Distortion (or not) of digital volume control.

[jan.didden]

I've been wondering some time now whether digital volume control is as good, or as bad, as they say. Depending on who you ask.
So I thought I'd measure the distortion of the digital volume control on my miniDSP FLEX digital.
Feed coaxial from the AP SYS2722 digital generator, output Toslink to the AP digital analyzer.
I measured at 0dB with 1dBV input and with the control turned down to -20dB and -30dB.
Frequency slightly offset to see the individual traces.
I can't find any flaw with it. (24 bit, 48kHz, triangular dither).

Comments?

Jan

 

[RayDunzl]

My experience with digital volume control:

By the time you turn it down far enough to create a SNR problem you can't hear the program any more anyway.

 

[kemmler3D]

Digital volume control will generally work at 24 or 32 bits I think? So you can get pretty quiet before the noise starts to get audible. I think even if you at 16 bits, you really need to throw away quite a few bits before noise becomes objectionable. 12 or so is still better than you think.

 

[RayDunzl]

Long ago...

While considering "losing bits" with volume control...

I went for the worst case scenario.

I took a "well known" recording and reduced it to "one bit" - if the signal was positive it went to the "up" level, or negative to the "down" level.

Basically just capture zero crossings and one level positive or negative representing one digital bit.

A short segment of the waveform is seen in the background while it plays.

https://soundcloud.com/ray-880875693%2Fonebitpaul
I thought it was remarkably intelligible.

In fact, this is the longest version I was able to get past the hall monitor that prevents you from posting known commercial tunes - longer versions were rejected for copyright violation by their recognition software.

I didn't dispute the claim.

 

[jan.didden]

@RayDunzl yes you're probably right. Harmonics at -165dB wrt 1V?
Still more than 100dB SINAD.
As to your 1bit file, if you would dither that you'd probably just hear a noisy track.

Jan

 

[RayDunzl]

As to your 1bit file, if you would dither that you'd probably just hear a noisy track.

Well, I think it's just crude representation of one bit, in that there are only two levels (three if any of the original samples was zero) raised to an audible level.

Dither would just wiggle the flats around their level for the most part.

SoundCloud performed some compression algorithm and maybe dithered it themselves, since the corners as displayed are no longer the original 90 degree square.

Leading edges look pretty square but the trailing edges don't.

And dither is too small to see at this scale.

 

[danadam]

As to your 1bit file, if you would dither that you'd probably just hear a noisy track.

I think the way he generated this file was by applying such a high gain that every sample clipped, so I don't see how dither could be used there.

And if you wanted to generate such 1-bit file really by lowering the volume, then dither is a must, because without it you are left with only a few non-zero samples, in places where the original file had peaks, and the rest are all zeros.

 

[RayDunzl]

I think the way he generated this file was by applying such a high gain that every sample clipped, so I don't see how dither could be used there.

Correct.

Digitally amplified (maybe it was +100dB), until any sample positive or negative went full scale, then the volume reduced to a listenable level.