Become a Site Supporter and Never see Ads again!

Author Topic: 32Bit Float recording - The Technical view  (Read 23712 times)

0 Members and 2 Guests are viewing this topic.

Offline shijan

  • Trade Count: (0)
  • Taperssection Newbie
  • *
  • Posts: 24
Re: 32Bit Float recording - The Technical view
« Reply #60 on: October 05, 2019, 07:57:41 AM »
Any info how 32Bit Float will work with compressors and software limiters in real life?
If your source have a lot of really loud splashes and you bring back loudest part to visible range in one click, other parts of track gain became too low. Simple normalize tool for selected regions only or manual gain adjust with curve is ok for simple situations, but for instant random loud splaches you need some kind of automatic limiter. So will those software plugins see that hidden dynamic range in 32Bit mode? Is it possible that software limiters in 32Bit mode will produce same result as analogue limiters or even better?

UPDATE: I downloaded samples from SoundDevices website and play with very simple AU limiter plugin in Cocos Reaper. All you need do is set attack/release time. Seems works great. Artifacts-free real time recovery!  :headphones:
« Last Edit: October 05, 2019, 08:28:13 AM by shijan »
2x Oktava MK-012+Rycote BBG -> MixPre -> Edirol R-09HR

Offline Ozpeter

  • Trade Count: (0)
  • Taperssection All-Star
  • ****
  • Posts: 1399
Re: 32Bit Float recording - The Technical view
« Reply #61 on: October 05, 2019, 08:48:32 AM »
When a recording on the unit is going well into the red, what happens to the monitoring?  Do you hear distortion that won't be heard after normalising, or... how does it work?  I guess it's down to where in the chain the monitor signal comes from.

Offline shijan

  • Trade Count: (0)
  • Taperssection Newbie
  • *
  • Posts: 24
Re: 32Bit Float recording - The Technical view
« Reply #62 on: October 05, 2019, 11:46:20 AM »
When a recording on the unit is going well into the red, what happens to the monitoring?  Do you hear distortion that won't be heard after normalising, or... how does it work?  I guess it's down to where in the chain the monitor signal comes from.

My guess you will hear distortion that won't be heard after normalising. MixPre limiters are disabled in 32 bit mode.
Here are some processed examples (original, normalized only, non normalized with software limiter plugin):
https://www.dropbox.com/sh/rm6c9sdgimeoy98/AACXdJ3tpd5Mi5QJYyO7gKG7a?dl=0

It is also interesting what strategy will work better for sound "highlights" recovery?
a. normalize all track and then use compressor to boost silent parts withiout clipping loudest parts.
b. don't normalize track and recover hidden loudest parts with limiter.

2x Oktava MK-012+Rycote BBG -> MixPre -> Edirol R-09HR

Offline voltronic

  • Trade Count: (40)
  • Needs to get out more...
  • *****
  • Posts: 4104
Re: 32Bit Float recording - The Technical view
« Reply #63 on: October 15, 2019, 06:28:35 PM »
DSatz, Gutbucket, and a few others raised some excellent points about tempering expectations regarding 32-bit float recording.  This is valuable information we all need to pay attention to, but as has been said before: the real test is in the implementation.

To that end, I would like to hear from other 32-bit float recorder owners if they have run into anything of concern.

So far, recordings from my Zoom F6 have been flawless.  These recordings have swung very wide dynamic ranges, and I do not hear any noisefloor modulation, artifacts, or anything else that shouldn't be there.  If there is anything nasty going on, it is buried way down in a level that I cannot hear it, nor can my spectral analysis show it.
« Last Edit: October 16, 2019, 05:39:12 AM by voltronic »
I am hitting my head against the walls, but the walls are giving way.
- Gustav Mahler

Acoustic Recording Techniques
Team Classical
Team Line Audio
Team DPA

Offline dallman

  • Site Supporter
  • Trade Count: (33)
  • Taperssection All-Star
  • *
  • Posts: 1812
  • Gender: Male
    • Clifford Morse
Re: 32Bit Float recording - The Technical view
« Reply #64 on: October 16, 2019, 01:38:14 AM »
DSatz, Gutbucket, and a few others raised some excellent points about tempering expectations regarding 32-bit float recording.  This is valuable information we all need to pay attention to, but as has been said before: the real test is in the implementation.

To that end, I would like to hear from other 32-bit float recorder owners if they have run into anything of concern.

So far, recordings from my Zoom F6 have been flawless.  The recordings I have made have swung very wide dynamic ranges, and I do not hear any noisefloor modulation, artifacts, or anything else that shouldn't be there.  If there is anything nasty going on, it is buried way down in a level that I cannot hear it, nor can my spectral analysis cannot show it.
I agree, my recordings have sounded excellent and very much like the 24bit recordings I have been making for years. I have played with running hot and I have played with recording conservatively and it does not seem to make a difference. I have seen very positive tests where things have really been pushed but I am not looking to make a point when I record, I am looking to grab as good a recording as I can, so I am not pushing into anything crazy. I have used both the F6 and the MixPre6II and I am happy with both decks performance and sound.
Support Live Music: Tape A Show Today!
Deck>possibly something here> Mics

Offline rick.lang

  • Trade Count: (0)
  • Taperssection Regular
  • **
  • Posts: 50
32Bit Float recording - The Technical view
« Reply #65 on: November 26, 2019, 01:16:57 PM »
Excellent feedback, thanks! 

Good to hear if I had kept my order with Zoom that all would end well.  But I switched on the delay because I couldn’t be sure it would be ready for my late September and October recordings where I used the MixPre-6 II successfully.

Offline jerryfreak

  • No PZ
  • Trade Count: (31)
  • Needs to get out more...
  • *
  • Posts: 6205
  • The plural of anecdote is not data
Re: 32Bit Float recording - The Technical view
« Reply #66 on: January 13, 2020, 11:51:19 AM »
got into a discussion on another board where it was asserted that 32-bit float offers similar 'precision' to 24-bit. their discussion was in regard to DSP.

heres the bulk of my post. Whycome i got what look like rounding errors if the container can represent 1500 dB of data to the sample?


----------------------------------------------------------------------------------------------------------------------------------------
im not really mathematically equipped to discuss it at that level

though it seems they are talking about precision of calculation of 32-bit float in DSP

i could kind of see this in a way. for a given mantissa value, there is a finite 'precision' those numbers can represent. and then the exponent in a way 'slides' that scale of representable numbers up and down, but it doesnt go any 'deeper' or more precise in calculations

seems like almost two different discussions. (DSP vs representing large differences in headroom)

im not super concerned about how the LSB in a DSP is handled, as much as i am about not clipping my signal by running out of data 'container capacity' for a better recording.

i did a test in soundforge. i generated a 1 sec 1khz sine at FSD and saved that file
then i used 'volume' effect in soundforge to adjust the volume by -120.00dB. saved that file
then i used volume again to adjust the volume +220.00 dB, which took it to an expected +100 dB signal on playback , saved that file
then i used volume to reduce it exactly -100.00 dB which put it at an expected 0dB peak

when i inverted this over the original 0dB generated file, it canceled it.... almost. there is some resulting random noise with peaks at -135 dB

so it can be argued that i indeed represented this waveform at ~24 bit precision at volume levels ranging over 200+ dB difference. however at the end of the day, the DSP precision of that file after multiple computations was indeed right around 24-bit.

could be multiple accumulated rounding errors at the 32-bit noise floor summing up to make some random low-level noise?

______________________________________________________________________________________________________

responses from that forum:

Right, it's about what you could call instantaneous dynamic range. 24 bit signed integers effectively have 23 bits plus sign but any integer gain adjustment could either clip or throw away bits. Each operation in 32 bit floating point essentially normalizes the result to keep 23 bits plus sign in the result. There numerical noise issues when doing a large number of re-scales or effects but these are usually minor.

and another response, emphasis mine

This interactive demo should be self-explanatory:
https://www.h-schmidt.net/FloatConverter/IEEE754.html

An implied bit when exponent bits are non-zero (1), (23) bits of mantissa and (1) sign bit.

The benefit of floating point is it has a near constant precision at different scales, that makes digital processing and gain staging much safer and consistent.
« Last Edit: January 13, 2020, 01:00:27 PM by jerryfreak »
Unable to post or PM due to arbitrary censorship of people the mod doesn't like. Please email me using the link in my profile if you need to connect

Offline DSatz

  • Site Supporter
  • Trade Count: (35)
  • Needs to get out more...
  • *
  • Posts: 3347
  • Gender: Male
Re: 32Bit Float recording - The Technical view
« Reply #67 on: January 16, 2020, 12:54:18 AM »
jerryfreak, nice experiment. Thank you for it.

I don't agree with the last part of the final statement (in boldface); it even seems propagandistic to me--an attempt to make the shifting digital noise floor seem like a virtue when it isn't one. I don't mean that it's necessarily a defect, either, because if it's low enough at all times, no one will hear it shifting. But "it successfully evades detection" is the best that can be said about it if so.

--best regards
« Last Edit: January 16, 2020, 08:40:14 PM by DSatz »
music > microphones > a recorder of some sort

Offline voltronic

  • Trade Count: (40)
  • Needs to get out more...
  • *****
  • Posts: 4104
Re: 32Bit Float recording - The Technical view
« Reply #68 on: March 22, 2020, 02:01:04 PM »
DSatz, Gutbucket, and a few others raised some excellent points about tempering expectations regarding 32-bit float recording.  This is valuable information we all need to pay attention to, but as has been said before: the real test is in the implementation.

To that end, I would like to hear from other 32-bit float recorder owners if they have run into anything of concern.

So far, recordings from my Zoom F6 have been flawless.  These recordings have swung very wide dynamic ranges, and I do not hear any noisefloor modulation, artifacts, or anything else that shouldn't be there.  If there is anything nasty going on, it is buried way down in a level that I cannot hear it, nor can my spectral analysis show it.

Bumping this thread for anyone considering a 32-bit float dual-ADC recorder...

6 months in, and I can report zero glitches to date with my Zoom F6 running in this mode.  While I can't speak for the specifics of the exact implementation on other units, the input gain on the F6 is fixed in this mode.  I tend to turn up the faders a bit just so I can see levels popping up during setup, +20 to +40 depending on which mics I'm using.  Keep in mind that this is only affecting post-ADC fader level as the gain is fixed.  This means that the resulting file may have areas significantly beyond 0 dBFS, but in this format you can just pull it down and post; there really isn't any clipping.

For me, I am never going back to 24-point fixed unless another 32-bit float dual-ADC device I am using down the road doesn't work as well as this one does.  I have posted on some other threads that I am often both performer and recordist for my concerts, so I cannot monitor my recordings which have wide dynamic ranges, and peak levels can be unpredictable.  Because of this, this new implementation truly is a game-changer for me, because I no longer have to rely on safety tracks and can use all of my inputs without have to think about levels.

Yes, one still needs to be careful not to overload the input stage, as has been discussed.  I don't own very sensitive mics so it's a non-issue for me.  If it is a concern for you, the Zoom F6 has a Line + Phantom mode which has a much higher max input level, or you can go with the Sound Devices MixPre-II series which has a much higher max level in all recording formats.

I am eagerly awaiting to see how the Tentacle Track E recorders perform once released.
I am hitting my head against the walls, but the walls are giving way.
- Gustav Mahler

Acoustic Recording Techniques
Team Classical
Team Line Audio
Team DPA

Offline jerryfreak

  • No PZ
  • Trade Count: (31)
  • Needs to get out more...
  • *
  • Posts: 6205
  • The plural of anecdote is not data
Re: 32Bit Float recording - The Technical view
« Reply #69 on: May 21, 2020, 01:18:22 AM »
ive come across some good recent posts from other forums with respect to bit-depth and fixed vs float, that i thought i would share here

It appears the most state of the art gear uses 32-bit fixed, not float

here is an example of a modern dual-ranging ADC, the Merging ADA8S, with some astonishing specs, using its 'dual gain' topology. it literally is utilizing close to 24bits of resolution through its entire signal path which would put it quite a bit above the specs of the 32-bit float recorders we are discussing here

https://youtu.be/IMkUcWr2-cg?t=629

heres a direct demo of a sine wave recorded peaking at -110dBU and boosted 80dB in post. the new ADA8S is the one on the left and is pretty astonishing:

https://www.merging.com/products/interfaces


some comments ive culled on fixed vs float re:this gear and its implementation

audiosciencereview:

Basically the floating point approach is just an array of ADCs (or a multichannel ADC) recording at different levels. The old Sony PCM-D100 used a similar approach -- recording two copies at the same time, but with a difference of 12dB. Of course, it only supports 24-bit integer and DSD.

While 32-bit float "only" has 25 bits of precision, don't forget it wins 32-bit integer when sample values (yes, sample values, not the overall level of the whole file) are below -42.1442dBFS (7 bits below full scale) and above 0dBFS (integers will clip). Floating point on the other hand always store 25 bits of data within the ~1500dB range. Since no combination of ADC, mic and preamp can achieve 25 bits of instantaneous dynamic range, the floating point approach is much more sensible.


from Gearslutz, someone asked Merging tech support the following question

Quote
Will the new ADA8 also be 32-Bit like Anubis? And speaking of 32-Bit conversion, Anubis specs doesn't say 32-Bit float, so it's fixed correct? And if that's the case, what made you choose 32-Bit fixed instead of float, I'm not saying one is better than the other, I'm just curious as to why you made the decision to use 32-Bit Fixed, and not Float?

response:
In terms of raw resolution, 32 bit fixed point is much better than 32 bit floating point (which only gives the equivalent of 25 bit resolution).
In other words, if converting a 32 bit fixed point signal to a 32 bit float signal and then back again to 32 bit fixed point, one would lose the 7 least significant bits from the original 32 fixed point number.

In terms of Dynamic range however (headroom, noise floor), 32 bit float is of course incomparably better (about 1500 dB) than fixed point (about 186 dB) but in most real situations 1500 dB is uselessly large.

And with respect to using a full 32 bit fixed point path in the Anubis it does make most sense, since the AD chip is able to deliver up to 32 bit fixed data and the DA chip is able to eat up to 32 bit fixed data as well.
So making sure to preserve all those 32 bit during the whole signal processing chain including through the mixing engine offers maximum transparency in the Anubis (and soon in the new ADA8 as well).


at the end of the day, we are stuffing 12-16bits of information from our microphones into a box. at the end of the day it doesnt practically matter if the box you put it in has 24 bits, 25 bits, or 32 bits of space. All that matters is that you get it in the box  ;D

as mentioned above when your sample data is below -42dB, 32bit-float offers theoretical advantage. Youd be hard pressed to hear any quantization noise over the noise floor of the room, the mics, and the analog input of your recorder, however.

-------------

here is an example of a modern ADC chip, the 8-channel AK5578
you can see how multi-channels of a chip are typically combined to boost S/N
S/N: 121 dB (8-to-4 mode: 124 dB, 8-to-2 mode: 127 dB, 8-to-1 mode: 130 dB)
https://www.akm.com/content/dam/documents/products/audio/audio-adc/ak5578en/ak5578en-en-datasheet.pdf



 
« Last Edit: May 21, 2020, 03:13:17 AM by jerryfreak »
Unable to post or PM due to arbitrary censorship of people the mod doesn't like. Please email me using the link in my profile if you need to connect

Offline justink

  • Trade Count: (4)
  • Taperssection All-Star
  • ****
  • Posts: 1973
  • Gender: Male
Re: 32Bit Float recording - The Technical view
« Reply #70 on: May 21, 2020, 07:12:13 AM »
ive come across some good recent posts from other forums with respect to bit-depth and fixed vs float, that i thought i would share here

It appears the most state of the art gear uses 32-bit fixed, not float

here is an example of a modern dual-ranging ADC, the Merging ADA8S, with some astonishing specs, using its 'dual gain' topology. it literally is utilizing close to 24bits of resolution through its entire signal path which would put it quite a bit above the specs of the 32-bit float recorders we are discussing here

https://youtu.be/IMkUcWr2-cg?t=629

heres a direct demo of a sine wave recorded peaking at -110dBU and boosted 80dB in post. the new ADA8S is the one on the left and is pretty astonishing:

https://www.merging.com/products/interfaces


some comments ive culled on fixed vs float re:this gear and its implementation

audiosciencereview:

Basically the floating point approach is just an array of ADCs (or a multichannel ADC) recording at different levels. The old Sony PCM-D100 used a similar approach -- recording two copies at the same time, but with a difference of 12dB. Of course, it only supports 24-bit integer and DSD.

While 32-bit float "only" has 25 bits of precision, don't forget it wins 32-bit integer when sample values (yes, sample values, not the overall level of the whole file) are below -42.1442dBFS (7 bits below full scale) and above 0dBFS (integers will clip). Floating point on the other hand always store 25 bits of data within the ~1500dB range. Since no combination of ADC, mic and preamp can achieve 25 bits of instantaneous dynamic range, the floating point approach is much more sensible.


from Gearslutz, someone asked Merging tech support the following question

Quote
Will the new ADA8 also be 32-Bit like Anubis? And speaking of 32-Bit conversion, Anubis specs doesn't say 32-Bit float, so it's fixed correct? And if that's the case, what made you choose 32-Bit fixed instead of float, I'm not saying one is better than the other, I'm just curious as to why you made the decision to use 32-Bit Fixed, and not Float?

response:
In terms of raw resolution, 32 bit fixed point is much better than 32 bit floating point (which only gives the equivalent of 25 bit resolution).
In other words, if converting a 32 bit fixed point signal to a 32 bit float signal and then back again to 32 bit fixed point, one would lose the 7 least significant bits from the original 32 fixed point number.

In terms of Dynamic range however (headroom, noise floor), 32 bit float is of course incomparably better (about 1500 dB) than fixed point (about 186 dB) but in most real situations 1500 dB is uselessly large.

And with respect to using a full 32 bit fixed point path in the Anubis it does make most sense, since the AD chip is able to deliver up to 32 bit fixed data and the DA chip is able to eat up to 32 bit fixed data as well.
So making sure to preserve all those 32 bit during the whole signal processing chain including through the mixing engine offers maximum transparency in the Anubis (and soon in the new ADA8 as well).


at the end of the day, we are stuffing 12-16bits of information from our microphones into a box. at the end of the day it doesnt practically matter if the box you put it in has 24 bits, 25 bits, or 32 bits of space. All that matters is that you get it in the box  ;D

as mentioned above when your sample data is below -42dB, 32bit-float offers theoretical advantage. Youd be hard pressed to hear any quantization noise over the noise floor of the room, the mics, and the analog input of your recorder, however.

-------------

here is an example of a modern ADC chip, the 8-channel AK5578
you can see how multi-channels of a chip are typically combined to boost S/N
S/N: 121 dB (8-to-4 mode: 124 dB, 8-to-2 mode: 127 dB, 8-to-1 mode: 130 dB)
https://www.akm.com/content/dam/documents/products/audio/audio-adc/ak5578en/ak5578en-en-datasheet.pdf

The way I interpret the two bolded statements is that if I'm exporting from my DAW in 24bit, it seems that 32bit float is the best of everything and there might not be a need to ever upgrade a recorder after this.
Mics:
DPA 4023 (Cardioid)
DPA 4028 (Subcardioid)
DPA 4018V (Supercardioid)
Earthworks TC25 (Omni) 

Pres and A/D's:
Grace Design Lunatec V3 (Oade ACM)
Edirol UA-5 (bm2p+ Mod)

Recorders:
Sound Devices MixPre10 II
Edirol R-44 (Oade CM)
Sony PCM‑M10

Offline jerryfreak

  • No PZ
  • Trade Count: (31)
  • Needs to get out more...
  • *
  • Posts: 6205
  • The plural of anecdote is not data
Re: 32Bit Float recording - The Technical view
« Reply #71 on: May 21, 2020, 07:53:22 AM »
well, it depends on what youre happy with. technically im sure somebody said the same thing when good HQ pre/AD combos like the V3 came out. As easily as digital technology can become dated, 20 years later you can still go make world-class recordings with a V3, in a studio, yet alone a relatively noisy live show environment.

Heck a sax>SBM setup from the 90s was putting down phenomenal dynamic range for our purposes esp with the noise-shaping of the SBM

that said, perhaps you could have highlighted this part as well:
Since no combination of ADC, mic and preamp can achieve 25 bits of instantaneous dynamic range

it seems that the ADC portion is ahead of the rest of the signal chain, for now (forever?). However, unlike previous improvements in digital technology, the 32-bit float architecture only improves the sound when compared with a less-than-ideal setup of lesser bitrate. In other words there is no sonic advantage to a 32-bit file vs a 24-bit file at proper levels. Which depending on setup (background noise, mic noise, amount of gain used, etc) could mean anything where your peaks dont hit -12dBu to where your peaks dont hit -48dBu. Again, its a container for your data, if you are carefully putting your data in the container there will be no difference. if you are missing the container with your data its a huge problem. and an easy fix in this case, with the auto-ranging ADCs which "move the container for you" as needed.

will you want to upgrade someday? sure... when a recorder that sounds as good as your mixpre is the size of your phone. that will be another 20+ years though, and will likely only offer diminishing returns, if any, over your current setup in terms of overall sonic quality for our purposes

next stop is microphone signal path, it seems that paused for a bit with the digital technology, but i would imagine thats what we see next. Every manufacturer has their own way of powering their mics, from the permanently polarized DPAs that take a trivial amount of voltage/current, to the schoeps who have cut the power requirements of their preamps by nearly an order of magnitude in a few decades

couple these advances with digital microphone preamp technology (advantage being less power required to generate a hot signal to compete with noise), as well as the ever evolving ADC and op-amps, and, basically its only going to get better as far as tech.

so i guess to circle back, you asked  "there might not be a need to ever upgrade a recorder after this."

was there a need before this?  :bigsmile:
« Last Edit: May 21, 2020, 08:02:38 AM by jerryfreak »
Unable to post or PM due to arbitrary censorship of people the mod doesn't like. Please email me using the link in my profile if you need to connect

Offline Gutbucket

  • record > listen > revise technique
  • Trade Count: (15)
  • Needs to get out more...
  • *****
  • Posts: 15700
  • Gender: Male
  • "Better to love music than respect it" ~Stravinsky
Re: 32Bit Float recording - The Technical view
« Reply #72 on: May 21, 2020, 10:07:14 AM »
^ Good summary!

Quote
auto-ranging ADCs "move the container for you" as needed
musical volition > vibrations > voltages > numeric values > voltages > vibrations> virtual teleportation time-machine experience
Better recording made easy - >>Improved PAS table<< | Made excellent- >>click here to download the Oddball Microphone Technique illustrated PDF booklet<< (note: This is a 1st draft, now several years old and in need of revision!  Stay tuned)

Offline voltronic

  • Trade Count: (40)
  • Needs to get out more...
  • *****
  • Posts: 4104
Re: 32Bit Float recording - The Technical view
« Reply #73 on: May 21, 2020, 12:00:25 PM »
Regarding the mic signal path:

Remember AES42 digital mics, like the Neumann D line?  For those that don't know, these mics have a 28-bit ADC right in the mic body.  The specs on them are amazing, and the few recordings I have heard made with them are extremely transparent sounding.  It doesn't seem that standard ever got wide adoption though - I hear Neumann is discontinuing that series.  Sound Devices always supported AES42 back through the 7-series. 

The silver bullet for audio recording in my mind would be AES42 mics with the type of 32-bit multi-ADC converters we are discussing built right in.  Now you have the mic manufacturers doing all of the "container moving" onboard, and you can simplify your recording box to a bit bucket that only has to supply 10V phantom to the mics.  Lower noise all around, less chance of interference on your cable runs, lower power requirements, etc.
I am hitting my head against the walls, but the walls are giving way.
- Gustav Mahler

Acoustic Recording Techniques
Team Classical
Team Line Audio
Team DPA

Offline jerryfreak

  • No PZ
  • Trade Count: (31)
  • Needs to get out more...
  • *
  • Posts: 6205
  • The plural of anecdote is not data
Re: 32Bit Float recording - The Technical view
« Reply #74 on: May 21, 2020, 12:05:42 PM »
yes schoeps digital version of CMCs (CMD i beleive?) just disappeared as well about 7 years ago

i’m sure they could do it better with today’s tech. those came out in like 07-08 IIRC. that’s forever ago in digital

Unable to post or PM due to arbitrary censorship of people the mod doesn't like. Please email me using the link in my profile if you need to connect

 

RSS | Mobile
Page created in 0.102 seconds with 44 queries.
© 2002-2024 Taperssection.com
Powered by SMF