DAC with bit perfect volume control

[earlevel]

This below is an excerpt from the jriver wiki about volume control in their media center.

The precision offered by Media Center's 64bit audio engine is billions of times greater than the best hardware can utilize. In other words, it is bit-perfect on all known hardware...

Oh, someone actually touts this as a feature in a wiki

To demonstrate the incredible precision of 64bit audio, imagine applying 100 million random volume changes (huge changes from -100 to 100 dB), and then applying those same 100 million volume changes again in the opposite direction.

Amazingly, you will have the exact same signal at 32bit after 200 million huge volume changes as when you started.

As if there is an actual need for this

In other words, this incredible number of changes results in a bit-perfect output at 32bit, which is the highest hardware output bitdepth (most high-end hardware is 24bit).

And...yes, as if 32-bit converters exist...or that we can even do 24...

Please pardon the diversion, I thought "bit perfect" could only be trivia for discussion boards. But it's wiki material. Next thing you know, it will be an escalating battle between manufacturers—"with our proprietary extended precision volume control, you could do billions of volume changes and undo them and it would come back exactly, even for future 48-bit converters."

OK, anyway, thanks for another definition of "bit perfect". None of them are important, but at least I know there is another claim. I'm most disturbed that some people would think this matters in any way, while there are so many other things in the chain that can actually be heard.

 

[bennetng]

"Bit-perfect" by definition should be something like lossless codecs, which means reversible.

Here's an intuitive interpretation of that:

If you start with a 16-bit source and you have a 24-bit volume control that dithers the LSB, then you've added 7 bits to the "right" of the 16 MSB (OK maybe 7.5 bits but let's say 7 for simplicity). That should give up to 7*6 = 42 dB of attenuation before the attenuation changes any of the original 16 MSB in a way that is not perfectly reversible.

In this case, if you digitally attenuate it by anything less than 42 dB, when you turn it back up again you have the exact same bits.

The difference is that lossless codecs decode to the original bit-depth, with the original values, a digital volume control doesn't. It is only reversible if you turn up the volume with the exact amount, and convert to the original bit-depth again without using dither, it is not what happens in a DAC.

Your later comments also mentioned about something like 30dB attenuation. Of course how much attenuation is needed depends on audio content and equipment matching. If unfortunately 30dB is the absolute minimum required attenuation then an additional analog attenuation stage makes sense. One can even use a fixed 30dB analog attenuation and do the rest with digital if their DACs don't work in the RME way. I only use digital attenuation and my volume range is about 10-20dB, with ReplayGain enabled in the player and -15LUFS as my preferred average volume, none of the music I listen to uses more than 30dB attenuation. Also the benefit to avoid floating point induced clipping. No matter how many bits a fixed point volume control is, if 0dBFS is aligned to the MSB, then it will always clip when playing content with 0dBFS+ values. What contents can contain 0dBFS+ values? Download and try to play with this:
https://1drv.ms/u/s!AvzB71jO7t0-gYwfVGulbuFbVFZcmg?e=uN2p8Z

It also comes with the question about the used dynamic range in audio materials. Contents with higher dynamic range often require less attenuation, and vice versa, so the actual dynamic range is also compensated with this. For example, it would be rather pointless to use something like Merzbow to justify the constant use of high attenuation.

 

[bennetng]

denying reference threads.

Indeed I did not, because like I said, I consider that topic to be a reference on that matter

If you denied others' references, why blame others also denied your references? It is mutual.

 

[MRC01]

... We listen to the changed volume, which would not have the same bits. ...

The real question is, was there any loss of information? Just cuz the bits aren't the same, doesn't necessarily mean you've lost information.

 

[chris719]

The real question is, was there any loss of information? Just cuz the bits aren't the same, doesn't necessarily mean you've lost information.

I’m not sure that’s the real question. Anytime you attenuate below the actual full (true) resolution of a converter you have “lost information”. What’s a few LSBs among friends, though.

 

[MRC01]

Ah, but it's useful to know what that threshold is, where the control becomes lossy. It may be so wide you never hit it, or it could be so narrow you always hit it. Whether or not it's actually audible, most folks here are intellectually curious and just want to know.

 

[chris719]

Ah, but it's useful to know what that threshold is, where the control becomes lossy. It may be so wide you never hit it, or it could be so narrow you always hit it. Whether or not it's actually audible, most folks here are intellectually curious and just want to know.

That depends on the resolution of your converter. You cannot represent a 20-bit number with less than 20-bits, for example.

 

[Blumlein 88]

That depends on the resolution of your converter. You cannot represent a 20-bit number with less than 20-bits, for example.

Quite a few DAC chips are accepting 32 bit inputs. Yes, they can in no way match that performance at the output. But there you go. It appears that sigma delta DACs hit correct output all the way to 24 bits even though it may be in noise at the output. I don't know if any do better than that or not.

Maybe an interesting test to run if I can get 26 or 27 bit outputs within the noise of a DAC at the output.

 

[RayDunzl]

The real question is, was there any loss of information?

If I turn down the volume,, the quietest sounds recede from audibility, so, though the "information" may not be lost to a meter, it will be lost to me.

 

[MRC01]

If I turn down the volume,, the quietest sounds recede from audibility...

Not at first, but if you keep turning it down lower, at some point they will. But at what point exactly? Different thresholds:
1. The quietest sounds become imperceptible
2. The quietest sounds are masked by analog noise
3. The quietest sounds are lost digitally (data gets randomized or truncated)
Ideally, the thresholds happen in the above order. But in non-ideal systems they could happen in any other order.