[Glassman]

there is no 'digitize in 24bit', there are no 24bit ADC at all, everything is low bit delta sigma, which means anywhere from 1-5bit operating at anywhere between 3-6MHz, the resulting PCM is just a product of digital filtration and decimation, it's just as easy to output 16bit as it's to output 32bit or whatever, but it says nothing about how precisely was the analog input captured, the reason for using 24bit as an output format is just that there's no chance there's any useful information past 24bit and that 24 is exactly three bytes.. it's that simple.. it also makes no difference to output 48kHz, 96kHz or 192kHz and in fact whatever frequency less that the actual frequency of the sigma delta modulator is running at, but why to do that when there is nothing but ultrasonic noise at those frequencies? again, it's that simple..

 

[Syzygies]

Quote:

Originally Posted by rickcr42 http://www.digido.com/portal/pmodule...er_page_id=27/ Easy to understand language,very non technical,dead on

This article does a great job of explaining dithering. I take issue with one claim: If one is working alone as a complete "mastering studio", sometimes one has to manipulate the sound at 24 or 32 bits before dithering down to 16 bits, against the advice of the article.

In reading various assertions in this thread, remember a crucial rule of science: We are here to use our analytical skills to explain what we empirically observe, not to will onto the universe what must be true because it stands to reason. Too many arguments made in this thread fall into the second category.

What my ears heard knocked me flat on my back, the difference is dramatic. Whatever the actual mechanism in D to A and A to D converters, the "so-called" 17th bit carries significant audio information. This is an empirical statement no one can wish away by pure reason.

 

[tangent]

This discussion is diverging from my point, and some of you have missed it entirely. (Probably my fault...not trying to irritate anyone by pointing it out.) I am not saying that 16 bits is sufficient. I simply want people to be aware of what it means to add just one bit. A particular 20 bit DAC may indeed sound a lot better than a different 18 bit DAC, but I'll bet more is at work than the number of bits the DAC chip can play with.

Rick's comment about 48 bits is exactly the sort of thing I'm attacking here. Relative to a 1V signal, 48 bits allows for over a 5000 dB dynamic range! And 96 bits is just unspeakably silly. That allows for a number something like 16 orders of magnitude higher than needed to count all of the atoms in the universe.

 

[Glassman]

some additional comments:

sound intensity is not linear, but logarithmic, just as light intensity and such, so the linear scales like distances or weights might look impressive, but are not really appropriate when comparing to hearing..

96bit A/D is a piece of cake just like 24bit A/D, the problem is you will only have random noise in the 'lower' 75bits

the 24bit output of 24bit A/D converter is already just a result of mathematical computations, processed at certain precision, most likely 32bit and then rounded to the nearest whole-byte bit depth that makes sense.. less than 24bit is nonsense as there is useful signal there and more than 24bit is just as well nonsense as there is no useful signal at all.. I know you are an analog guy, but you should know what's under the hood of the digital devices Rick

24bit digitization is approaching physical limits most likely, it makes more sense to focus on the interstage precision losses when doing math to the signal.. I'd say 64bit precision for filtering the delta sigma modulator output could make sense and all the following processing steps also.. just to reduce even remote possibilities of signal degradation because of rounding losses..

 

[Hibbsy]

Tangent - very nice.

Now, all of you, get back to analogue!!

 

[rickcr42]

Quote:

Rick's comment about 48 bits is exactly the sort of thing I'm attacking here. Relative to a 1V signal, 48 bits allows for over a 5000 dB dynamic range! And 96 bits is just unspeakably silly. That allows for a number something like 16 orders of magnitude higher than needed to count all of the atoms in the universe.

would only be silly in the contest of the playback,the DAc side.
Remeber what digital is and that is a sample of the analog signal and not ALL of that signal.When the headroom on the recording side is considered there is no headroom at all.Zip
Unlike anlaog which will either compress a signal or distort that signal,sometimes in a good way if distortion can be considered good,digital just throws out a nasty square wave.
So you already need to begin by losing bits by keeping the signal well below the 0db point or a peak will destroy the entire recording.If this "headroom" is set to -6db you now have 18 bits with a 24 bit ADC and even then you better have a peak limiter somewhere.Music is not a steady state lab test tone but a dynamic and unpredictable beast.
At the other end you do not want to be at the last bit ,the noise floor so again you lose bits in reality,not in the lab or some guy "paper" who's job it is to convince you why the new chip is the ultimate.
I can flood this joint with data sheets that tell one story while the actual sound tells another story entirely and that real story not a good one.
24 bits is barely adequate in the real world for recording.

Can a 16 bit or 18 bit DAC sound better than a 20 or 24 bit DAC ?
Obviously.Implementation is everything and a comprimised design one well will kick the hell out of a potentiallly surperior design done so so.Everything involved has a sound and when built to sound good over what looks good on the scope only you most times have a better end product.why else do you think three DACs using identical chips have such varying performance in actual system use ?

Not the bits but the design.Most of the best DACs I have heard use 16 bit chips and I use no higher than 18 bits though i have done 20 and 24.Why 18 ? Why 16 ? because along with the bits comes all the extra garbage that large scale integration allows these knucklheads to put on the chip and because it is large scale and must keep both the power down and the heat low that means analog done in cmos.

A simple 16 bit DAC done with a simple Class-a output stage will sound more true to the actual music than the etched sound (digititis) of the modern 24 bit DACs unless the designer intentionally builds an output stage to correct the faults of the chip design.Kinda seems backwards to me.

just my opinion though don't mean squat

 

[hegestratos]

Quote:

Originally Posted by tangent That's something like 16 orders of magnitude higher than needed to count all of the atoms in the universe. [...] With 96 bit audio, you could record the universe at atomic scale.

There is a lot more than 96 bits of information in the universe, thanks. Counting the number of particles in the universe and describing what they're doing is something entirely different. First of all, with 10^79 particles in the universe, you'd need 263 bits to label them all uniquely. You'd need to write down 10^79 of those 263-bit numbers to do it. And by then, you've only given them an identifier.

In everyday life, we're used to working with linear scales. Our ears don't measure sound intensity linearly. I'd go a bit further than Glassman: the examples in the original article are misleading and nonsensical. Measuring the difference between 64kg and 65kg on a linear scale is equal to measuring the difference between a value of 2^16 and 2^16.25 on a logarithmic scale.

Of course the whole systems is only as strong as the weakest link. Increasing the number of bits of the DAC will make some other part of the system the limiting factor, if it wasn't already.

 

[jefemeister]

yes, as long as you can keep your quantization noise uncorrelated and make sure it's below the noise floor of your analog stage, all is well from a number of bits standpoint. IMO, sample rate only needs to be increased until an analog pulse can be clearly represented in the time doamin (such as a -6dB 3us spike)