On dac-chip volume control guidance

[JohnYang1997]

In short with digital volume control, if you can hear noise at full volume, you will hear noise when fully attenuated.
With passive pot(for example), you will not hear noise when signal fully attenuated.

 

[bennetng]

Yes that’s why the video said some ESS (older chip?) dac noise floor was -135 dB. I was wondering what it was on newer ESS chips and also recent AKM dac chips - not finding this in spec sheets.

People who ask a question like this probably cannot hear their DAC's analog noise floor. If they can hear it there is no need to ask.

Here are some files:

original
Original 8-bit file at near 0dBFS (loud)

8-bit fade
Volume fade saved as 8-bit

16-bit fade
Volume fade saved as 16-bit

Now the noise is clearly audible.

None of the DAC measured here has 135dB DNR when using a -60dBFS input even in 2019, so ESS's measurement parameter/standard is probably different from the one that ASR uses.

[EDIT]
Every 24 or 32-bit DAC add zeros to the lower bits, otherwise volume will be different when playing audio data with different bit depths. Just like when you send a 1080p signal to a 4k TV, it will be scaled to full screen instead of showing the picture at a corner with 1/4 size. In the case of audio doing so would be disastrous because 24-bit data would be 48dB louder than 16-bit data.

Therefore apart from some typos and imprecise numbers, that ESS presentation is only describing a typical PCM digital volume control. There are block diagrams in some product datasheets showing where the volume control is located, for example:

https://statics.cirrus.com/pubs/proDatasheet/CS4392_PP3.pdf
https://statics.cirrus.com/pubs/proDatasheet/WM8740_v4.4.pdf

 

[bennetng]

See here for a direct comparison of Roon digital volume V an analogue pot.

Spot the difference.

https://www.audiosciencereview.com/...-dig-or-not-to-dig-that-is-the-question.5951/

What I witnessed is the analog pot created some spikes by itself, and turning down the pot reduced them. The noise floor remains the same.

 

[March Audio]

What I witnessed is the analog pot created some spikes by itself, and turning down the pot reduced them. The noise floor remains the same.

The 8kHz and 16khz spikes are usb packet noise, so not actually to do with the volume control specifically.

The noise floor will be changing but it's so low its below the noise floor of the measurement system.

So real world with normal sensitivity speakers it's a non issue.

 

[bennetng]

The 8kHz and 16khz spikes are usb packet noise, so not actually to do with the volume control specifically.

So did you find a way to avoid those noise when using that analog volume control?

 

[March Audio]

So did you find a way to avoid those noise when using that analog volume control?

It was a quickly thrown together test done without my USB isolator. It's just a measurement artefact, essentially a ground loop. The additional resistance in the shield wires of the volume box increases the voltage of those ground currents enough to make them just visible. Also the levels are about - 140dB. Completely inaudible.

 

[HammerSandwich]

If you specify the file format for 32bit float, yes, for 16 and 24bit PCM, I think no, from past experiments. I don't think the export/import is necessary when Audacity is set for 16 or 24 bit PCM.

Example:
Second track has gone -60dB and +60dB. No export/import, and no dither. Now it has flat tops and stair-steps

Wish I'd caught that. Thanks, Ray.

Below, showing the difference between a 16bit source, using 24bit integer volume control, and attenuating/amplifying by 60dB, results in a difference (noise?) between the 16bit input and 24bit output at least 80dB down.

Agreed 100%, even if my earlier post seems to disagree. Worrying about losing resolution this way misses the fundamentals of noise, gain staging, etc. You'll need a LOT of attenuation before even a 16-bit signal's SNR matches tape or vinyl. Unless your math has a bug, it's simply a question of noise versus gain. Same as in analog world, right?