Seems to be that with 42dB of attenuation, any competent digital attenuator should simply move signal 42dB closer to the noise floor without introducing any non linearity or loss of bits, that is after all what dither is for.
If (And it should be a given) the word lengths are correctly scaled and any word length reduction is correctly dithered, then there is no loss of fidelity apart from that due to moving closer to the system noise floor.
I don't doubt that there is sand out there that gets it horribly wrong, but any competent designer can come up with a digital gain control that is LINEAR all the way down into (and even below) the noise floor.
Regards, Dan.
This solution will add more distortion, but it might be distortion you like (I know I do 🙂 ).
You could use a transformer to do the attenuation, for example the Sowter 9150.
You could use a transformer to do the attenuation, for example the Sowter 9150.
For a 16-bit system, dither will move the (already just moderate) noise floor upwards as well, worsening the result regardless of the 'correctness' of the dither. Dither always does that by its very nature. And you wouldn't know which manufacturer enabled dither and which left it off. You'd be surprised at a few big name brands that faltered at these basic tasks.
With a 32-bit control this will not be an issue because the result would still be better than the analog part of the system. We seem to be unable to exceed 21 ENOB in any kind of analog output stage, so anything above 23-24 bits effective is just fine.
Loss of dynamic range is an issue with both digital and analog volume controls, and is a result of poor system gainstaging. The attenuation required here seems excessive, and a buffered volume control is one way out of that, but it introduces many new active devices and power supplies in the chain.
If you go into the gain control multiplier as say 16 bit with maybe something like 1.15 for the control port then you get a 32 bit signed value out of the multiplier. If you then dither this to say 24 bits (so it fits in an standard AES stream, more then is really useful) then you have lost pretty much nothing.
Even 21 ENOB is a biggish ask, but I know my amps/room/speakers/ears are not really 21 ENOB (123dB range approximately), so why worry about it?
Regards, Dan.
Even 21 ENOB is a biggish ask, but I know my amps/room/speakers/ears are not really 21 ENOB (123dB range approximately), so why worry about it?
Regards, Dan.
It is important to know your definitions and how to read specs.
There is no such thing as a 32-bit DAC (and you would be hard pushed to even get to 24 bits for typical audio levels) and there is never going to be one.
The reason is simple - assuming the limits of analog when it comes to the smallest signals not requiring exotic technology, we are at around -128dB referred to 1V which is about 1/4 of a uV RMS noise, or around 22 bits of resolution assuming 1V RMS as full scale. It is EXTREMELY hard to push the noise down, so what one is left with is increase the full scale. Fitting 32 bits into this gives us about 1kV full scale. Very impractical and also futile - unless you want to kill your listeners with a dynamic range below audibility on one end, and obliterate the city block they are in, on the other.
32-bit means that 32-bit digital processing is used in order to maximally preserve the actual dynamic range of the final digital to analog conversion, i.e. prevent rounding and other cumulative and non-cumulative errors creeping up over the analog noise floor. The relevant figure is the dynamic range of the DAC, and the best case for a SABRE chip is around 134dB referred to 4V rms full scale, when it is configured in mono differential mode. Any further analog processing will only make it worse. So what is it in bits? Well, a bit less than 23. And remember, this is bend-over-backwards cutting edge.
Add 42dB, i.e. 7 bits of attenuation, and you have just pushed 7 bits out of 23 into the noise floor (which is a good result, given that attenuating a 24-bit input signal by 7 bits still leaves one extra bit out of the 32-bit processing width, so there is no loss in the digital domain), and you are left with 16 bits. Which is fine if your sources are CD based.
It could be convincingly argued that even with 24-bit material, where you lose 1 bit to begin with and 7 more at 42dB attenuation, you are unlikely to get even 16 real bits of real dynamics, let alone more, which has to do with how recordings are made in the real world and limitations on things like microphone preamps. Using some simple analysis tools in a sound editing program will tell you a lot here. But the main thing to remember is that 24 bits was originally an upgrade from 16, for the very same reason 32 is an upgrade from 24 - to prevent processing degradation from creeping up into the usable dynamic range.
That being said, using a tailor-made analog attenuator MAY get you a better result assuming the rest of your system has a real >16 bits dynamic range, but given that it's gain is high (or you would not need attenuation in the first place), this is a real question. A good attenuator will introduce some noise and distortion but it is unlikely to do so at -96dB or so (equivalent to 16 bits).
There is no such thing as a 32-bit DAC (and you would be hard pushed to even get to 24 bits for typical audio levels) and there is never going to be one.
The reason is simple - assuming the limits of analog when it comes to the smallest signals not requiring exotic technology, we are at around -128dB referred to 1V which is about 1/4 of a uV RMS noise, or around 22 bits of resolution assuming 1V RMS as full scale. It is EXTREMELY hard to push the noise down, so what one is left with is increase the full scale. Fitting 32 bits into this gives us about 1kV full scale. Very impractical and also futile - unless you want to kill your listeners with a dynamic range below audibility on one end, and obliterate the city block they are in, on the other.
32-bit means that 32-bit digital processing is used in order to maximally preserve the actual dynamic range of the final digital to analog conversion, i.e. prevent rounding and other cumulative and non-cumulative errors creeping up over the analog noise floor. The relevant figure is the dynamic range of the DAC, and the best case for a SABRE chip is around 134dB referred to 4V rms full scale, when it is configured in mono differential mode. Any further analog processing will only make it worse. So what is it in bits? Well, a bit less than 23. And remember, this is bend-over-backwards cutting edge.
Add 42dB, i.e. 7 bits of attenuation, and you have just pushed 7 bits out of 23 into the noise floor (which is a good result, given that attenuating a 24-bit input signal by 7 bits still leaves one extra bit out of the 32-bit processing width, so there is no loss in the digital domain), and you are left with 16 bits. Which is fine if your sources are CD based.
It could be convincingly argued that even with 24-bit material, where you lose 1 bit to begin with and 7 more at 42dB attenuation, you are unlikely to get even 16 real bits of real dynamics, let alone more, which has to do with how recordings are made in the real world and limitations on things like microphone preamps. Using some simple analysis tools in a sound editing program will tell you a lot here. But the main thing to remember is that 24 bits was originally an upgrade from 16, for the very same reason 32 is an upgrade from 24 - to prevent processing degradation from creeping up into the usable dynamic range.
That being said, using a tailor-made analog attenuator MAY get you a better result assuming the rest of your system has a real >16 bits dynamic range, but given that it's gain is high (or you would not need attenuation in the first place), this is a real question. A good attenuator will introduce some noise and distortion but it is unlikely to do so at -96dB or so (equivalent to 16 bits).
- Status
- Not open for further replies.