avatar

Mike Rivers

Aqua Marine

Posts: 2,513 Member Since: 13/10/2012

Lead

Apr 19 17 5:56 PM

Tags : :

I was going to ask about this in the DSD vs 192 kHz PCM discussion, but I see that's gone over the hill already.

A friend e-mailed me today asking what I thought of the new Sound Devices MixPre series that has 32-bit conveters. I looked it up, and I see that AKM makes a few for audio (they've been around for instrumentation for a while, for very high resolution at low sample rates, like 5 samples/second). The noise specs for the chip don't look a whole lot better than what we've been used to seeing for the past few years.

Have I not been paying attention? I didn't think there was anything but noise beyond what you could resolve with 21 or 22 bits. Is this just "mine is longer than yours?" Or is high resolution low level noise better (subjectively) than no lo level noise at all?



For a good time, call mikeriversaudio.wordpress.com

Quote    Reply   
avatar

dcollins

Platinum Blonde

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

#1 [url]

Apr 19 17 7:39 PM

This has been discussed on this forum in the past. 32 bits is the data frame. It’s impossible, for a number of reasons, to achieve very much more than 20 bits at audio sampling rates.

20 bits is one part-per-million error, which ain’t much. Although I hear that 24bits at 384kHz is where it’s really at.

 davecollinsmastering.com


Quote    Reply   
avatar

chrisj

Gold Finger

Posts: 972 Member Since:22/02/2011

#2 [url]

Apr 20 17 5:33 AM

People get confused because capture data and processing are very much not the same thing. Yes, if you encoded the output of a converter you'd end up not needing more than 20-22 bits at best, but then as soon as you try to do math on the stuff suddenly you need more and more or it gets crunchy and thin and soulless.

I doubt there's any benefit to what's currently out there as '32 bit audio converters' because at this stage it's clearly marketing trickery, and even if it became practically relevant (as in, the converter requires that much to represent its theoretical low level signal retrieval), in practice fully exploiting 24 bit would be about the same in terms of human experience.

Chris Johnson, airwindows.com

Quote    Reply   
avatar

Mike Rivers

Aqua Marine

Posts: 2,513 Member Since:13/10/2012

#3 [url]

Apr 20 17 5:55 AM

dcollins wrote:
This has been discussed on this forum in the past. 32 bits is the data frame. It’s impossible, for a number of reasons, to achieve very much more than 20 bits at audio sampling rates.
 


That's what I thought. Because of rounding, truncation, and fixed vs floating point arithmetic, a 32-bit "word window" is the equivalent of cutting off your calculation of pi to 1000 decimal places to six rather than two. But Sound Devices isn't a company that thrives on marketing trickery.

I wonder if there's any advantage of having a 32-bit word come out of the converter chip and go directly into the the 32-bit pipe, rather than a 24-bit word coming out of the chip and going into the same 32-bit pipe. Could there be an improvement in speed that data moves into the processing chain (like putting out a fire with a 4-inch fire hose rather than a 1-inch garden hose) that results in shaving a few micoseconds off the system latency?



For a good time, call mikeriversaudio.wordpress.com

Quote    Reply   
avatar

barry hufker

Diamond Forever

Posts: 12,120 Member Since:26/01/2011

#5 [url]

Apr 20 17 1:16 PM

Sound Devices certainly isn't a company to hype or mislead.  I've used their recorders for years and I like them very much.  They sound good and are ruggedly dependable.

Doing a quick search for 32 bit a/d converters lead me to this link.  I don't know if this is the chip is the one S/D is using but it seems to be a high quality chip of its type, so here's the link  http://www.akm.com/akm/en/product/detail/0055/

And below is a photo of the information...
image
image
 

Last Edited By: barry hufker Apr 20 17 1:19 PM. Edited 2 times.

Quote    Reply   
avatar

zmix

Aqua Marine

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

#6 [url]

Apr 20 17 1:21 PM

I remember when I was in elementary school there were a group of kids who were convinced that you could tell how fast a car was capable of going by looking at the highest displayed number on the face of the speedometer dial.

Quote    Reply   
avatar

Mike Rivers

Aqua Marine

Posts: 2,513 Member Since:13/10/2012

#8 [url]

Apr 20 17 2:09 PM

barry hufker wrote:
Doing a quick search for 32 bit a/d converters lead me to this link.  I don't know if this is the chip is the one S/D is using but it seems to be a high quality chip of its type, so here's the link  http://www.akm.com/akm/en/product/detail/0055/
 


That's the one I was looking at when I was trying to figure out if there was anything special about its noise or distortion specs.  Meh! But it might just be a very good converter chip close to the top of the heap of the many very good converter chips out there today.



For a good time, call mikeriversaudio.wordpress.com

Quote    Reply   
avatar

zmix

Aqua Marine

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

#9 [url]

Apr 20 17 2:12 PM

mikerivers wrote:
barry hufker wrote:
Doing a quick search for 32 bit a/d converters lead me to this link.  I don't know if this is the chip is the one S/D is using but it seems to be a high quality chip of its type, so here's the link  http://www.akm.com/akm/en/product/detail/0055/
 


That's the one I was looking at when I was trying to figure out if there was anything special about its noise or distortion specs.  Meh! But it might just be a very good converter chip close to the top of the heap of the many very good converter chips out there today.

Yep.

 -108dB = ~18 bits


Quote    Reply   
avatar

chrisj

Gold Finger

Posts: 972 Member Since:22/02/2011

#10 [url]

Apr 20 17 2:31 PM

Well hang on—32-bit digital filters? If that's new, that could amount to something. Effectively they're saying the decimation process is running at 32 bit. That's digital math: that's going to have some kind of effect, assuming other converters aren't already calculating at 32 bit.

And that, I couldn't tell ya :)

Chris Johnson, airwindows.com

Quote    Reply   
avatar

barry hufker

Diamond Forever

Posts: 12,120 Member Since:26/01/2011

#11 [url]

Apr 20 17 11:38 PM

For comparison, here is the data from what appears to be Texas Instrument's best 24 bit A/D converter, the PCM4220.  http://www.ti.com/product/PCM4220
I searched for an equivalent or superior product from Analog Devices and couldn't find one, tho' I looked for quite a while.  As I understand it, TI owns Burr-Brown and National Semiconductor.

image

Last Edited By: barry hufker Apr 21 17 12:02 AM. Edited 2 times.

Quote    Reply   
avatar

extrememixing

Platinum Blonde

Posts: 2,319 Member Since:02/02/2011

#12 [url]

Apr 21 17 1:05 AM

I don't know how to read the spec sheets. But I do know that if your digital recordings or your mixes don't sound good, there is a chance that it's not the converters that you are using.

Steve

Quote    Reply   
avatar

Mike Rivers

Aqua Marine

Posts: 2,513 Member Since:13/10/2012

#14 [url]

Apr 21 17 6:13 AM

chrisj wrote:
Well hang on—32-bit digital filters? If that's new, that could amount to something. Effectively they're saying the decimation process is running at 32 bit. That's digital math: that's going to have some kind of effect, assuming other converters aren't already calculating at 32 bit.
 


Did you look at the full data sheet for the AKM chip?  There are plots of the filter response that show the ringing near the cutoff frequency. Perhaps that tells you something useful if you have similar plots of conventional 24-bit converters to compare them to. I don't fully understand them because there's no key that tells what the numbers in parentheses or the blue and purple plots represent, but I can guess.

image



For a good time, call mikeriversaudio.wordpress.com

Quote    Reply   
avatar

dcollins

Platinum Blonde

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

#15 [url]

Apr 23 17 10:37 PM

That part actually looks worse than the 5394 that preceded it. Granted the new one does do 768kHz.

 davecollinsmastering.com


Quote    Reply   
avatar

rodaffleck

Silverado

Posts: 177 Member Since:26/01/2011

#17 [url]

Apr 24 17 9:35 AM

mikerivers wrote:

chrisj wrote:
Well hang on—32-bit digital filters? If that's new, that could amount to something. Effectively they're saying the decimation process is running at 32 bit. That's digital math: that's going to have some kind of effect, assuming other converters aren't already calculating at 32 bit.

 


Did you look at the full data sheet for the AKM chip?  There are plots of the filter response that show the ringing near the cutoff frequency. Perhaps that tells you something useful if you have similar plots of conventional 24-bit converters to compare them to. I don't fully understand them because there's no key that tells what the numbers in parentheses or the blue and purple plots represent, but I can guess.

image


The numbers in parentheses are listed in the tables that preceed the figures on the data sheets:

1. Passband limit
2. Passband ripple
3. Stopband
4. Stopband attenuation

I'm not sure what you mean by ringing near the cutoff. 

Dave Collins, out of curiosity, which specs in particular look worse than its predecessor, and would any of it actually make a difference in real world terms? A cursory glance suggests stopband attenuation is poorer on the new one but I'm not convinced you need anything better than 100+ dB anyway, and if it provides improved pre-ringing then maybe it's a worthy compromise. Also having a hex-speed clock for an ADC seems ridiculously unnecessary but that's just me.

Quote    Reply   
avatar

nobturner

Gold Finger

Posts: 388 Member Since:01/02/2011

#19 [url]

Apr 30 17 9:10 PM

I just got a song to mix that was recorded in pro tools at 96k/32-bit. Many of the tracks hit 0db FS constantly. At the very least, working in 32 bits should discourage that kind of meter-banging, no? I've just cranked the clip gain down on all of the tracks by a passel, so that I don't have to keep the faders too low to easily make small adjustments. But it appears to be a case of the "more is better!" mentality. I don't see much reason to stress my CPU for no audible improvement, but then, I'm just the mix engineer.

Quote    Reply   
avatar

zmix

Aqua Marine

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

#20 [url]

Apr 30 17 10:41 PM

nobturner wrote:
I just got a song to mix that was recorded in pro tools at 96k/32-bit. Many of the tracks hit 0db FS constantly. At the very least, working in 32 bits should discourage that kind of meter-banging, no? I've just cranked the clip gain down on all of the tracks by a passel, so that I don't have to keep the faders too low to easily make small adjustments. But it appears to be a case of the "more is better!" mentality. I don't see much reason to stress my CPU for no audible improvement, but then, I'm just the mix engineer.

Ahem:



http://prorecordingworkshop.lefora.com/topic/19408225/want-hear-possible-explanations-export-32-bit-files?page=1#.WQauVcm1tL8




Quote    Reply   
Add Reply

Quick Reply

bbcode help