor noises which may produce intermodulation effects in the audible bandwidth.  Electronics are subject to thermal noise at very least, and harmonic distortions.  DACs are mixed digital and analog circuitry in a single chassis, which in itself is a challenge.

 

Higher sample rates, and oversampling overcome some of the real world limitations of how things actually operate.

[beerandmusic]

23 minutes ago, barrows said:

In a "perfect" world, where electronics all acted "perfectly" with no errors or losses' and where digital filters were "perfect", with infinite stop band attenuation and no artifacts, then 16 bit 44.1 kHz sample rate would result in perfect output from a DAC.

 

But nothing in the real world works "perfectly", everything is subject to errors in one way or another:  digital filters produce artifacts (alias products and or ringing), analog circuitry is subject to interferences from high speed processor noises which may produce intermodulation effects in the audible bandwidth.  Electronics are subject to thermal noise at very least, and harmonic distortions.  DACs are mixed digital and analog circuitry in a single chassis, which in itself is a challenge.

 

Higher sample rates, and oversampling overcome some of the real world limitations of how things actually operate.

 

I haven't read through yours or davides posts yet, so i will get to them later, but this posting captured my attention.

I have lived in the digital world for my entire career and things are more "predictable", so maybe this is where my logic fails in the audio world....but i want to concentrate on where things break down....i think most people would agree that most dacs would produce the same output for a simple sine wave, so it is a good starting point....anyway...thanks for your input as always...will continue to digest.

 

edit to add:: since you believe if everything was perfect that 44.1khz should be sufficient to capture the audible spectrum, I personally think that 96K would likely be more ideal as 44.1K is very close to our boundaries and that 96K "may" be a rudimentary step that would leave no room for argument, and plus it is a multiple of 48k which I understand is more ideal for video, that 96K "may" be a "sweet spot" for both recording and reconstruction.

[beerandmusic]

I may pop in from time to time simply to post a thought that entered my mind, that may or may not be applicable, but at the moment I post it, I believe it is applicable, and don't want to forget the thought.....

 

this is such a moment, and just want to share this thought....

 

Part of the problem with DAC design may be the amount of different inputs it has to resolve for, that if the design could concentrate on a limited amount of inputs, the design could be better perfected....along this thought, if 44.1K fully encapsulates our audible spectrum, that if the DAC only had to resolve for example 44.1, 48, and perhaps 96k, it may be easier to perfect the design to accurately reproduce the input signal, if it had less variables to deal with.

[beerandmusic]

another thought...i think bit depth is important for dynamics, and usually when i do critical listening, i look for dynamics in very quiet background sounds, and is where i typically find differences.  I understand there is a theorem (nyquist) for sample rate and this is probably a very stupid question that i could probably google to find the answer, but want to continue to demonstrate that I am a layman, and that no question is a stupid question, so..the question is, does this theorem take into consideration bit depth?

 

again, this is just a rambling thought that crossed my mind, that i do not wish to forget, so just posting it here for future consideration....

[barrows]

14 minutes ago, beerandmusic said:

I may pop in from time to time simply to post a thought that entered my mind, that may or may not be applicable, but at the moment I post it, I believe it is applicable, and don't want to forget the thought.....

 

this is such a moment, and just want to share this thought....

 

Part of the problem with DAC design may be the amount of different inputs it has to resolve for, that if the design could concentrate on a limited amount of inputs, the design could be better perfected....along this thought, if 44.1K fully encapsulates our audible spectrum, that if the DAC only had to resolve for example 44.1, 48, and perhaps 96k, it may be easier to perfect the design to accurately reproduce the input signal, if it had less variables to deal with.

 

In most DACs, the actual conversion stage does operate in either a single rate (for those with an asynchronous upsampling section) or at a high rate related to the two base frequencies (usually 16x rate).  So they are already optimized for a given rate.

 

The exception would of course be NOS style DACs, but these are in the minority.

 

Indeed it is correct to think that if a DAC is optimized for certain sample rate, it may be able to more easily deliver the expected performance level.  For example, on my ESS 9038 DAC build I optimized all settings of the ESS chip for DSD 256 input, and then I also optimized the analog output filter to suit this rate as well.

 

This is what I have been thinking about:

 

Single rate DAC optimized for single rate (DSD 256) input

"Perfected" DSC style discrete conversion stage

"Perfected" optical Ethernet input with NAA, ROON RAAT, and DLNA (for old schoolers) capability (no other input)

onboard, by passable, analog volume control

A robust output stage capable of driving any amplifier directly without compromise

 

Such a DAC, made really well, in the USA, with high quality, ought to be able to be produced for under $10K, perhaps even $7K would be possible, given that we are leaving out unnecessary "features".  It would require the owner to oversample in software to DSD 256, but that is fairly easy to obtain these days from many playback softwares.

 

I would even want to make an "one box" version, with amplifier modules as well.

[beerandmusic]

^^^  I think that once the "end to end" D->A design is perfected, and mass produced, that it could be manufactured based on BOM for very little money....of course, initial runs, to be profitable and to carry engineering costs would be astronomical...but a lot of knowledge is already known...

My thinking is that since we have a finite digital signal, the entire analog signal could be accurately mathematically reconstructed via software (and i am not sure i understand any need of high rate dsd to reconstruct a 44.1k recording), and that the actual analog out really wouldn't need expensive power or isolation circuitry....but this is a different topic, and just a concept in my mind at the moment...I would have to hash the thoughts around a lot, before i could say this confidently....but currently i have much lower level understanding needed before i get to that point.

 

 

 

 

[barrows]

The reason for the DSD 256 conversion rate is because this makes the actual conversion process much much simpler and more accurate.  Resulting in improved sound quality.

And who said we should start with 16/44.1 music recordings, those are compromised to begin with.  I will choose a higher rate file any time I have the option to purchase it, because there is an advantage to higher rate of recording.

 

BTW, right now I am listening to a DSD 128 file of Debussy Piano works recorded by Wave Kinetics records, played by Ilya Itin, through the native single bit discrete conversion of the Bricasti M3 DAC.  It sounds AMAZING!

[beerandmusic]

45 minutes ago, barrows said:

The reason for the DSD 256 conversion rate is because this makes the actual conversion process much much simpler and more accurate.  Resulting in improved sound quality.

And who said we should start with 16/44.1 music recordings, those are compromised to begin with.  I will choose a higher rate file any time I have the option to purchase it, because there is an advantage to higher rate of recording.

 

BTW, right now I am listening to a DSD 128 file of Debussy Piano works recorded by Wave Kinetics records, played by Ilya Itin, through the native single bit discrete conversion of the Bricasti M3 DAC.  It sounds AMAZING!

 

I agree, i don't think we should start with 16/44.1 but that is what mostly exists...but my current understanding is there likely wouldn't be any benefit to higher than 96K sampling rate( and unsure of if there is a theorem representing optimal bit depth?).

 

I have many native DSD256 and DSD128 files and have downampled them out of curiosity and couldn't hear differences, whereas a quieter source and better bit depth have made an audible difference. I also agree, It may make no sense to downsample if we have the technology to "ACCURATELY" record and reproduce at higher rates, then why not (other than wasted power).  My issue is that there are bigger issues to resolve for than potentially unnecessary higher sampling rates. 

 

If i play a 320k mp3 file from a good player, it will sound a lot better than a DSD256 file playing from a noisy local pc.  We need to identify why we hear differences in different dacs and why the source matters so much.

 

I really am kind of spacing out right now, and not sure where I am trying to go with this entire thread, as I have new discoveries since i started it (grin).  I guess I believe that a well designed DAC even with a "lowly" 96K should be as capable as a quad dsd dac, and you have already surrendered that in a perfect world a 44.1K dac would be sufficient.

 

I am curious as to what is the threshold of the break down is....if you believe a 44.1K should be able to be 100% accurate in reproductions of a 44.1k recording. There has to be a point where one can confidently say that under certain "defined conditions" that a 44.1K dac will be perfect.... again, I don't even like using 44.1K as i believe in "golden ears" and room for correction, and why not 96K....i am only using 44.1k for this hypothetical dac because that is cd rate and what most recordings were created and it is of lowest denomination as a "starting" point....an am just as happy to just start with 96K since so far it seems there is a concensus that every dac can reproduce a 1Khz test signal with relatively 100% accuracy.

[beerandmusic]

REFOCUS::  We can discuss higher recordings later...I think for now i want to stick to resolving for original recordings at 44.1K still...I personally am not ready to move on from that at this time for my purposes....i think it is agreed, by at least many, that original recordings recorded at 96K with higher bit depths, may have benefits...... but most of what is in libraries today is 44.1k....so sticking with that "for now".

 

On that assumption, i believe most people would agree it is "possible" to accurately reproduce the original signal with a 96K sample rate.  The downside to upsampling higher is the possibility that better clocks, better power, and more advanced filtering is needed, which in of itself could add need for more money and wasted power if it is not necessary to "accurately" reproduce an original recording done at 44.1K.

[beerandmusic]

7 hours ago, yamamoto2002 said:

 

When 1 second of 1kHz signal is sampled by 44100Hz, 44100 sampled values is produced. (This is obvious🙂 )

 

When we look closer to those 44100 samples, first 441 sample values are repeated 100 times. This is caused by periodicity of sine function : sin(x + 2π) = sin(x). This 441 samples contain exact 10 cycles of sine wave.

 

Most of sampled values are irrational numbers.

When those values are quantized to 16bit or 24bit integer PCM, sample values becomes something like the following table.

 

In digital domain, PCM signal is stored/transferred as a list of those integer values. One second of PCM consists of 44100 integer values.

 

 

amazing that much data is used in a 1 second sampling of a simple 1k sine wave.

I am confused as how does this data look digitally in a flat file?

[yamamoto2002]

5 hours ago, beerandmusic said:

The reason, i wanted to know about a "simple" 1K signal is because in the audio world it is simple, finite, and predictable.

 

There is actually infinite there in the discrete-time sampled 1kHz sine signal:

• Most of sample values are irrational numbers. This means, to represent one sample value exactly (for example sample#148 = sqrt(3)/2 ), infinite number of digits are needed. 16bit quantization truncates the value of first 16bit of the number to create finite digit rational number and quantization noise is generated by the truncation.
• Pure 1kHz signal is, by definition, has infinite length. If it is truncated to finite length (1 second), other frequency components appear on the truncated edge and signal is contaminated. You may hear click noise at the truncated edge.

[beerandmusic]

8 minutes ago, yamamoto2002 said:

 

There is actually infinite there in the discrete-time sampled 1kHz sine signal:

• Most of sample values are irrational numbers. This means, to represent one sample value exactly (for example sample#148 = sqrt(3)/2 ), infinite number of digits are needed. 16bit quantization truncates the value of first 16bit of the number to create finite digit rational number and quantization noise is generated by the truncation.
• Pure 1kHz signal is, by definition, has infinite length. If it is truncated to finite length (1 second), other frequency components appear on the truncated edge and signal is contaminated. You may hear click noise at the truncated edge.

thanks.  I guess i should have said the digitized flat file would be more simple and easier to compare the original analog input to the analog output (assuming differences would be easier to be measured?)  There may be an easier test for accuracy, but as a layman, that just seemed like a good example as a starting point.

[pkane2001]

To give you an idea of the scale of the error, here's a 1kHz sinewave captured from a non-audiophile DAC (purchased for $20 used).

 

There are two overlapping waveforms here. Pink is the measured, captured digital waveform, while blue is the original, mathematically constructed sinewave:

 

The DAC isn't great as far as measurements are concerned. Here's that same 1kHz capture in the frequency domain, showing distortions (all peaks that are not at 1kHz are distortion), fairly high noise floor, and some stray noise signals. No doubt some of these are responsible for the tiny difference above.

 

[yamamoto2002]

3 hours ago, beerandmusic said:

amazing that much data is used in a 1 second sampling of a simple 1k sine wave.

I am confused as how does this data look digitally in a flat file?

 

I choose WAV file as an example to explain it.

 

WAV file stored 1 second (truncated) 1kHz sine can be created using WaveGene.
Generated WAV file can be read using any Hex editor. I'm using HxD.

 

 

Click the following images to magnify

 

 

 

[beerandmusic]

29 minutes ago, yamamoto2002 said:

 

I choose WAV file as an example to explain it.

 

WAV file stored 1 second (truncated) 1kHz sine can be created using WaveGene.
Generated WAV file can be read using any Hex editor. I'm using HxD.

 

 

 

 

thanks, i realized it would have a header, and assume an acknowlege of some type (even for isosynchronus?), it appears what is actually recorded is the offset rather than a value, and it does appear linear.  What does the first negative number represent, and will it eventually repeat?

I just noticed your tagline Developer of PlayPcmWin...will have to check it out.

 

Edit to add:: I just downloaded your player and it sounds great.

 

 

[fas42]

10 hours ago, davide256 said:

I have found that DSD upsampling matters with CD quality choral pieces; there is normally a fuzziness associated with massed voices that up sampling removes, allows inner

counterpoint singing to be heard clearly vs covered up. I've heard this also with better PCM gear but the price tag becomes unaffordably expensive.

 

 

 

"Fuzziness associated with massed voices" is as good as any other subjective measure, of what digital playback often gets wrong. Very expensive PCM playback usually has the engineering to resolve this failure of the playback chain to demonstrate truly accurate presentation of the recording - carefully applied tweaking to normal priced gear is another method, but is only open to some people.

[beerandmusic]

13 hours ago, davide256 said:

I have found that DSD upsampling matters with CD quality choral pieces; there is normally a fuzziness associated with massed voices that up sampling removes, allows inner

counterpoint singing to be heard clearly vs covered up. I've heard this also with better PCM gear but the price tag becomes unaffordably expensive.

 

 

what is an example cd i can test what you are speaking about?  Can you share a cd, track and time that you are talking about? 

[davide256]

1 hour ago, beerandmusic said:

 

what is an example cd i can test what you are speaking about?  Can you share a cd, track and time that you are talking about? 

John Williams soundtrack for the movie "Rosewood", 2nd track "Look down Lord". Massed voices for the entire track have a fuzz to the sound without upsampling

and an interspersed baritone counterpoint becomes vague and hard to distinguish without up sampling

[yamamoto2002]

10 hours ago, beerandmusic said:

it appears what is actually recorded is the offset rather than a value, and it does appear linear. 

 

WAV file has the start and the end (Unlike digital radio broadcast stream where there is no start and no end) and WAV file format use it as the means to organize data.

 

Offset is the count of bytes from start of the file and it usually counts from zero.

 

On the HxD Editor screenshot of my previous post, 8-digit hexadecimal offset number is added by HxD Editor for human reading convenience.

 

Actual WAV files do not contain offset values in the file, but it can be calculated by counting data bytes from the file start. On the accompanying
table of the slides has also offset columns and I added it for readers to compare easily the table rows with HxD screenshot hexadecimal dump.

 

Actual WAV file data is something like this (expressed as hexadecimal number sequences, which is shown in HxD Editor screen) :
 52494646AC580100574156...

 

One 1-digit hexadecimal number is actually 4-bit data and it can be expressed as a 4-digit binary number:
    hex 0 ⇔ bin 0000, hex 1 ⇔ bin 0001, hex 2 ⇔ bin 0010, hex 3 ⇔ bin 0011
    hex 4 ⇔ bin 0100, hex 5 ⇔ bin 0101, hex 6 ⇔ bin 0110, hex 7 ⇔ bin 0111
    hex 8 ⇔ bin 1000, hex 9 ⇔ bin 1001, hex A ⇔ bin 1010, hex B ⇔ bin 1011
    hex C ⇔ bin 1100, hex D ⇔ bin 1101, hex E ⇔ bin 1110, hex F ⇔ bin 1111

 

Therefore this WAV file data also can be expressed as a binary digits sequence:
 0101001001001001010001100100011010101100010110000000000100000000010101110100000101010110...

 

On storage media,  value 1 may expressed as the hole (of the punched card), magnetized, charged, high resistance values, etc.

 

10 hours ago, beerandmusic said:

What does the first negative number represent, and will it eventually repeat?

 

As you said, on the "Looking into 16bit PCM data part of the WAV file" slide, last PCM data ends with minus number, this is because I tried to explain 16bit integer value can express minus number as well. Actual WAV file goes on.

 

1kHz 1 second 44100Hz PCM starts from 441 unique sample values and it is repeated 100 times. Please refer my first post of this thread what it is like this 441 sample values are.

[Audiophile Neuroscience]

45 minutes ago, yamamoto2002 said:

One 1-digit hexadecimal number is actually 4-bit data and it can be expressed as a 4-digit binary number:
    hex 0 ⇔ bin 0000, hex 1 ⇔ bin 0001, hex 2 ⇔ bin 0010, hex 3 ⇔ bin 0011
    hex 4 ⇔ bin 0100, hex 5 ⇔ bin 0101, hex 6 ⇔ bin 0110, hex 7 ⇔ bin 0111
    hex 8 ⇔ bin 1000, hex 9 ⇔ bin 1001, hex A ⇔ bin 1010, hex B ⇔ bin 1011
    hex C ⇔ bin 1100, hex D ⇔ bin 1101, hex E ⇔ bin 1110, hex F ⇔ bin 1111

 

Therefore this WAV file data also can be expressed as a binary digits sequence:
 0101001001001001010001100100011010101100010110000000000100000000010101110100000101010110...1001..

 

Fixed that for you, duh ! 🙄

[yamamoto2002]

4 minutes ago, Audiophile Neuroscience said:

 

Fixed that for you, duh ! 🙄

Thanks! I'm surprised you actually read it 😁🙏

[beerandmusic]

2 hours ago, yamamoto2002 said:

 

1kHz 1 second 44100Hz PCM starts from 441 unique sample values and it is repeated 100 times. Please refer my first post of this thread what it is like this 441 sample values are.

 

ok, thanks now it makes a lot more sense to me....only "value" columns are the actual data in the flat file, and you are showing it expressed in hex vs binary.  As simple as the signal is, it still requires a lot of unique bits (smile)....still it should be able to be converted to an analog wave form out with relative precision, and should the same no matter what dac was used or how it is upsampled.

[beerandmusic]

 

For my purposes, the fact that Barrows conceded that with "perfect" electronics, you can't better sound than what a 44.1K PCM dac would do, and that no upsampling would give you better sound....I was even willing to concede a 96K PCM dac would do....(and actually is closer to what my belief is).

 

That acknowledged, that also allows for my statement "the better the source" the less upsampling will make any difference.

 

I never saw an answer regarding if there is a theorem for what bit depth is required for human hearing, but i found this

https://www.soundguys.com/audio-bit-depth-explained-23706/

so if 20bits is the magic number, i will just go with 24 to be safe.

 

I guess to surmise, then it would be accurate to say the worlds greatest dac, will compensate for imperfect electronics, and present an accurate reproduction of the of the original analog recording, and does not need to be any more sophisticated than a 24 bit 96K PCM DAC.

 

Not sure I need to go any further with this topic...but it may raise a new topic (smile).

[beerandmusic]

^^^^ Thinking more along these lines, I am not sure there is anyone more knowledgeable about usb and pcm than Mike Moffat (Schiit) and if 24bit 96K dac is capable, I am not sure I need to put my money anywhere else (provided i ultimately decide on USB).

 

I actually will always have both USB and enet solutions because USB is just more convenient for playback functions, and is less finicky. 

 

I also believe that "imperfect electronics" is more prominent in usb than ENET, which is why I will always have an enet solution as well for more critical listening...as has always been my contention.

 

I will always keep in mind though that if "imperfect electronics" are kept in check that all competent dacs "should" sound the same if they are doing their job...you should never have to "pay" for a dac that sounds better, the correct way it should be said, is that you pay for a dac that better compensates for "imperfect electronics" to more accurately recreate the original recorded signal....