32 bit music

[adamdea]

Mostly the second law. It is all tied up with thermal noise and information entropy.

I have done some quick calculations. Assuming an amplifier input with 2V sensitivity into 50K Ohms running at room temperature:

32 bits at 6db/bit is a dynamic range of around 216dB. At full scale (32-bits), the amplifier input current is around 40uA. The current for 1 bit output is 9.3E-15A.

I calculate the Johnson noise current alone at 300K as 8.14E-11A, making just this noise current around 8,700 times greater than the signal.

To make use of 32-bit resolution, you would need to cool the whole of your audio system and your ears to below the temperature of liquid helium.

Don't try this at home!

not wishing to be arsey, but 32*6=192.
This makes me wary of the rest of your calculations. But it seems to be widely accepted that thermal noise in a dac is around -120 dB yielding 20 bits of effective snr. 24 bits is therefore excessive.

[chris217]

not wishing to be arsey, but 32*6=192.
This makes me wary of the rest of your calculations. But it seems to be widely accepted that thermal noise in a dac is around -120 dB yielding 20 bits of effective snr. 24 bits is therefore excessive.

Quite right. I must have had a thinkahead buffering problem in my brain and calculated 36*6. Apologies for the mistake.

[barnacle bill]

Quite right. I must have had a thinkahead buffering problem in my brain and calculated 36*6. Apologies for the mistake.

What would the new calcs be, still liquid helium cool?

[adamdea]

Quite right. I must have had a thinkahead buffering problem in my brain and calculated 36*6. Apologies for the mistake.

No worries, I shouldn't really have mentioned it, but was overcome with a pedantic urge. As a matter of interest do your figures agree with the -120db figure I have seen?

[RichG]

There are DAC chips, such as the Sabre 9018 (used by Weiss, W4S, etc) that have a quoted DR of 135dB. This will inevitably drop off to some degree when an output stage is applied, but in a good design can still be in the region of 128dB or so. IMHO this is way beyond audibility.

The Sabre chip also has 32 bit playback, however the main reasoning for this is not to allow input of 32 bit source files, but for signal processing (ie up-sampling / volume control) with minimal impact on SQ.

[tuga]

There are DAC chips, such as the Sabre 9018 (used by Weiss, W4S, etc) that have a quoted DR of 135dB. This will inevitably drop off to some degree when an output stage is applied, but in a good design can still be in the region of 128dB or so. IMHO this is way beyond audibility.

It's also beyond playbackability...at least with any of the domestic sound reproduction speakers that I've heard of, even for enourmous mamoths such as these:

Ric

[ncdrawl]

32 bit is commonly used in DAW workstations for heavy processing(32 bit floating point actually) and there are some advantages for that...but for playback? no.

[ncdrawl]

It's also beyond playbackability...at least with any of the domestic sound reproduction speakers that I've heard of, even for enourmous mamoths such as these:

Ric

what are those blue ones?

[Werner]

good design can still be in the region of 128dB or so. IMHO this is way beyond audibility.

With DACs, perhaps. As for ADCs: the best I've measured so far in real-world situations manage 108-110dB, unweighted, 20kHz band.

The Sabre chip also has 32 bit playback, however the main reasoning for this is not to allow input of 32 bit source files, but for signal processing (ie up-sampling / volume control) with minimal impact on SQ.

Even that is stretching it a bit (pun intended), and I am nearly confident that the very same chip architecture reduced to 24 bit would yield the same performance.
There is, however, a tiny shred of justification in the scenario where a signal processor preceeds the DAC chip. In such case it would be a shame (at least theoretically) to reduce the processor output to 24 bits, only to carry it over to the DAC's oversampling filter where internal word size is expanded to 48 bits or whatever. Might as well then throw 32 bit over the interface... it would give the chip and system designers a warm and fuzzy feeling.

[RichG]

what are those blue ones?

jbl 4350

[barnacle bill]

Sorry for dragging this thread back up.

The best 24bit music file (2L.no) I've analyzed mannaged a whooping 12bit or 72dB of dynamic range.

If a 24 bit file only has a 12 bits of info and a true dynamic range of 72db why not put onto a CD, wouldn't that reach a greater audience?

What's needed is a few recording engineers and industry "suits" who are willing to use a bit more than 6dB of it.

How does this work? Are the zero's in the digital signal just pads to bring the dynamic range to 96db's?

[browellm]

How does this work? Are the zero's in the digital signal just pads to bring the dynamic range to 96db's?

96dB is the theoretical maximum that the format can carry (actually it's a bit more as it's possible to record information below the noise floor).

No-one actually records, or would want to record on to a CD using the whole range. The mics used couldn't resolve it, ambient noise will also reduce what's actually available.

then you have to consider what's actually listenable at home. You don't want to be having to turn the volume control up and down all the time, and having concert hall dynamics isn't practical.

The best non-classical recordings (from a DR perspective) have about 15dB of range. Classical may even have a little more. Remember this is the actual difference between the quietest parts and the loudest parts of the recording, not the range available on the medium.

And this is why CD as a format (and indeed LP!) is perfectly fine (in terms of DR).

[tuga]

The best 24bit music file (2L.no) I've analyzed mannaged a whooping 12bit or 72dB of dynamic range.

If a 24 bit file only has a 12 bits of info and a true dynamic range of 72db why not put onto a CD, wouldn't that reach a greater audience?

Because......24bit files can be sold at a higher price...24bit files require new D/ACs...24bit files require new transports...24bit files require a lot of storage...24bit files require dedicated "bit-perfect" software...24bit files require experimenting with firewire/usb/s-pdif/toslink cables...and people audiophiles are willing to buy yet another copy of "Kind of Blue"



P.S.: and CD is dead anyway...

- - - Updated - - -

96dB is the theoretical maximum that the format can carry (actually it's a bit more as it's possible to record information below the noise floor).

No-one actually records, or would want to record on to a CD using the whole range. The mics used couldn't resolve it, ambient noise will also reduce what's actually available.

then you have to consider what's actually listenable at home. You don't want to be having to turn the volume control up and down all the time, and having concert hall dynamics isn't practical.

The best non-classical recordings (from a DR perspective) have about 15dB of range. Classical may even have a little more. Remember this is the actual difference between the quietest parts and the loudest parts of the recording, not the range available on the medium.

And this is why CD as a format (and indeed LP!) is perfectly fine (in terms of DR).

More than half of my "classical" music CDs have around 45-50dB of DR.
Since my room's noise-floor is somewhere near 35dB, the loud bits are played back quite loud and I presume that many non-audiophile people wouldn't like this.

[Werner]

If a 24 bit file only has a 12 bits of info and a true dynamic range of 72db

This is not what is happening. The definition of dynamic range depends on the context.

When discussing music, DR means the difference between the loudest passage and the quietest passage, where a passage itself means something of a few seconds or more (i.e. the gaps between the notes don't count).


When discussing signal theory DR means the difference between the loudest signal a channel can pass, and the channel's noise floor. No information can exist below the noise floor.

The quality reproduction of a musical programme with, say, a dynamic range of 20dB requires a channel with much much more dynamic range. Otherwise the gaps between the notes, or even the music's quieter parts, hug / drown in the channel noise.

[tuga]

Here's the example I mentioned before:

FLAC 24/96 Britten: Simple Symphony, Op. 4 (source http://www.2l.no/)

(p.s. saving the plot cropped the original which went down to 69dB)

[adamdea]

This is not what is happening. The definition of dynamic range depends on the context.

When discussing music, DR means the difference between the loudest passage and the quietest passage, where a passage itself means something of a few seconds or more (i.e. the gaps between the notes don't count).


When discussing signal theory DR means the difference between the loudest signal a channel can pass, and the channel's noise floor. No information can exist below the noise floor.

The quality reproduction of a musical programme with, say, a dynamic range of 20dB requires a channel with much much more dynamic range. Otherwise the gaps between the notes, or even the music's quieter parts, hug / drown in the channel noise.

If you go back to Barnacle Bill's point, the implication of this is that he is correct:-

The 2L recording would have sounded just the same in 16 bit, except that the quantisation noise would be higher.

For the reasosn you have given it would not work in say 10 bit, where the noise level would become intrusive, but the 16 bit version should be indistinguishable especially with psycho-acoustically correct noise shaping. I believe that there have been several studies showing exaclty what BB says ie that a proper 16 bit downsampling (what is the right verb -down quantising?) would be indistinguishable and that the contrary has never been demonstrated.