The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
32-bit converters??? Is ESS Sabre better than Apogee 16x??? Audio Interfaces
Old 5th July 2016
  #61
Quote:
Originally Posted by azzzy View Post
I am confused. Don't current 24 bit converters use floating point?
By definition, you can really only use floating point math to process a signal already in the digital domain, so a converter has to be fixed point.

Maybe read some more about what floating point means, and it should be clear!
Old 5th July 2016
  #62
Lives for gear
 

Quote:
Originally Posted by chrischoir View Post
I'm not denying nor implying anything. I'm just telling him that modern ADC ICs sample in the Mhz speeds. When modern ADCs read incoming voltages they process them at super fast high frequencies, subsequently they gather loads of info (aka sample points). It's not redundant or useless information as he stated. You can read anything you want into that, I'm stating fact and I'm only questioning HIS statement and not sampling theorems. Understood?
You are mixing two very different things : the physical implementation of the converter and the resulting data.

One must see the converter as a black box, with an input and an output. The innards and how they work is irrelevant to the topic. Why oversampling is used in order to facilitate digital FIR filtering or to allow noise shaping in the upper bands and all other technical considerations are irrelevant to the topic at hand.

The point is : more samples does not mean more information or more content if the signal is bandwidth limited. This has been demonstrated a long time ago. Although it seems counterintuitive, it really is understandable if you have a precise knowledge of the mathematics behind signal processing.

Quote:
Originally Posted by chrischoir View Post
I'm not implying or stating 44Khz is not ample to convert analog audio and reproduce digital audio. I'm simply correcting him that when sampling occurs in modern devices, it's not as simple as reading in voltages at 44K and sending them out in 44k data streams. What goes on in between is pretty complex. So please don't put words in my mouth. Thank you in advance, Sir.
No correction needed as I knew that already.
Old 5th July 2016
  #63
Lives for gear
 
toledo3's Avatar
 

Quote:
Originally Posted by psycho_monkey View Post
By definition, you can really only use floating point math to process a signal already in the digital domain, so a converter has to be fixed point.

Maybe read some more about what floating point means, and it should be clear!
(bolded by me)

FWIW, you could use fixed point math to process in the digital domain.

For example, in iOS, the canonical format for processing is 8.24 bit fixed point PCM.

On the other note- it is entirely possible to implement a floating point ADC and has been done, it's just not done often, because according to most outlooks it just isn't needed.

I know you're a mod here, not my intent to put anyone on the spot or anything. Just...maybe lighten up on the chastising posters when you are in fact incorrect yourself.
Old 5th July 2016
  #64
Quote:
Originally Posted by toledo3 View Post
(bolded by me)

FWIW, you could use fixed point math to process in the digital domain.

For example, in iOS, the canonical format for processing is 8.24 bit fixed point PCM.

On the other note- it is entirely possible to implement a floating point ADC and has been done, it's just not done often, because according to most outlooks it just isn't needed.

I know you're a mod here, not my intent to put anyone on the spot or anything. Just...maybe lighten up on the chastising posters when you are in fact incorrect yourself.
You're misinterpreting my first statement - I didn't mean to suggest you couldn't use fixed point maths in digital audio processing, only that it doesn't make sense to use floating point at the conversion point.

In what application did this floating point converter exist? I can't imagine it was audio, since a theoretical floating point DA would potentially overload even the most forgiving analogue stage, and a floating point ad would be pointless since as discussed all it would capture is extra noise.

I'm happy to be proved wrong, but every 24bit converter is fixed point, as is the 32bit ones in question. I'm not lecturing the OP, just clarifying!
Old 6th July 2016
  #65
Gear Addict
Quote:
Originally Posted by psycho_monkey View Post
I'm not lecturing the OP, just clarifying!
Personally I am happy to learn regardless of how the information is presented (this whole fixed vs. floating point thing is completely new information to me, for example). But I must say it's very unpleasant when a thread - instead of being purely informative - turns into "I know better than you, now go learn, boy" type dialog. This is not specifically about the latest comment - this whole thread quickly devolved into a pissing contest. Let's chill, people, eh?
Old 6th July 2016
  #66
Lives for gear
 
Whack Doofa's Avatar
Quote:
Originally Posted by azzzy View Post
Personally I am happy to learn regardless of how the information is presented (this whole fixed vs. floating point thing is completely new information to me, for example). But I must say it's very unpleasant when a thread - instead of being purely informative - turns into "I know better than you, now go learn, boy" type dialog. This is not specifically about the latest comment - this whole thread quickly devolved into a pissing contest. Let's chill, people, eh?
I'm not sure what it is that makes audio engineers absolute "****** nozzles" when it comes to discussing sound.

I think it's something to do with the subjectivity of audio (particularly music) recording. It's very hard to be 100% correct much of the time. (Although I'm sure someone will correct me on this)

It's nowhere near as prevalent in other creative arts.

I was guilty this evening. Bristling when someone dared to try and teach me something I already knew!!

Ego, Ego everywhere, and not a drop to.... something.

EDIT to add. As I was reading the thread, I wanted to point out that you were both right about tape but coming from it from opposite angles.
Old 6th July 2016
  #67
Lives for gear
 
Whack Doofa's Avatar
Quote:
Originally Posted by IanBSC View Post
This view of bit depth, though ubiquitous, doesn't seem right to me. If I run a 24 bit signal and a 16 bit signal through a tube device with a high noise floor, or print either to tape, they do not sound the same.
For this reason I am open to the possibility that 32 bit conversion could improve sound quality, even if it exceeds the dynamic range of the converter or analogue signal. But, I want to hear it first.
Everytime I've tried I have been able to null 16 bit and 24 bit files. Even down to the silence, which I was surprised about. A couple of times I have heard tiny spits of noise on very quiet fade outs, but other than that absolute silence.
It makes sense when you consider the theory behind wave reconstruction. It also gives me confidence to roll my eyes when people declare the difference between 16 vs 32 bit as "night and day".
That veil appears again!

All this said, I'm not 100% sure that all 16bit files will null with their 24bit equivalents. Just pointing out my observation.
Old 6th July 2016
  #68
Lives for gear
 
toledo3's Avatar
 

Quote:
Originally Posted by psycho_monkey View Post
In what application did this floating point converter exist? I can't imagine it was audio, since a theoretical floating point DA would potentially overload even the most forgiving analogue stage, and a floating point ad would be pointless since as discussed all it would capture is extra noise.

I'm happy to be proved wrong, but every 24bit converter is fixed point, as is the 32bit ones in question. I'm not lecturing the OP, just clarifying!
Kinetis. There is another company too that I'm forgetting the name of. I've run across it while designing some interactive visual art.

The dynamic range of a 32 bit fixed point or even 24 bit fixed is way, way more than enough for audio.

It is also possible to use existing fixed point chips in combination to achieve a floating point output, as has been done many times and described in various white papers and tech sites, going back quite a ways.

Floating point ADC is also used in video sensors in an optical application, found in things like infrared cameras or similar, for detecting terrain or even for obtaining meshes for projection mapping.

The big thing in the statement that popped out to me, was the point that processing has to be floating point which then means that ADC/DAC must be fixed, when in reality processing can be and is often "fixed point"/integer. Especially if a programmer wishes to use integer and coefficients to get more accurate results than are sometimes possible with float, or even double float. That is more of a real world scenario.

Supposing that that it was float processing though - as it often is, it doesn't then mean that output would *have* to go to a fixed point chip. It could also be floating. It would be entirely unnecessary in my estimation, but it could, with no harm. Floating point d/a wouldn't necessarily overload a subsequent analog input stage, it just depends on the calibration....it could just as easily be setup to be able to resolve data that analog never could.

To be really clear, I am in no way saying that floating point ADC/DAC is necessary or makes sense for any of the audio that any of us does in audio. That's more hypothetical. Big part of why it hasn't worked it's way towards being common, despite people screwing around with it for decades now. All that is really needed is the fixed on input, float for DSP (or integer sometimes as I pointed out)... and then fixed on output. But that doesn't mean it *has* to happen that way, especially with the DSP side of things.

I'm not saying I'm Einstein when it comes to this stuff, I just know what I've had to run across for work... which often involves pulling off something "never before done" for various interactive installs and interactive/visualization bits for tours, and reading a ton of white papers and tech stuff from all sorts of disciplines, to try to think of some new combos. Some of the facts occasionally manage to stick in my brain!

Last edited by toledo3; 6th July 2016 at 07:36 AM.. Reason: typo
Old 6th July 2016
  #69
Lives for gear
 
IanBSC's Avatar
Quote:
Originally Posted by Whack Doofa View Post
Everytime I've tried I have been able to null 16 bit and 24 bit files. Even down to the silence, which I was surprised about. A couple of times I have heard tiny spits of noise on very quiet fade outs, but other than that absolute silence.
It makes sense when you consider the theory behind wave reconstruction. It also gives me confidence to roll my eyes when people declare the difference between 16 vs 32 bit as "night and day".
That veil appears again!

All this said, I'm not 100% sure that all 16bit files will null with their 24bit equivalents. Just pointing out my observation.
I will admit, I do not do null tests. I'm not sold on how useful it really is. However, I hear some very clear and consistent differences between 16 and 24 bit files, particularly in terms of reverb tails, separation, subtle dynamics and masking. Obviously, this is not everyone's finding.

Viewpoints on this subject usually emerge from a couple unstated perspectives: 1) Modern conversion is already good enough because of DNR and THD specs, so claimed improvements are irrelevant, 2) Modern converters still don't sound quite as good as straight wire, so I'm open to any potential improvements if they turn out to be audible, 3) My conversion sounds ok, but I'm curious about these manufacturer claims and what they mean.

I'm in the second camp, but I will trust a claim when I hear the end result.
Old 6th July 2016
  #70
Lives for gear
 

of course a 16 and 24 bit file won't null. the difference is around -70 dbfs or so. this is sufficient to appear as silence, but in reality we're talking about a whole lot of quantization noise. not desirable for the aesthetically minded.

back to the question at hand, does the 32 bit filter sound 'better'? perhaps a tiny bit, but not really something to worry about.
Old 6th July 2016
  #71
Lives for gear
 
Brian Campbell's Avatar
 

This stuff is market driven as manufacturers try to figure out ways to sell gear, "more is better" is always a good bet.
The sad part is we have more dynamic range available today than ever before yet lots of guys are not using it and insist on slamming and squashing the crap out of everything. It's ridiculous.
Anyway no pissing contest here, ranting yes, pissing no...

Last edited by Brian Campbell; 7th July 2016 at 03:53 PM..
Old 6th July 2016
  #72
Lives for gear
 
IanBSC's Avatar
Quote:
Originally Posted by Timesaver800W View Post
back to the question at hand, does the 32 bit filter sound 'better'? perhaps a tiny bit, but not really something to worry about.
I've used a number of pieces of gear that have these chips, and I would put the improvement with the kind of incremental improvement you expect every 7-10 years along with better clocks, circuit designs, etc. They're just better chips.

A 32 bit recording and playback chain might be another story.
Old 6th July 2016
  #73
Quote:
Originally Posted by IanBSC View Post
I will admit, I do not do null tests. I'm not sold on how useful it really is. However, I hear some very clear and consistent differences between 16 and 24 bit files, particularly in terms of reverb tails, separation, subtle dynamics and masking. Obviously, this is not everyone's finding.

Viewpoints on this subject usually emerge from a couple unstated perspectives: 1) Modern conversion is already good enough because of DNR and THD specs, so claimed improvements are irrelevant, 2) Modern converters still don't sound quite as good as straight wire, so I'm open to any potential improvements if they turn out to be audible, 3) My conversion sounds ok, but I'm curious about these manufacturer claims and what they mean.

I'm in the second camp, but I will trust a claim when I hear the end result.
One could argue that if you're doing carefully controlled listening tests to ascertain those differences, that actually tells you more about if a technical difference is relevant.

After all, there's plenty of situations where a measurable difference isn't audible. I've never heard of a situation where an unmeasurable difference can be repeatedly identified on a listening test.
Old 6th July 2016
  #74
Quote:
Originally Posted by toledo3 View Post
Kinetis. There is another company too that I'm forgetting the name of. I've run across it while designing some interactive visual art.

The dynamic range of a 32 bit fixed point or even 24 bit fixed is way, way more than enough for audio.

It is also possible to use existing fixed point chips in combination to achieve a floating point output, as has been done many times and described in various white papers and tech sites, going back quite a ways.

Floating point ADC is also used in video sensors in an optical application, found in things like infrared cameras or similar, for detecting terrain or even for obtaining meshes for projection mapping.

The big thing in the statement that popped out to me, was the point that processing has to be floating point which then means that ADC/DAC must be fixed, when in reality processing can be and is often "fixed point"/integer. Especially if a programmer wishes to use integer and coefficients to get more accurate results than are sometimes possible with float, or even double float. That is more of a real world scenario.

Supposing that that it was float processing though - as it often is, it doesn't then mean that output would *have* to go to a fixed point chip. It could also be floating. It would be entirely unnecessary in my estimation, but it could, with no harm. Floating point d/a wouldn't necessarily overload a subsequent analog input stage, it just depends on the calibration....it could just as easily be setup to be able to resolve data that analog never could.

To be really clear, I am in no way saying that floating point ADC/DAC is necessary or makes sense for any of the audio that any of us does in audio. That's more hypothetical. Big part of why it hasn't worked it's way towards being common, despite people screwing around with it for decades now. All that is really needed is the fixed on input, float for DSP (or integer sometimes as I pointed out)... and then fixed on output. But that doesn't mean it *has* to happen that way, especially with the DSP side of things.

I'm not saying I'm Einstein when it comes to this stuff, I just know what I've had to run across for work... which often involves pulling off something "never before done" for various interactive installs and interactive/visualization bits for tours, and reading a ton of white papers and tech stuff from all sorts of disciplines, to try to think of some new combos. Some of the facts occasionally manage to stick in my brain!
Interesting stuff. Good to learn! As you say, largely irrelevant for audio though where the end result has to be amplified and reproduced I guess.

You have misunderstood my point though re fixed/floating point processing though. Of course much internal processing is fixed.
Old 6th July 2016
  #75
Quote:
Originally Posted by azzzy View Post
Personally I am happy to learn regardless of how the information is presented (this whole fixed vs. floating point thing is completely new information to me, for example). But I must say it's very unpleasant when a thread - instead of being purely informative - turns into "I know better than you, now go learn, boy" type dialog. This is not specifically about the latest comment - this whole thread quickly devolved into a pissing contest. Let's chill, people, eh?
Apologies if that's how it came across. It was more meant as a helpful tip. It's quite clear, if you understand the difference between fixed and floating point - why a floating point ADC is unnecessary (especially when a true 24bit fixed converter isn't possible in practice).

Which is of course the crux of the argument against a 32bit fixed converter
Old 6th July 2016
  #76
Lives for gear
 
IanBSC's Avatar
Quote:
Originally Posted by psycho_monkey View Post
I've never heard of a situation where an unmeasurable difference can be repeatedly identified on a listening test.
Doesn't it come down to whether or not the correct things are being measured, or measured appropriately?
Old 6th July 2016
  #77
Lives for gear
 

Quote:
Originally Posted by IanBSC View Post
I've used a number of pieces of gear that have these chips, and I would put the improvement with the kind of incremental improvement you expect every 7-10 years along with better clocks, circuit designs, etc. They're just better chips.

A 32 bit recording and playback chain might be another story.
agreed. a 32 bit output (float) from ad, and 32bit input to da would be cool, as it would ensure proper dithering (these daw manufacturers get it wrong). 64 bits even cooler.
Old 6th July 2016
  #78
Lives for gear
 

unmeasurable and unmeasurable with software or ap analyzer are distinct concepts.
Old 6th July 2016
  #79
Gear Addict
Quote:
Originally Posted by psycho_monkey View Post
Apologies if that's how it came across.
No worries. Cheers.
Old 7th July 2016
  #80
Lives for gear
 
acreil's Avatar
 

Quote:
Originally Posted by psycho_monkey View Post
In what application did this floating point converter exist? I can't imagine it was audio, since a theoretical floating point DA would potentially overload even the most forgiving analogue stage, and a floating point ad would be pointless since as discussed all it would capture is extra noise.
It was done fairly frequently in the 1970s through early 90s, when high resolution converters were sometimes prohibitively expensive. You could say it's a way to get 16 bit dynamic range on a 12 bit budget, back when this was actually a big deal. The actual audio processing in these cases is fixed point.

I've been compiling a list of examples recently. The EMT 250, Lexicon 224 and 200, Publison DHM 89 and a number of things made by Casio, Yamaha and Technics used floating point converters. Yamaha used them most prolifically, including many custom ICs. Dynamic range is 15-22 bits with a 10-16 bit mantissa. They don't really perform that well and weren't generally used in high end applications. In these cases it's mainly a matter of dynamic range vs. cost.
Old 7th July 2016
  #81
Lives for gear
 

Quote:
Originally Posted by acreil View Post
It was done fairly frequently in the 1970s through early 90s, when high resolution converters were sometimes prohibitively expensive. You could say it's a way to get 16 bit dynamic range on a 12 bit budget, back when this was actually a big deal. The actual audio processing in these cases is fixed point.

I've been compiling a list of examples recently. The EMT 250, Lexicon 224 and 200, Publison DHM 89 and a number of things made by Casio, Yamaha and Technics used floating point converters. Yamaha used them most prolifically, including many custom ICs. Dynamic range is 15-22 bits with a 10-16 bit mantissa. They don't really perform that well and weren't generally used in high end applications. In these cases it's mainly a matter of dynamic range vs. cost.
No. All audio converters _used_ have been fixed point.

Last edited by pentagon; 7th July 2016 at 05:05 PM.. Reason: Getting too personal
Old 7th July 2016
  #82
Lives for gear
 
Clockwise's Avatar
I've been reading this thread and getting confused...

I thought A/D or D/A conversion is fixed point only,because any signal above 0dBFS is clipped. Floating point utilises extra bit depth for "virtual" headroom to avoid digital clipping and it's done in the digital domain, no?
Old 7th July 2016
  #83
Quote:
Originally Posted by IanBSC View Post
Doesn't it come down to whether or not the correct things are being measured, or measured appropriately?
Absolutely. If someone can repeatedly and definitely identify differences between 2 sources in controlled testing that appear identical when measured, something isn't being measured correctly.

I've never heard of this happening, but I'm not saying it's not possible!

Quote:
Originally Posted by Timesaver800W View Post
unmeasurable and unmeasurable with software or ap analyzer are distinct concepts.
Not really different concepts - just that one isn't in depth enough. If it's there, it HAS to be measurable - even if we haven't worked out how to do it yet. Surely?
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump