DAC or Streamer with volume controlled AES/Coax output

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hello,


I am searching for an Preamp/DAC/Streamer with USB-Audio-Input and an digital AES or Koax-Output where the volume of the digital output can be controlled digitally in 32bit or higher.
There are a lot of streamers with digital outputs but none of them can control the volume of them especially not in 32bit (lossless) or higher.
Is there any product available which can do that or maybe any diy-solution how to do that?


The background is that I want to build a new active speaker system build on the new hypex fa-modules (fa253/503). To avoid doulbe AD-DA-converstation I want to use the digital inputs (aes or koax) on the Fusion-Amps. Therefor I need an good volume control on the source (32bit or higher).
 
Strom Audio

I was in contact with the Storm Audio support and they told me that volume is controlled in DSP chain which is floating point 32bit. Signal is then getting converted to PCM and output in 32bit via AES/EBU.
They told me that now it depends if your AES input can handle 32bit if not will be "downsampled" to 24bit.
 
ok I have read that if you we are doing the volume control digitally you will lose 1bit per 6db attentuation.


So in case the Storm-Audio will do volume control in 32bit FLOAT in the DSP-chain but the AES-Input (hypex fusion) only allows 24bit. So let´s make an example when I play 16bit file. This file gets upsampled in the DSP to 32bit and volume control will be made here in 32bit.
Then the file gets downgraded to 24bit because AES-Input can only handle 24bit. Right?



Regarding the 1bit/6db loss rule I mentioned above.
What is now the calculation starting point for calculating the bit-loss for the attentuation? The source-file (16 bit), the dsp chain/volume control (32bit) or the ouptut bitrate (24bit)?
Maybe someone with more technical background can bring a bit more light into this confusing scenario.


In general to the active speaker guys here. How are you doing your volume control? Digitally at DSP-step or analog after DA-converstation?
 
I was in contact with the Storm Audio support and they told me that volume is controlled in DSP chain which is floating point 32bit. Signal is then getting converted to PCM and output in 32bit via AES/EBU.
They told me that now it depends if your AES input can handle 32bit if not will be "downsampled" to 24bit.

Does anyone know if there has been some fairly recent addition to the AES3 standard to enable 32 bit audio? A subframe always had 32 bits, but eight of them were reserved for status data.
 
ok I have read that if you we are doing the volume control digitally you will lose 1bit per 6db attentuation.


So in case the Storm-Audio will do volume control in 32bit FLOAT in the DSP-chain but the AES-Input (hypex fusion) only allows 24bit. So let´s make an example when I play 16bit file. This file gets upsampled in the DSP to 32bit and volume control will be made here in 32bit.
Then the file gets downgraded to 24bit because AES-Input can only handle 24bit. Right?



Regarding the 1bit/6db loss rule I mentioned above.
What is now the calculation starting point for calculating the bit-loss for the attentuation? The source-file (16 bit), the dsp chain/volume control (32bit) or the ouptut bitrate (24bit)?
Maybe someone with more technical background can bring a bit more light into this confusing scenario.


In general to the active speaker guys here. How are you doing your volume control? Digitally at DSP-step or analog after DA-converstation?

Your 16-bit data would get converted to floating point without any losses, just the same data represented differently. The 16-bit data have a quantization noise floor of about -93 dB with respect to full scale when the recording is made with dithering, so the same holds after the conversion to floating point.

After volume control, rounding the result to 32-bit floating point (assuming a 24-bit mantissa, see Single-precision floating-point format - Wikipedia ) would add a quantization error that is always some 146 dB below the signal. That is, when the signal gets soft, so does the quantization error.

Rounding to 24 bits then adds another quantization error. If done without dithering, this one is around -146 dB with respect to full scale, so it is very soft but doesn't get any softer when the music does. With dither, the error would be around -141 dB, but it would resemble extremely soft noise rather than extremely soft distortion.

All in all, when the attenuation of the volume control is less than 48 dB, the biggest source of quantization errors is the 16 bit format of the original data. With more than 48 dB of attenuation, the other quantization errors start to dominate.
 
thanks for your detailed explanation. I don´t really understand all of your technical stuff, but what I can hear is that if this combination of digital volume control (FLOAT 32bit and the 24bit output to AES) is done with dithering it could be really audiophile solution and no need for a analog volume control?

As I understand you it will maybe be just an "issue" when attentuate more than 48db. But for example you have an speaker with let´s say 100db/1w/1m and attentuate by 48db it will result 52db/1m and then maybe in 3 meters away you will lose 8-10db more which results in 44db at listening position. I don´t think that someone will hear music at such low volume.


The storm audio was just an example. I have now just seen the new "MiniDSP SHD Studio" which seems to be the optimal stereo preamp to connect digitally (AES) with the fusion amps (fa503).
I was already in contact with MiniDSP and they are doing the volume control in the DSP with 32bit FLOAT with dithering and will then output 24bit to AES. So this seems to be a very good choice. Another cool feature is that is an ARM processor with Volumio installed and Dirac Live is also included.
 
I think your calculation is not correct, mainly because those Hypex amplifiers have a maximum power well above 1 W. I'm not sure whether my calculation below is correct either, but at least it should not be too far off.

Suppose you have a mono system with two 250 W amplifiers (woofer and squacker) and a 100 W amplifier (tweeter). When playing back a full-scale sine wave at full volume, it would either end up in one of the loudspeakers at full volume or in two at less than full volume, if it happens to be close in frequency to a crossover point.

A 250 W sine wave into a 90 dB at 1 W, 1 m loudspeaker would produce a sound pressure level of 114 dB at 1 metre distance. With two channels (stereo), the sine waves of left and right may add in phase, in antiphase or anything in between, but more practical audio signals will normally add in power. Hence, the maximum SPL increases by 3 dB to 117 dB at 1 metre in free space.

In a room you have reverberation. With a reverberation radius of 1 metre, the sound pressure level at large distance from the speakers would again be 117dB SPL.

Subtracting 48 dB, the level below which the quantization of the 24 bit output dominates over the quantization of the 16 bit source would be 69 dB SPL. That's actually the SPL of a full-scale sine wave played back at the same volume setting as whatever you are listening to. Music will typically have a higher crest factor than a sine wave, so maybe it will be 50 dB to 60 dB SPL of music.

A more useful way to look at it: with a full-scale level of 117 dB, the 141 dB dynamic range of a dithered 24 bit digital interface would translate into a -24 dB SPL quantization noise floor. That's way below the threshold of audibility, so quantization noise of the 24 bit interface will never be a problem, no matter what volume you listen at.

About dither: I'm in favour of dithering because rounding with dither sounds much better than rounding without dither in those cases where the quantization error is audible. For a 24 bit interface it is mainly of academic interest, because with or without dither, it is very unlikely that you will ever hear the quantization error of the 24 bit interface, simply because it is so soft. If you should ever be able to hear the quantization error, when you have your ear practically touching the tweeter in a perfectly quiet environment while playing extremely soft music, then it will sound slightly louder but much less annoying with dither.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.