avatar

soapfoot

Ruby Baby

Posts: 7,316 Member Since:04/02/2011

#201 [url]

Dec 24 16 3:26 PM

OK I must pause for a question here--

Has the SOTA advanced such that converters exist that can capture or play back greater than 24 bits performance? I was always under the impression that the practical limit was around 21 bit performance. Has this changed?

brad allen williams

Quote    Reply   
avatar

gtoledo3

Aqua Marine

Posts: 4,115 Member Since:23/10/2013

#202 [url]

Dec 24 16 5:32 PM

I think you're right that the practical limit winds up being around 21 bit, Brad.

I know there are 32 bit ADC chips made by multiple companies (TI, Analog Devices, a couple boutique manufacturers).... not sure what end companies incorporate them. I'm certain there is at least one end manufacturer that did, because I've looked into it before and found one (or two I think?). The names don't come to mind immediately.

On the other side, Cranesong's Avocet IIA has a 32 bit capable DAC, supposedly... and I tend to believe that if Dave Hill is putting his name on it, that it can do it. Haven't used it personally.

Quote    Reply   
avatar

gtoledo3

Aqua Marine

Posts: 4,115 Member Since:23/10/2013

#203 [url]

Dec 24 16 5:43 PM

I strongly suspect that the only reason PT can print 32bit float files was for bug testing, it got left in, and here we are.

I have no proof, but that's my guess and I'm sticking to it... 

:)

Quote    Reply   
avatar

dcollins

Platinum Blonde

Posts: 2,343 Member Since:27/01/2011

#204 [url]

Dec 24 16 7:40 PM

gtoledo3 wrote:
I think you're right that the practical limit winds up being around 21 bit, Brad.

I know there are 32 bit ADC chips made by multiple companies (TI, Analog Devices, a couple boutique manufacturers).... not sure what end companies incorporate them. I'm certain there is at least one end manufacturer that did, because I've looked into it before and found one (or two I think?). The names don't come to mind immediately.

On the other side, Cranesong's Avocet IIA has a 32 bit capable DAC, supposedly... and I tend to believe that if Dave Hill is putting his name on it, that it can do it. Haven't used it personally.
32 bits is just the data frame size.  It is not related to the actual conversion accuracy which ranges up to somewhat more than 20 actual bits.  It's a physical impossibility to get much more for many reasons......  But if you're selling DAC's you can bet they are advertised as "32 Bits."



Quote    Reply   
avatar

tom eaton

Platinum Blonde

Posts: 1,335 Member Since:25/01/2011

#206 [url]

Dec 24 16 9:47 PM

The newest Benchmark DACs have 32 bit converters as well, and they implement volume control in 32 bit and hand that off to the converter...which delivers about 21 bits of s/n. I never liked the sound of their dac1, but they seem to be serious about trying to lower the noise on the playback side. Very curious to hear their amp which claims 130dB unweighted s/n!

t

Quote    Reply   
avatar

ArtSta

Silverado

Posts: 66 Member Since:30/04/2014

#207 [url]

Dec 25 16 1:30 PM

Late to the party...

dandan wrote:
So how does Pro deal with getting a gain changed track back down again to 16 Bit for CD? 



 

Well, I am pretty sure pros have all the source files at hand when they need it :), as simple as it sounds :).

 

zmix wrote:
and asserts that 24 bit audio really doesn't require any dither, (which most of us know already).



 

I don’t agree. Truncation to 24 bit also requires dither. Here it is why: https://www.gearslutz.com/board/11361110-post69.html
zmix wrote:
Certainly noise can "self dither" and yes pure white noise can also dither.

When we're talking about TPDF, yes, it works "better" in the sense that it can accomplish the same function as random white noise, but at a largely inaudible level.



 

I don’t agree here as well. I have prepared some files to show why.
As DC wrote, just noise can't decorrelate properly or enough (but can mask some distortion for sure).

white noise added before truncation: https://drive.google.com/file/d/0ByU7lL8D-5TablJmd0NERkFKRFU/view?usp=sharing
white noise added before truncation (slightly lower noise level): https://drive.google.com/file/d/0ByU7lL8D-5Tac05EX00zVGdjUFE/view?usp=sharing
tpdf dither for comparison: https://drive.google.com/file/d/0ByU7lL8D-5TaN0Z5a3FLZjdNMUE/view?usp=sharing

Btw. GS examples use Mbit+. 

Happy Holidays and 2017!
Art

Quote    Reply   
avatar

dcollins

Platinum Blonde

Posts: 2,343 Member Since:27/01/2011

#209 [url]

Dec 25 16 7:16 PM

bob olhsson wrote:
AES/EBU truncates to 24 bits so the 32 bit converters are probably only 32 bit using their USB input.
32 bits is very common in the I2S interface.  Also, the DAC chip itself may be using 32 bit processing for the filter, volume control, ASRC, etc.

The new ESS chip is squeezing another couple dB of THD+N out, but I fear the push to ever-lower supply voltages and dissipation requirements (most new technology is driven by mobile phones, not audio DAC's, after all) will make these converters the ones to beat for the foreseeable future.



Quote    Reply   
avatar

zmix

Aqua Marine

Posts: 4,066 Member Since:20/01/2011

#210 [url]

Dec 26 16 10:08 AM

dcollins wrote:
bob olhsson wrote:
AES/EBU truncates to 24 bits so the 32 bit converters are probably only 32 bit using their USB input.
32 bits is very common in the I2S interface.  Also, the DAC chip itself may be using 32 bit processing for the filter, volume control, ASRC, etc.

The new ESS chip is squeezing another couple dB of THD+N out, but I fear the push to ever-lower supply voltages and dissipation requirements (most new technology is driven by mobile phones, not audio DAC's, after all) will make these converters the ones to beat for the foreseeable future.




Yes, that's exactly right, ESS claim  of their flagship ES9018:

"The SABRE 32 Reference audio DAC’s 32-bit Hyperstream architecture can handle full 32-bit PCM data via I2S input..."
"The SABRE 32 consumes less than 100mW"
And they also list "64-bit accumulator and 32-bit processing" in the datasheet

The THD is listed as -120dB in Mono, Stereo or 8 channel mode (which is 19.93 bits) and in Stereo mode they claim a dynamic range of 133dB (22 bits!!)

As we know, dither makes the resolution infinite, and as Paul Frindle noted in the Oxford manual, properly dithered signals can be recovered 120dB below the noise floor.
We can see in the ESS ES9018 DAC, which is the current state of the art converter technology, and designated as a "32-bit" DAC,  that it is producing THD levels 24dB above even the theoretical quantization error of 24 bit PCM. 

Mark my words, this is a truly remarkable feat.

But is it 32 bits?  No.

It will certainly never reproduce truncation distortion in a 24 bit file, and probably not in a 20 bit file.

Viva la Resolution..!!


Quote    Reply   
avatar

chrisj

Gold Finger

Posts: 972 Member Since:22/02/2011

#214 [url]

Mar 1 17 11:23 AM

That's me… Bob O. didn't like it as much as the simple highpass TPDF 'PaulDither' so I made this one.

http://www.airwindows.com/not-just-another-dither/

Comes in 16 bit (NotJustAnotherCD) and 24 bit (NotJustAnotherDither) versions. The video ought to give you some idea of how it works: pretty sure it's the highest performance dither out there.

[url=

Uses no noise at all. It's all Benford realness calculations and noise shaping. And it gates to digital black if given digital black (which has its uses).

Again, listen to the video, try it (it's free, I am patreon supported). If you can do better, I'll use that. NJAD was a hell of a breakthrough. And yes it still 'naturalizes'.

Chris Johnson, airwindows.com

Quote    Reply   
avatar

Hermetech Mastering

Gold Finger

Posts: 385 Member Since:22/02/2011

#215 [url]

Mar 1 17 5:53 PM

Great plugins and videos Chris, I'm going to try out NJAD instead of my usual TPDF dither. A couple of questions though, I am not so clear on a few points, and perhaps you may be able to help me understand things better:

1) In my usual mastering work flow, I put a 24 bit TPDF dither plugin after the client's source file and a digital EQ (which expands anything to 32 bit float), and right before it goes out the transfer DAC (24 fixed) to the analogue chain. Aside from the (very good) arguments stating you may not need to dither when going from 32 bit float to 24 bit fixed point, it seems to sound better with the dither. I've previously only ever used TPDF in this situation, as I heard/read that any noise shaping benefits will be negated any time anything else is done to the file further down line. I later capture the file again after the analogue chain and process it with a digital limiter etc. The question is, does this apply to this situation, i.e. just before going out the DAC to the analogue chain? In short, would you recommend I use NJAD in this situation, or stick to "basic" TPDF?

2) A similar question, the other place I use dither is right at the end of mastering, after the final digital limiting, to arrive at the 24 fixed source file I send to the client. In this situation I have always just used TPDF, because I know somewhere down line either the client, label or listener are going to make mp3s, or use a digital volume control to alter the sound in some way. Again, I have heard that anything other than TPDF (e.g. noise shaped dither) will somehow "negate" itself, and the only "safe" dither to use is TPDF. Is this really the case? And as such, would you recommend NJAD knowing that things like digital volume adjustments are likely to be made to the file?

Hope I'm making myself clear, and thanks in advance for any further info!

Gregg Janman, Hermetech Mastering

Quote    Reply   
avatar

chrisj

Gold Finger

Posts: 972 Member Since:22/02/2011

#216 [url]

Mar 1 17 6:57 PM

Yeah, I still would. The reason being, if you compare NJAD to high energy noise shapers you'll see it just produces less energy. You're quite right that you can't use typical noise shapers (numerous others, and several of my own) in that context, because they're fragile and depend on the noise shaping not being touched (in particular, you're throwing a lot of energy that can't be clipped or limited, and messing with it will cause problems).

NJAD's a different beast. It has NO noise source in it at all: it's doing another sort of calculation, and then its noise shaping is not a high order, more a sort of bias into the already-unusual calculation (what it does is work out a Benford Realness calculation and round up or down based on what will be more statistically typical).

I'm pretty confident you can go out to analog using NJAD and process there: that's like an ideal case and the only concern is 'will it gate in a way I don't like' which nearly any background noise in the mix would prevent. As for the final digital limiting to 24 bit output, I'm likewise pretty confident there because this is not a fragile arrangement: the basic processing is trying to create a distribution of values that's as statistically random as any TPDF would produce.

The concern with heavy noise shaping is very much that the output will require huge overshoots, values way beyond 'truncated to ceiling or floor' in order to generate the desired output, and screwing with that will cause ugly artifacts. When I meter it in testing (or in the videos) the amplitude including peak energy is less than TPDF: if you used a high energy noise shaper, the peak energy would get way higher and that's your tip-off that there could be trouble downstream. I think you're safe though I invite observations on whether that's as safe as I say it is :)

Chris Johnson, airwindows.com

Quote    Reply   
avatar

Hermetech Mastering

Gold Finger

Posts: 385 Member Since:22/02/2011

#217 [url]

Mar 2 17 3:40 AM

Thanks for that very enlightening answer, Chris. That certainly puts my mind at ease and encourages me to try out NJAD properly.

I tried NJAD in my media player last night (I have a "Headphone Processing" chain in Foobar2000, which is a Gain drop, which reminds me, I ought to be using your bit gain shift plugin here instead... plus an EQ and a Crossfeed, with 24 bit Dither at the end), and it seemed to make things sound better. Not done any blind A/Bs yet though.

Today I will try swapping it out with my regular dithers (TBDither set to TPDF before the analogue chain, and the built in Dither in the Limitless limiter, right before the final file render, again plain old TPDF), in the mastering chain, and report back.

Gregg Janman, Hermetech Mastering

Last Edited By: Hermetech Mastering Mar 2 17 3:43 AM. Edited 1 time.

Quote    Reply   
avatar

chrisj

Gold Finger

Posts: 972 Member Since:22/02/2011

#219 [url]

Mar 2 17 2:08 PM

That's why my original guidelines for coming up with the name included 'it shouldn't use the word dither'. Naturalize did end up using a noise source. NJAD uses no noise source at all, it's completely procedural and will give the same results if you give it the same input stream of samples. Does it 'work'? I guess you gotta be the judge of that. It sure ACTS like it works :)

Chris Johnson, airwindows.com

Quote    Reply   
Add Reply

Quick Reply

bbcode help