32 bit floating point

[benw]

so- logic is 32 bit floating point. which means that it cannot clip internally. right? which means that if i have a microphone plugged into my mixer, going into my interface and then into logic, i cannot get a signal to clip when i play it back inside logic? right? or possibly wrong as the desk might clip?? so...the only point in recording thru mixer and interface into logic and then playing back thru speakers where logic would clip would be if my speakers caused it to clip... right? or not possibly??

 

Can anyone help me out with this?

 

Cheers,

Ben

[shivermetimbers]

so- logic is 32 bit floating point. which means that it cannot clip internally. right? which means that if i have a microphone plugged into my mixer, going into my interface and then into logic, i cannot get a signal to clip when i play it back inside logic? right? or possibly wrong as the desk might clip?? so...the only point in recording thru mixer and interface into logic and then playing back thru speakers where logic would clip would be if my speakers caused it to clip... right? or not possibly??

 

Can anyone help me out with this?

 

Cheers,

Ben

YES! The signal can clip internally with only a Mic (screaming loudly) connected to an interface (Turned all the way up) then into Logic. I just did it. If you add more pre amplification you will get more clipping.

 

You can also use plug-ins like compression to boost the signal and get it to clip.

 

Speakers are connected to your Amp and if the signal is hot enough out of Logic and you have your amp cranked, it is possible to over drive the speakers.

[benw]

so this whole idea of 32 bit floating point thanng presenting clipping... at which point in the signal path DOES it present clipping?

 

Cheers,

Ben

[shivermetimbers]

The 32 bit is more about the processing. It isn't going to present clipping.

 

Overloading your input signal is a different thing altogether.

 

Here is some good reading:

 

http://www.applied-acoustics.com/techtalk2.htm

 

The bottom line is to antenuate the input signal to get the best level without distorting it. The input levels are something you have control over. How you process it before, during, or after you record it could possibly add to an overdriven signal.

 

If the signal is too low, you may end up amplifying noise along with your signal. So trying to keep it at that '0' or '-3db' level is a good practice.

 

If you are trying to keep the clipping out of the signal path, then pay close attention to your settings and don't overload the signal at any point in the path.

 

Are you having problems with clipping?

[benw]

no, no problems but i read a rather confusing thread about the actual use of 32 bit floating point and how it prevented clipping without explaining where or how or why it prevented clipping. when you say 0 or -3db level, would that be the same as 0 or -3 on the level meter if its set to measure rms?

 

thanks a lot for your help, as you can see im a bit confused bout this whole thing but ill check out the link!

[David Nahmani]

When recording audio, Logic takes the 0 and 1s sent by the converter, representing the 24 bit fixed point audio, and places them in an audio file, THAT's IT!!! You can place the fader wherever you want on your Logic mixer, that doesn't change a thing as far as recording audio. If the converter says "01001110", then that's what Logic puts in the digital audio file. There is NO processing whatsoever, no 32 bit floating point conversion.

 

So if your converter sends a clipped signal to Logic, that's what Logic puts in the digital audio file. So if your mic signal is too loud, you're going to overload your A/D converters, and a clipped signal is recorded. Logic has nothing to do with it, it just places the clipped signal provided by the converter into an audio file. It's the dumbest job in the world, and Logic performs it flawlessly.

 

When you play back, Logic immediately converts the signal to 32 bit floating point, and starts processing it, changing volume, pan position, adding plug-ins, and summing the various audio files. It is virtually impossible to clip a 32 bit floating point audio signal. So as long as you stay in 32 bit floating point, you don't have to worry about any distortion taking place.

 

So when do you have to worry about distortion? When you go back to 24 bit fixed point. Which might be when using a 24 bit fixed point 3rd party AU plug-in (in which case you could be clipping its input, so stick a gain and a meter in front), or when you're going back out of the audio interface, so at the output of the Out 1-2 object.

[benw]

thanks a lot david, that definitely makes it a lot clearer.

[nikkik]

Here we go again...

 

First, in the link provided, in point number 4: I disagree with a few points, mostly stemming from the use of the phrase,"..more accurate..." Pertaining to sample rate, a higher rate does not present a "more accurate" recording; at least, not without a qualifier of what "more accurate" means. In this case, I am hoping the author meant "more accurate" to mean "greater range of frequencies." For recording a signal that is bandlimited (naturally, or thru filtering) to say, 20K, then a sampling rate of 44.1k will record just as accurately as 1,300,564k. PERIOD.

 

Floating vs Fixed.... That article inofitself lacks the content and explanation to properly argue for either "camp" IMO. I would suggest reading more complete works.

 

Recording levels: Bigger sigh...

The first step is to understand analogue gain staging, and proper use of preamps and such. Proper system calibration, digital or analogue. And then when recording using a DAW, it can be simple good practice to allow some "cushion." I would not record a signal hotter than -6dB peak. 16-bit is plenty for almost any musical recording. 24-bit, for operational "freedoms" and safety "cushioning" is great. Understanding how a DAW works, fixe dor flaoting, can enable one to appreciate the reasoning behind certain habits. "Why not record as 32-bit if we are going to use a 32-bit float DAW?" Rough to explain simply and completely. People using Steinberg products opt for this actually. I do not see any reason for it personally. If one understood 32-bit float, they might agree with me. I am curious what lengths they will go to when their app goes 64-bit float... lol...

 

Use proper gain staging, record at 24-bit to allow for more headrom than you will ever need. And if noisefloor is an issue when raising a level after recording at even -12dB peak, then you have some pretty horrible equipment...

 

I would suggest to anyone wanting to actually understand the tools they are using, grab Nika Aldrich's "Digital Audio Explained..." book, and Bob Katz's book, "Mastering Audio..." Properly "tune" your room, or at least get it as close to being accurate and fair. Bass trapping is so incredibly important IMO. Eliminate bad reflections, diffuse, but do not make the room dead. A room that is too dead is..well...dead. No life.

[Ashermusic]

When recording audio, Logic takes the 0 and 1s sent by the converter, representing the 24 bit fixed point audio, and places them in an audio file, THAT's IT!!! You can place the fader wherever you want on your Logic mixer, that doesn't change a thing as far as recording audio. If the converter says "01001110", then that's what Logic puts in the digital audio file. There is NO processing whatsoever, no 32 bit floating point conversion.

 

So if your converter sends a clipped signal to Logic, that's what Logic puts in the digital audio file. So if your mic signal is too loud, you're going to overload your A/D converters, and a clipped signal is recorded. Logic has nothing to do with it, it just places the clipped signal provided by the converter into an audio file. It's the dumbest job in the world, and Logic performs it flawlessly.

 

When you play back, Logic immediately converts the signal to 32 bit floating point, and starts processing it, changing volume, pan position, adding plug-ins, and summing the various audio files. It is virtually impossible to clip a 32 bit floating point audio signal. So as long as you stay in 32 bit floating point, you don't have to worry about any distortion taking place.

 

So when do you have to worry about distortion? When you go back to 24 bit fixed point. Which might be when using a 24 bit fixed point 3rd party AU plug-in (in which case you could be clipping its input, so stick a gain and a meter in front), or when you're going back out of the audio interface, so at the output of the Out 1-2 object.

 

I totally agree with everything David has written here.

 

Slightly OT I would also add as I always do that there are lots of good reasons besides avoiding distortion to observe traditional good mix practices even in a 32 bit float app by turning on pre-fader metering and keeping the levels from going into the red a lot and there is no downside to it as there a plenty of bits available even when the levels are showing pre-fader -6 to -18.

[shivermetimbers]

Here is another article about bits for anyone who likes the technical info.

 

http://mixonline.com/mag/audio_im_sixty_four/index.html

 

 

"...For digital audio, word length translates to dynamic range like this:

 

16-bit = 96 dB

24-bit = 144 dB

32-bit = 192 dB

64-bit = 384 dB"

[David Nahmani]

16-bit = 96 dB

24-bit = 144 dB

32-bit = 192 dB

64-bit = 384 dB"

That's with fixed point. The dynamic in 32 bit floating point is much higher.

[shivermetimbers]

500 at 32?

[ski]

It is virtually impossible to clip a 32 bit floating point audio signal. So as long as you stay in 32 bit floating point, you don't have to worry about any distortion taking place.

 

I beg to differ on this point... For example, it's extremely easy to ge EXS-24 to distort right at the audio instrument itself. I routinely achieve distortion because many of my EXS-24 instrument start out at a level of volume=0dB when really it should be set for -6 or -9 at a minimum. On patches which I'll play polyphonically (strings, voice, brass), if I don't remember to check the output level setting on EXS-24 and I start playing thick chords, I'll be into the red and hearing distortion very quickly.

 

I can't say that I fully understand how the theory of how a 32-bit float format inherently prevents distortion from occurring, but I can state for an absolute fact that my ears tell me that audio instruments -- particularly EXS24 -- will easily and regularly distort if care isn't taken to watch the gain settings.

 

It's based on this real-world experience that makes me suspect of the 32-bit float prevent distortion argument. Even if mathematically/theoretically this should be true, my ears tell me differently.

 

Sorry if I've burst anyone's bubble on this, but in my experience, the theory doesn't hold up.

 

-=iS=-

[shivermetimbers]

The theory holds up.

 

What it is dealing with is how the waveform is processed and not necessarily the sound coming out of your speakers. You will definitely distort the output signal if you push the processing limits in terms of a volume output. The internal audio of my imac handles 16 bit and my Audio interface handles 24 bit. 32 bit will exceed that limit.

 

With 32 bit you have more headroom to play around with your signal processing stage without loosing the quality at lower levels. The waveform will not distort in the 32 bit domain inside the computer, but when you try to force that 32 bit wave form at a higher db level through a 24 bit soundcard, it will clip and you hear the result as distortion.

 

At 32 bit (floating) You have the freedom to process the signal, but you must remember to keep the final signal down to a level that a 24 or 16 bit system can handle.

 

What goes on inside the computer is of no concern to me as long as what comes out sounds good.

 

 

"...Sixteen bits isn’t a lot of space to represent a sound, and the small size becomes a problem if you want to modify digital sounds. Applying an effect to the sound can be restated as “doing complex math on a series of samples.” Every time you apply effects on 16-bit integers, you lose 3dB of sound in the affected area due to the low resolution of the sample. Complex math with 16-bit integers requires a lot of rounding decisions to be made along the way, and that’s one place you will lose sound. The other place you will lose sound is at the upper end when samples start to clip, or exceed the maximum voltage the 16-bit integer is capable of representing."

 

http://macvoip.com/stn/?p=41

[David Nahmani]

It is virtually impossible to clip a 32 bit floating point audio signal. So as long as you stay in 32 bit floating point, you don't have to worry about any distortion taking place.

 

I beg to differ on this point... For example, it's extremely easy to ge EXS-24 to distort right at the audio instrument itself.

 

(...)

 

Sorry if I've burst anyone's bubble on this, but in my experience, the theory doesn't hold up.

 

You haven't bursted any bubble at all, you just mentioned that in some cases you can hear distortion. You haven't demonstrated that this distortion is caused by the 32 bit floating point audio engine.

 

There are many, many ways to produce distortion in a signal chain (the original poster, for example, talked about A/D input distortion). You found one of them. I can guarantee you that is not generated by overloading any "stage" in the "gain structure" of the 32 bit floating point audio engine, since.. well they don't have stages.

 

Ski, I have demonstrated that you can't distort a 32 bit floating point engine by the theory AND the practice. There is a thread somewhere out there where I explain the principles if you want to conduct your own test.

 

Unlike topics like workflow, or techniques, distortion really is not a topic where you can have an opinion, it's a scientific fact: a signal is distorted when it is different from the original signal it was derived from. If you compare two signals and they are identical, no distortion has been produced. By the way, I have also discussed it with the Logic architect and he confirmed my findings.

 

However, if you have actually found something that creates a distortion that doesn't go away when turning the Out 1-2 fader down, I encourage you to upload your song to this thread so we can see if it has anything to do with 32 bit floating point, or if (like I suspect), it is caused by something else altogether.

[ski]

Thanks for your replies guys. I appreciate the info. Time for me to dig out a old file where distortion was occuring at the instrument level in a very obvious manner...

[stevenson]

the reason ski was hearing distortion is probably because he had his esx preferneces set to 'sample storage - original' instead of 32 bit float. you have the option with esx in order to save memory in large lay-ups.

[ski]

the reason ski was hearing distortion is probably because he had his esx preferneces set to 'sample storage - original' instead of 32 bit float. you have the option with esx in order to save memory in large lay-ups.

 

Funny you should bring that up... As I've been thinking about this thread this morning it occured to me that, as you brought up, that I don't have the 32-bit option enabled. So indeed I've been wondering if that might be a contributing factor here...

[Fred B]

the reason ski was hearing distortion is probably because he had his esx preferneces set to 'sample storage - original' instead of 32 bit float.

 

No, that shouldn't matter. The samples are converted anyway when used. I guess the distortion was not caused by clipping, but by specific issues of Logic's plugin handling, like buffer initialization, MIDI chasing conflicts etc.

[stevenson]

No, that shouldn't matter. The samples are converted anyway when used. I guess the distortion was not caused by clipping, but by specific issues of Logic's plugin handling, like buffer initialization, MIDI chasing conflicts etc.

 

actually you may be right. i think the samples are held in memory and converted on the fly - but i am not sure. if ski can have a look at his distorting esx instruments and try switching to 32-bit float we might learn something....

 

in any case, i have never had distortion from esx internally. except after large meals...javascript:emoticon(':oops:')