Bit perfection

[solderdude]

It's a bogus story. It only means that during the upsampling process only the intermediate calculated values are 'invented' and the actual sample values are the ones used from the actual data. So a 4x oversampled 44.1 (let's face it, the majority of folks have mostly 44.1) only has 1 out of every 4 samples 'bit perfect' (that's what they mean with bit perfect) the following 3 are calculated.
This is sent to the DAC chip(s) and the output is sample and hold.
This means that 'bit-perfect' sample value is held 5.6us where in reality the original sample, when taken, was of a much, much shorter duration.
Then, these 'stairsteps' are followed by low-pass filter (to get rid of the stairsteps) which turns it more into a smoother signal.
At a specific point in time the actual output voltage will be exactly the same as the sample value suggests and the rest of the time it is not.

Compare that to any other DAC and these DAC's analog output too will have the exact same output voltage as the sample value dictates at a specific point in time.

So... the megaburritocombo filter story is just blabla where a lot of folks value the schiit story as 'relevant' and accept that 'explanation' as the reason why they subjectively prefer Schiit MB.
That and believing the Schiit filter is 'more accurate' and more 'musical' as all other resampling algorithms.
Something like freedom of religion exists so if one buys into advertising talk one is free to do so.

 

[Veri]

I guess if you are talking about "bit perfect" you could talk about multibit NOS DACs. Schiit claims that their multibit filter is also bit perfect though it does perform oversampling.

Black magic? Audio sorcery?

 

[Theo]

If I get it correctly, bit perfect is a safeguard against bad SRC or volume change software programming... If the SRC/volume change is done properly, there should be no audible difference with a "bit perfect" stream with an analog volume control...

 

[solderdude]

Not exactly. A statement something is bit-perfect is not a safeguard by itself. Doing a test for it is a safeguard. It ensures data is handled correctly.

When someone states something is 'bit perfect' it merely states that the data that has gone through the device has not been altered in any way.
A PC for instance can give 'bit perfect' output but as mentioned chances are it isn't. Depends on multiple factors on the PC.

When digital volume control is used the output by definition is not bit-perfect any more, but as also has been stated the sound quality may still be as good as when no digital volume control is used (when done properly).

SRC is not bit perfect by definition (the output differs from the input) but when done properly has no detrimental effect to the sound quality and may even 'improve' when the DAC in question, for instance' rolls off in the treble at 44.1 in an audible way but does not at say 88.2 or 176.4.

Bit perfect and not bit perfect can sound indistinguishable but when things are f'ed up they may not be.

 

[Don Hills]

...
This means that 'bit-perfect' sample value is held 5.6us where in reality the original sample, when taken, was of a much, much shorter duration.
Then, these 'stairsteps' are followed by low-pass filter (to get rid of the stairsteps) which turns it more into a smoother signal.
At a specific point in time the actual output voltage will be exactly the same as the sample value suggests and the rest of the time it is not.
...

True, the filter "smooths the stairsteps". I think this process deserves more credit than it usually gets.
As you said, the ADC takes "snapshots" of the input. Between those times, the input voltage continues to vary. The problem is that the DAC chip outputs the sample voltage continuously until the next sample, leading to the famous "stairstep". The magic of the filter on the output of the DAC is that it recreates the voltages that existed between the original samples. (In theory, perfectly. In practice, audibly perfect in any competent DAC.) So the people who prefer filterless DACs "because they are more accurate" are simply wrong.

 

[graz_lag]

I guess if you are talking about "bit perfect" you could talk about multibit NOS DACs. Schiit claims that their multibit filter is also bit perfect though it does perform oversampling. However you still have pre and post ringing with Schiit's filter. I'm not sure of NOS DAcs though.

What happens inside the DAC circuitry is out of scope as fas as the "bit-perfect definition" is concerned.
The DAC does whatever it needs to do to convert the signal that is brought in.
The "bit-perfect" term signifies the digital flow travels unaltered from the source (typically the PC) to the DAC's USB input, so with no equalization and/or DSP, with no upsampling, with no mixing, with no volume control, etc ...

 

[LF78]

DSP volume/EQ/etc., done properly, isn't an issue

This is correct, but how is it related to wanting a bit-perfect path though?

 

[sergeauckland]

This is correct, but how is it related to wanting a bit-perfect path though?

You can't have both. If you want bit-perfect from source through to DAC, then you can't have any DSP operating, no volume or EQ control as this will change the bits. If you want DSP, whether simple volume or including EQ, room correction or whatever, then it won;t be bit=perfect at the input to the DAC, but if done properly, then the effect entirely positive, as it does what one wants it to do.

The DAC in turn, 'just' converts the bits its given to an analogue voltage in a 1:1 manner to the limits of its performance envelope.

S.

 

[garbulky]

What happens inside the DAC circuitry is out of scope as fas as the "bit-perfect definition" is concerned.
The DAC does whatever it needs to do to convert the signal that is brought in.
The "bit-perfect" term signifies the digital flow travels unaltered from the source (typically the PC) to the DAC's USB input, so with no equalization and/or DSP, with no upsampling, with no mixing, with no volume control, etc ...

I understand. It's just that poster was talking about 5 bit DS dacs not being "bit perfect". Well if he's after that then there's other things that might meet his definition - at least a little closer.

 

[Theo]

True, the filter "smooths the stairsteps". I think this process deserves more credit than it usually gets.

This wouldn't work if it were not for the anti-alias filter in the ADC or used when down sampling, making sure that the "stairsteps" correspond to the proper frequency content, ie that the smoothness hypothesis is correct.

 

[Theo]

If you want bit-perfect from source through to DAC,

Do we actually need bit perfect ?

 

[sergeauckland]

Do we actually need bit perfect ?

Need is different to want.
Some people want it, whether needed or not.

S.

 

[graz_lag]

Do we actually need bit perfect ?

One can have it for free, so why not ?
It's a piece of mind ...
One needs a meaning of downstream volume control, as the one in the PC is locked and thus not available.

 

[HammerSandwich]

This is correct, but how is it related to wanting a bit-perfect path though?

Post 24 spells it out pretty well. (Thanks, @solderdude.)

 

[LF78]

You can't have both.

Well, yes. I want bit perfection unless I decide to alter the data as per my wishes (volume control, EQ filters, etc...). I just don't want additional data corruption or changes that I didn't apply explicitly.

 

[ahofer]

Bits is bits rebuttal - http://archimago.blogspot.com/2019/08/musings-demo-why-bits-are-bits-lets-not.html#more

 

[sajunky]

Refering to the OP, digital audio is just a transport and it is very easy to achieve bit-perfect transmission, so it is sane to keep bit-perfect whenever is possible. For two reasons: preserving original sound quality (1) and easy troubleshooting (2). There are cases when some other reasons are more important, by example when streaming, so for preserving uninterrupted streaming we are changing from reliable TCP delivery to less reliable UDP (which is also bit-perfect, but conditionally).

There also are other cases like unsufficient bandwith that leads to using compressing algorithms. As for MQA, mentioned too, it requires bit-perfect channel for a completely different reason. MQA stream carries encrypted information for copy protection and distribution control and this digitally encoded information has to be preserved. It has nothing about sound quality.

 

[Ron Texas]

Properly done, upsampling is transparent. Equalization isn't bit perfect, but properly done improves things, sometimes a lot.

 

[scott wurcer]

Bit perfect can have different meanings in different situations, I found some of the things in Schiit's "white papers" unprofessional calling the work of some early digital pioneers "pablum". There is simple bit perfect where you don't want SRC performed on an audio stream without being told (this is in fact not necessarily bad). As for any process I would consider it bit perfect if it was reversible with no loss at all. Of course that can't be so when information is reduced, but say upsampling from 44.1 to 48K could conceivably be totally reversible so that the exact same values are returned going back to 44.1.

 

[mansr]

Bit perfect can have different meanings in different situations

No. Two bit sequences are either identical or not. No ifs, buts, or maybes.

upsampling from 44.1 to 48K could conceivably be totally reversible so that the exact same values are returned going back to 44.1.

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.