Jump to content

Floating Point v Fixed Point WAV Files


jon_tatooles

Recommended Posts

I believe we will start to see this format, which has been available in DAWS for some time, becoming more common in sound-for-picture applications. The benefit of the extended representation of dynamic range outweighs the penalty of the added data required.

 

Here is a write-up of what a 32-bit float file is.

 

https://www.sounddevices.com/32-bit-float-files-explained/

Link to comment
Share on other sites

2 hours ago, Michael Wynne said:

 It  does matter where you set your trim on your mic pres even when working in 32 bit because all the same signal to noise rules apply in the analog domain and you are always limited by the dynamic range of the mic and the analog stage of the mic pre.

 

Yes, lots of relevance here. The file container size has historically not been the limiting factor of dynamic range. 24-bits of resolution is excellent for the overwhelming majority of applications. And the outputs of a device, such as AES3, will be in 24-bit. And if you are sourcing from an AES3 signal, no benefit in 32-bit float, unless you inadvertently add digital gain and go over 0 dBFS. In that case the 32-bit file will maintain signal integrity whereas the 24-bit file will be unrecoverable.

 

The step in front of the file container is A-to-D conversion, which has historically been more limiting than the container. And in front of that, an analog gain stage is absolutely critical to high dynamic range. Our latest analog preamp topology's noise performance is extremely consistent across its gain range. There is a reason preamps (including ours) are specified at their maximum gain, since that is where most circuits perform best. A benefit to the work Matt Anderson has done over the years is that the noise performance of SD preamps is largely consistent across the gain range. That excellent noise performance, along with with tremendous A-to-D dynamic resolution, results in scaling of gain to the file being much less important, and nearly meaningless when you can write the full signal into a 32-bit container. 

 

If you think of it like digital camera sensors, the highest performance of those are essentially ISO-invariant. What we have with our latest is the equivalent of this in the audio domain.

 

And we will continue to support 24-bit recording, still the gold standard, proven workflow. 

 

Link to comment
Share on other sites

Thanks Jon, I just think it’s important to point out that your analog input signal level is not directly equivalent to your signal to noise level on any given mic preamp.  For instance if you were to record at a relatively low level into a analog mic pre that had 32 bit floating point converter stage, the signal to noise level of that that track when you bring it up in post would be much less ideal then say a properly gain staged analog input that maximizes the use of the analog pre amp to its fullest. I won’t deny that 32 bit has some excellent advantages like you mentioned but I think it’s important we send the clear message to younger up and coming mixers and recordists that proper gain staging on the front end mic pre does matter, no matter what the bit depth your recording in.

Link to comment
Share on other sites

 

5 minutes ago, Michael Wynne said:

I think it’s important we send the clear message to younger up and coming mixers and recordists that proper gain staging on the front end mic pre does matter, no matter what the bit depth your recording in.

 

Yes. Especially when a mixer or recorder is in a signal chain with other devices. 

Link to comment
Share on other sites

4 hours ago, Michael Wynne said:

proper gain staging on the front end mic pre does matter, no matter what the bit depth your recording in.

 

That‘s not quite true. I‘m not familiar yet with the specs of the SD 32-bit converters, but I‘ve mentioned here a few times before the StageTec TrueMatch converter. It’s a 28-bit converter and it achieves a dynamic range of (IIRC) 155dB. That’s more than any mic can deliver and its own noisefloor is lower than that of the mic. So the TrueMatch doesn’t actually have a preamp anymore. You just connect the mic and it will convert whatever comes out of your mic, whether it’s the mic‘s noise floor or the loudest sound the mic can convert, the TrueMatch just converts all of it truthfully. So, no more need for a preamp and thus for gain staging. 

Link to comment
Share on other sites

1 hour ago, Constantin said:

You just connect the mic and it will convert whatever comes out of your mic, whether it’s the mic‘s noise floor or the loudest sound the mic can convert, the TrueMatch just converts all of it truthfully. So, no more need for a preamp and thus for gain staging. 

If there is no mic pre in the signal chain and it truly goes directly mic level into a AD converter that can handle these low input levels then you are technically correct. But only in theory not practical application in production sound workflows.  No matter how extensive the 32 bit float point dynamic range is in the digital domain the fundamentals of proper gain staging and signal to noise still apply for all production sound systems on all analog input/output points in a 32 bit float workflow for those who wish to achieve outstanding audio.  What concerns me is this marketing hype of “input trims no longer matter”  It’s simply not true and it sends the wrong message for those who may not no better.  

Link to comment
Share on other sites

It’s not just in theory, I mean the TrueMatch actually exists and could be used on location, although it is a bit unwieldy. 

Nevertheless, I think that if you’re coming out of a wireless receiver and go into the recorder theoretically no gain would be required at tue recorder‘s input stage. If you do apply gain, like we probably all do, it is to move the signal away from the noise that is to be added after the preamp, like converter noisefloor and so on.  But applying this gain will bring up any noise existing in the signal already, as it came out of the receiver. And the separation of signal to noise, in terms of dB, will remain the same. 

This gain will also increase any inherent preamp noise. So, when the converter stage doesn’t add any noise and can’t be clipped, gaining at the recorder becomes a lot less important. And so, Aaton, Zaxcom, Sonosax, and now Sound Devices all say that gaining is not really required anymore. 

Link to comment
Share on other sites

24 minutes ago, Constantin said:

And so, Aaton, Zaxcom, Sonosax, and now Sound Devices all say that gaining is not really required anymore. 

I've never heard any of these manufactures say that that gaining is no longer required on a mic pre amp.   Maybe in the marketing literature as a way to sell a certain type of converter or recording system as far as dynamic range.  But that's only one small piece of a much larger picture.   Even if one thinks that they could hypothetically get away without proper pre amp gain based on this ridiculous application theory that input trim levels no longer matter, any under or over modulated input signal would therefore create more issues as it runs through the signal chain of the full system.  Maybe as an experiment try ignoring your input trim levels on a professional job and see how long you hold on to that job before post gets a hold of you.

Link to comment
Share on other sites

I've found the first test of curtis Judd quite impressive where he demos recordings where he puts the gain on the Zoom F6 all the way up and all the way down and all recordings where quite usable. Not that I'm saying that gain staging doesn't matter at all, but the results where really surprising to me. 

 

 

 

Greetings

Link to comment
Share on other sites

2 hours ago, Michael Wynne said:

Maybe as an experiment try ignoring your input trim levels on a professional job and see how long you hold on to that job before post gets a hold of you.

 

Well, I don’t have a 32-bit recorder, so I can’t try that. 

Here is a quote from the text Jon Tatooles linked to:

 from a fidelity standpoint, it doesn’t matter where gains are set while recording. Audio levels in the 32-bit float WAV file can be adjusted up or down after recording with most major DAW software with no added noise or distortion.“

 

Granted, they are not literally saying that gain pots are not required but the direction is clear. 

9 minutes ago, Michael Wynne said:

what about Scorpio?

 

Scorpio already has 32-bit converters. I don’t know whether fixed or float, though

Link to comment
Share on other sites

Good questions, though I don't have anything to tell you. IMO, choose your tools based on the capabilities they offer today, not what they can and can't offer in the future. In the SD world, we try hard to introduce new features and products when they are ready, though we certainly have missed posted dates.

Link to comment
Share on other sites

On 8/30/2019 at 4:58 PM, jon_tatooles said:

Here is an article with both the sound files and visual representations of files that were correctly, over, and under recorded to three separate tracks in the MixPre II. This gives an idea of the gain invariant nature of 32-bit recording.

 

https://www.sounddevices.com/noise-in-32-bit-float/

Jon, thank you for posting this comparison. I just listened and adjusted the raw poly file here in Pro Tools in a 96k, 32bit Float session with a Apogee Symphony.  I am hearing no discernible difference in the signal to noise input on the tracks when comparing the optimal gain setting to the one set 40db below.  Very impressive!

 

However, I will say that I thought I heard a better sonic performance and detail from the track recorded at the optimal level.  It was very subtle but noticeable to me.  Just to make sure I wasn't fooling myself I through it 180 degrees out of phase with matched levels and could still hear portions of the track coming through. One could still take the position that the pre amp / converter still prefers an optimal input level.   But I will not deny that the 32bit float has many advantages in recording and post.  Congrats to you and the team! 

Link to comment
Share on other sites

So functionally speaking, in terms of everyday use, mix as you normally do by trying to correctly have all phases of audio in your chain properly set level wise. Then, rest assured that any surprise super loud level, or, level really low by an actor or sound can be readjusted in post... back to a nice workable largely undamaged and-or perfectly usable file.

Is this the SD version of a never clip style system??

 

I will say again, I sure hope we get a notification that this can be implemented Firmware update wise in my 6 Series recorders, and the new Scorpio... 

 

8 hours ago, jon_tatooles said:

IMO, choose your tools based on the capabilities they offer today, not what they can and can't offer in the future.

Of course, but once we already have our recorders, those after the fact additions to those recorders sure are nice... hence, Dugan Auto mix and the Auto mute, both of which I cherish.. Thanks Jon!!

Link to comment
Share on other sites

7 hours ago, afewmoreyears said:

I will say again, I sure hope we get a notification that this can be implemented Firmware update wise in my 6 Series recorders, and the new Scorpio... 

 

Scorpio I don’t know, because it does already have 32-bit converters. 6 series I‘d be astonished if it were possible, unless they had somehow installed a 32-bit converter without telling anyone. 

7 hours ago, afewmoreyears said:

is this the SD version of a never clip style system??

 

I know this may well lead to a heated discussion, but I personally believe that this is better than NeverClip, because here it is actually true. NeverClip is, afaik, still 24-bit. So if you manage to somehow clip NeverClip then it is and will remain clipped

Link to comment
Share on other sites

A comparison between 32bit float pre to NeverClip isn't just a simple math calculation. 

I remember the day when 32bit float DAW appeared, Protools was still using 48bit integar for their gear. The comparison was not an easy debate.

Here, I don't know how exactly a 32bit float AD converter can treat a signal. 32bit float is still a 24bit plus 8 bit of exponent. It's digital noise floor is the same as a 24bit. Even if it can handle a sgnal above OdB, its caractor depends how you design a physical dynamic range electronically. NeverClip can handle a signal goes over 24bit limit but their recorder stays on a 24bit file format (so you have to put it in a 24bit = it may clip).

And as we see several folks report that all digital signal chain contains also a digital noise floor, I believe it really depends on how it is physically designed the circuit.

Link to comment
Share on other sites

5 hours ago, Masaki Hatsui said:

32bit float is still a 24bit plus 8 bit of exponent. It's digital noise floor is the same as a 24bit

 

Yes, but the 8 bits are for extra headroom and increased dynamic range, so still an important bonus, I think. 

SD‘s quoted dynamic range of 142dB would be very hard to squeeze out of a 24-bit converter, so they either have multiple 24-bit converters or a single 32-bit converter. Either way, if implemented well, the advantages could be significant. 

Link to comment
Share on other sites

On 8/30/2019 at 8:40 AM, jon_tatooles said:

IMO, choose your tools based on the capabilities they offer today, not what they can and can't offer in the future.

 

Great advice. 🙂 It is nice when manufacturers back-port features where possible, but as we have seen many times (from many companies), even features that were promised when a product was announced do not always ship - sometimes they just cannot be made to work. If the hardware cannot do it, the hardware cannot do it.

 

As the owner of a 688, I look forward to the software update that will add DANTE support. 😉

Link to comment
Share on other sites

SO, another possibly dumb question: how would I ride my tracks using 32bf ?

I mean the gain on ISO tracks vs, say, L & R tracks according to what I see on the meters ? 

I don't get how it will all ''behave'', especially usual visual references I use for my gain structure..? 

I think I got the theory not too bad, but it's to translate that into my day-to-day workflow in relation to the machine I'm having difficulty to figure out.

Link to comment
Share on other sites

If someone thinks that the use of 32-bit floating point precludes them from the need to understand and apply proper gain-staging, they're akin to the kind of person who would lay an electric hair dryer on the edge of the tub while bathing.   

 

Translation:. Disaster could accompany your ignorance at any moment. 

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...