Bit perfection

[scott wurcer]

Only if you use an infinitely long sinc filter. In practice, there is always some error. The error can be made arbitrarily small, but never zero.

No, 64 bit double precision math can round to bit perfect results at 24 bits and certainly 16 bits. Remember bit perfect only matters to the supplied resolution, rounded to 16 bits if you start with 44.1/16 is all that matters.

I am assuming you keep the 48k data in double precision floats, so yes an unrealistic example.

 

[L5730]

When I demand that the integrity of the digital data is no manipulated and transformed, I expect exactly that. Null test input and output and get absolutely nothing as the result -bit perfect.

As all of my music is gain adjusted on-the-fly by ReplayGain in Foobar2000 and at 64 bit floating point precision and dithered down to 32 bit, what feeds the DAC cannot be bit perfect compared with what the original file is.
In Linux, I couldn't find a player that supported the features I make use of in Foobar2000 and also not-resample. I'd prefer all of music played at it's native sample rate and only the gain be adjusted.
I do concede that it might be very tough to consistently tell a re-sampled audio track from it playing at it's native sample rate, as there are very good SRC algorithms now.
I have read crazy things where someone think their music re-sampled to a higher frequency sounds better. I'd agree if the DAC was an old SoundBlaster which resampled everything to 48kHz, and did it badly! Re-sampling in software did a better job, but really re-sampling 44.1kHz to 96kHz sounding better than the original? I can't help but think of sighted bias here. If there is any audible difference maybe their prefer the error *shrug*.

 

[Bounce44.1]

Analog to Digital conversion never perfectly mirrors the Analog wave form. Quantization occurs with every change in the original digital capture. Manipulating this wave form should be intended and any other unintended manipulation should be frowned upon. Any hardware/software environment should never be allowed to manipulate the wave form unless specifically instructed to do so.

The Digital to Analog conversion attempts to "fill in the blanks" left when the wave form was converted from Analog. Voodoo I tell ya!!

 

[Degru]

Bit perfect primarily concerns the OS audio stack between the audio player application and DAC, which does indeed have an audible and measurable effect on audio. See here: http://archimago.blogspot.com/2015/11/measurements-windows-10-audio-stack.html

This is why many people prefer to use WASAPI/ASIO, and why many Android users use UAPP with an external DAC.

However, on operating systems with properly implemented audio stacks that either don't do resampling and/or have a much higher quality resampling algorithm such as Linux, iOS, and macOS, this is not a concern. On most Linux distros you will be getting "bit perfect" out of the box for most music since Pulseaudio tries to avoid resampling for 44.1 and 48khz and does not apply any weird filters to the audio like Windows does. Its default resampling algorithm is a little sub-par, but you can easily switch to a very high quality one in the config file.

iOS (and I think macOS) similarly tries to auto-switch sample rates based on what's playing and what the DAC supports, so no concerns there.

As far as the rest of the chain, as long as you don't have digital-to-digital conversion steps in between the source and DAC that may or may not be mucking things up (HDMI -> receiver -> optical, etc.) you shouldn't have issues if the OS audio stack is in order.

 

[BillG]

why many Android users use UAPP with an external DAC.

Poweramp now has high precision audio engine, that can talk directly to the hardware on applicable devices, thus bypassing the Android mixer. As for external DACs, very few of my acquaintances use them on their Android devices...

 

[Julf]

Analog to Digital conversion never perfectly mirrors the Analog wave form. Quantization occurs with every change in the original digital capture.

Yes, and the quantization error manifests itself as noise that even at 16 bits is lower than the noise floor of pretty much any recording, studio, microphone or listening room.

The Digital to Analog conversion attempts to "fill in the blanks" left when the wave form was converted from Analog. Voodoo I tell ya!!

Indeed. Mathematics (in this case the Shannon-Nyquist sampling theorem and sinc function) are almost like magic.

 

[BDWoody]

Yes, and the quantization error manifests itself as noise that even at 16 bits is lower than the noise floor of pretty much any recording, studio, microphone or listening room.



Indeed. Mathematics (in this case the Shannon-Nyquist sampling theorem and sinc function) are almost like magic.

One of the first areas I tried to learn more about when having my own internal obj/subj debate was Shannon-Nyquist. It answered a lot of questions, and relieved a lot of anxiety about what I was losing...since I wasn't losing anything.

Really awesome stuff.

A little learning goes a long way...

 

[Julf]

One of the first areas I tried to learn more about when having my own internal obj/subj debate was Shannon-Nyquist. It answered a lot of questions, and relieved a lot of anxiety about what I was losing...since I wasn't losing anything.

Really awesome stuff.

Indeed. It is a pity that so few audiophiles understand it.

 

[Bounce44.1]

Yes, and the quantization error manifests itself as noise that even at 16 bits is lower than the noise floor of pretty much any recording, studio, microphone or listening room.



Indeed. Mathematics (in this case the Shannon-Nyquist sampling theorem and sinc function) are almost like magic.

Point one: I find it shows itself as distortion which can be good or bad. Dither is added either to the low frequency or the high frequency to compensate for this and or bit depth change.

Point two: If this is so precise why do different dacs sound different?

 

[solderdude]

point 1:

point 2:
https://www.audiosciencereview.com/forum/index.php?threads/dac-types-and-their-sonic-signature.7959/
https://www.audiosciencereview.com/forum/index.php?threads/differences-in-dacs.7954/#post-193238
https://www.audiosciencereview.com/forum/index.php?threads/dac-frequency-response.7901/

In short when a DAC sounds different in a well performed blind test it isn't doing something quite right and (purposely or not) deviates way too much from how DA conversion is supposed to be done acc. to the rules set out for this task.

Bit perfect has nothing to do with this though. A connection can be bit perfect in all cases yet some DAC's will still sound equally different.

 

[Julf]

Point one: I find it shows itself as distortion which can be good or bad. Dither is added either to the low frequency or the high frequency to compensate for this and or bit depth change.

Dither is added to decorrelate the noise, and only applied when you reduce the number of bits, not the other way around.

Point two: If this is so precise why do different dacs sound different?

Sub-optimal filter algorithms (either bad design or on purpose to please audiophiles), and "voiced" analog stages.

 

[sajunky]

Sub-optimal filter algorithms (either bad design or on purpose to please audiophiles), and "voiced" analog stages.

I agree. It can be other reason you didn't mention and it seems to be most popular now. This is a focus on measurements and ignore how it sounds. Manufacturers follow this thrend and create artificial sounding products like Topping D30, DX3Pro, probably the same with couple D50's models. It is reported in this thread that default filter sounds harsh, while Minimum Phase filter sounds better. Default filter is tuned for the best measurements, while the other is tuned more relaxed. On the other side the well established community expert claims that chosing filter shouldn't matter if "all filters were 'proper' without severe roll-off or aliasing" and I agree with such statement. It looks like we are promoting an opportunistic behaviour.

 

[scott wurcer]

"all filters were 'proper' without severe roll-off or aliasing"

I see these as mutually exclusive, the slow roll off filters allow aliasing. I think it started with Wadia and their polynomial interpolation which made more "natural" looking transient and impulse response pictures (test waveforms that all violate Nyquist BTW). I wish someone would demonstrate how much "pre-ringing" actually occurs with real music waveforms. So maybe I'm missing what was meant by that comment.

 

[Julf]

I see these as mutually exclusive, the slow roll off filters allow aliasing. I think it started with Wadia and their polynomial interpolation which made more "natural" looking transient and impulse response pictures (test waveforms that all violate Nyquist BTW). I wish someone would demonstrate how much "pre-ringing" actually occurs with real music waveforms. So maybe I'm missing what was meant by that comment.

That was pretty much what I was referring to by "Sub-optimal filter algorithms on purpose to please audiophiles".

 

[solderdude]

I remember RayDunzl had a recording of some spoons that showed pre-ringing but I have no idea how well the input of his ADC was filtered.
It was visible in the digital file so must have been present at the recording phase.
On playback I expect it to be reproduced accurately.

The 'ringing' only occurs with frequencies near nyquist.
Only when using sharp digital EQ filters (linear phase) in the audible band you can get pre-and post-ringing in the audible range. (AFAIK)
The dreaded 'ringing' is only visible with (illegal) test signals which exist to show filter properties.
These signals can not exist in any recording.

But we are off topic... this thread is about bit perfect.

 

[sajunky]

I see these as mutually exclusive, the slow roll off filters allow aliasing. I think it started with Wadia and their polynomial interpolation which made more "natural" looking transient and impulse response pictures (test waveforms that all violate Nyquist BTW). I wish someone would demonstrate how much "pre-ringing" actually occurs with real music waveforms. So maybe I'm missing what was meant by that comment.

I saw such analysis somewhere that provide to the conclusion that pre-ringing is not really detectable in a properly mastered material. It can be heard only when clipping effects are deployed like commonly used in a modern pop music. I usually don't listen to a 'square' music, so assuming the assessment is correct, it doesn't concern me directly.

 

[sajunky]

That was pretty much what I was referring to by "Sub-optimal filter algorithms on purpose to please audiophiles".

Or to please those focusing only on measuring results.

 

[solderdude]

hard clipping will set of 'ringing'. The more headroom one has for intersample overs the higher the amplitude may be.
I saw a lot of 0dB clipping in DACs is pretty 'hard' in Archimago's scope plots which made ringing almost invisible with some DACs (at 0dBFS)

 

[solderdude]

Or to please those focusing only on measuring results.

But in that case only the ones that 'value' perfect needle pulse and squarewaves. The people that like filterless NOS and have ears that cannot detect the crap going on at highest frequencies.

 

[Cosmik]

Those who use DSP crossovers cannot have bit perfection because what gets fed to the amps must have been altered by DSP. So in the overall scheme of things, maintaining bit perfection all the way through will yield inferior results, because ultimately the crossover filtering will be done by coils of wire and plates of foil and paper, etc. So we have people worrying about -118dB of digital error while listening for it through Edwardian technology.

The enlightened DSP person will have a far superior system despite their lack of bit perfection all the way through.

However, it is worth maintaining bit perfection as far as the crossovers if possible. I am certainly phobic about unnecessary resampling.