Hi,
So say you have a 24/96 audio file. 24-bit provides 144dB of dynamic range between the noise floor and 0dBFS.
Say I then use a software DSP to break that into LF/MF/HF. Let's say, as an example, 20-400Hz for LF, 400-3000Hz for MF and 3000-20000Hz for HF.
Each of the three DAC channels can do 24/96.
Am I barking up the wrong tree here, does that give us stupid (wasted) amounts of dynamic range per channel?
16-bit provides 96dB of dynamic range between the noise floor and 0dBFS, so should the DAC channels just use 16-bit then?
Then there is DSP volume and volume levelling, my basic understanding of these comes from Roon and I seem to remember it converts the audio bitrate to a 64-bit float internally (I think CDSP is similar) and therefore DSP volume is lossless (because the dynamic range is huge) as far as human hearing is concerned?
I don't really understand how DSP volume and volume levelling impacts the DAC output bit depth? On my DAC8X a post by Renne over at ASR was complaining in the "Multichannel audio on a Pi will get a whole lot easier and cheaper!" thread that the DAC8X was only 24-bit and in their words he couldn't use it's 24-bit limitation because "24-bit dynamic + 8-bit volume control = 32-bit".
Apologies of this doesn't make any sense, I'm learning.
Thanks!
So say you have a 24/96 audio file. 24-bit provides 144dB of dynamic range between the noise floor and 0dBFS.
Say I then use a software DSP to break that into LF/MF/HF. Let's say, as an example, 20-400Hz for LF, 400-3000Hz for MF and 3000-20000Hz for HF.
Each of the three DAC channels can do 24/96.
Am I barking up the wrong tree here, does that give us stupid (wasted) amounts of dynamic range per channel?
16-bit provides 96dB of dynamic range between the noise floor and 0dBFS, so should the DAC channels just use 16-bit then?
Then there is DSP volume and volume levelling, my basic understanding of these comes from Roon and I seem to remember it converts the audio bitrate to a 64-bit float internally (I think CDSP is similar) and therefore DSP volume is lossless (because the dynamic range is huge) as far as human hearing is concerned?
I don't really understand how DSP volume and volume levelling impacts the DAC output bit depth? On my DAC8X a post by Renne over at ASR was complaining in the "Multichannel audio on a Pi will get a whole lot easier and cheaper!" thread that the DAC8X was only 24-bit and in their words he couldn't use it's 24-bit limitation because "24-bit dynamic + 8-bit volume control = 32-bit".
Apologies of this doesn't make any sense, I'm learning.
Thanks!
One thing - you need to differ between recorded music bit depth and the replay system bit depth capability - these are of course not the same thing. I can record with 8 bit data and replay in a 24 bit system, this system will play you 8 bit music resolution. If you reverse the situation, a 24 bit recording in a 8 bit system - you will hear an 8 bit system.
Volume control in the digital domain don't change the length of the replay system, it rather operates on the music samples and as you decrease the level, music carrying bits "disappear" For every 6 dB of attenuation, 1 bit disappear. One must however realise that as one lower the volume, the music starts to disappear into silence/noise so you simply don't need 24 bits of music data when you listen at 55 dBA SPL.
When you split a stereo channel into several ways, the bit depth of the system is retained in the different ways so no loss there - typically. Why limit the channel DACs to 16 in this day and age. No reason even if it could be OK...?
I would not be to to "bit" concerned. If you build a 24 bit reproduction system in all channels you will be fine. It's probably good if your DSP has an internal precision of 32b so that the calculations there don't waste any of the music precision.
Think about this.... a really quiet room is perhaps 30 dBA and a really loud sound pressure at mic is 115 dBA - this is 85dB worth of dynamics - its and extreme situation in an empty concert hall with a symphony orchestra playing at full blast - mic in 2nd row.. If you haven't heard this I think you should visit one.
Typical albums have perhaps 20 or maybe 30 dB of dynamics. Bits isn't the challenge... but don't be sloppy with them 😉
Good luck!
//
Volume control in the digital domain don't change the length of the replay system, it rather operates on the music samples and as you decrease the level, music carrying bits "disappear" For every 6 dB of attenuation, 1 bit disappear. One must however realise that as one lower the volume, the music starts to disappear into silence/noise so you simply don't need 24 bits of music data when you listen at 55 dBA SPL.
When you split a stereo channel into several ways, the bit depth of the system is retained in the different ways so no loss there - typically. Why limit the channel DACs to 16 in this day and age. No reason even if it could be OK...?
I would not be to to "bit" concerned. If you build a 24 bit reproduction system in all channels you will be fine. It's probably good if your DSP has an internal precision of 32b so that the calculations there don't waste any of the music precision.
Think about this.... a really quiet room is perhaps 30 dBA and a really loud sound pressure at mic is 115 dBA - this is 85dB worth of dynamics - its and extreme situation in an empty concert hall with a symphony orchestra playing at full blast - mic in 2nd row.. If you haven't heard this I think you should visit one.
Typical albums have perhaps 20 or maybe 30 dB of dynamics. Bits isn't the challenge... but don't be sloppy with them 😉
Good luck!
//
Last edited:
Apologies if this answer just adds to the confusion, but:
The dynamic range of modern sigma-delta a.k.a. delta-sigma audio DACs with a 24 or 32 bit input interface is not determined by 24 or 32 bit quantization noise. Their dynamic range is actually considerably smaller.
The digital signal doesn't directly drive DAC elements, like it used to in 1980's 14 to 20 bit DACs. Instead, it is digitally processed into a signal with a very high sample rate and very small word length that nonetheless has only little noise in the audio band (noise shaping). That signal gets converted to analogue.
The audio band noise is usually dominated by the analogue circuit noise. That makes sense, because adding a couple of bits to the digital part is cheap in terms of chip area and power, while reducing the analogue noise is not.
On top of that, quantization noise isn't really additive noise, but rather a weird kind of distortion that sounds like noise when certain conditions are met, and sounds pretty awful when those conditions are not met. It is therefore nice as well as cheap to reduce it to negligible levels. (There are ways to make quantization noise sound similar to additive noise without imposing conditions on the signal, this is called nonsubtractive dithering.)
The discussions about how many bits you need for digital volume control without requantization are largely academic, as long as the requantization noise is well below the DAC noise. Besides, when you apply filtering, like you do, you normally need to requantize anyway because of the filter's bit growth.
Bottom line: I agree with TNT. Just use DACs with a low enough noise floor and use plenty bits, where 24 should be plenty.
The dynamic range of modern sigma-delta a.k.a. delta-sigma audio DACs with a 24 or 32 bit input interface is not determined by 24 or 32 bit quantization noise. Their dynamic range is actually considerably smaller.
The digital signal doesn't directly drive DAC elements, like it used to in 1980's 14 to 20 bit DACs. Instead, it is digitally processed into a signal with a very high sample rate and very small word length that nonetheless has only little noise in the audio band (noise shaping). That signal gets converted to analogue.
The audio band noise is usually dominated by the analogue circuit noise. That makes sense, because adding a couple of bits to the digital part is cheap in terms of chip area and power, while reducing the analogue noise is not.
On top of that, quantization noise isn't really additive noise, but rather a weird kind of distortion that sounds like noise when certain conditions are met, and sounds pretty awful when those conditions are not met. It is therefore nice as well as cheap to reduce it to negligible levels. (There are ways to make quantization noise sound similar to additive noise without imposing conditions on the signal, this is called nonsubtractive dithering.)
The discussions about how many bits you need for digital volume control without requantization are largely academic, as long as the requantization noise is well below the DAC noise. Besides, when you apply filtering, like you do, you normally need to requantize anyway because of the filter's bit growth.
Bottom line: I agree with TNT. Just use DACs with a low enough noise floor and use plenty bits, where 24 should be plenty.