Gain staging within a DAW

[meLoCo_go]

I've got used to cascading my busses within a project, and recently started to wonder what happens when one of the busses clips that is attenuated downstream, so overall level should't exceed 0.
Like buss clips by 2 dB occasionally and is rooted to the next buss with say -4 dB gain.
With almost all modern DAW utilising floating-point math should I beware of internal clipping if end level do not exceed zero?

[otek]

This is a great topic - It's actually come up more than once for me lately, but I haven't reached any scientific conclusions. To a degree, it seems like most DAWs I've worked on sound better when you hit a few reds - not enough to clip the living shit out of it, of course, but little 2-3 dB overs.

I'm gonna do some experimentation tonight if I have the time (got a murder mix session coming) and report back.

[malice]

In 32 float, cascading your busses shouldn't result in clipping. The headroom of the system is enough to handdle this.

Now for 48 fixed, like in PT, I don't know ...

malice

[ffaudio]

Usually, unless I hear obvious distortion, I say to hell with the little red lights. If it sounds good then I'm happy.

Of course I'd like to hear about this from a more knowledgeable person on the computer scene.

[bunnerabb]

One of the main reasons I sum to a console OTB is dicking around with all of the internal gain staging ITB with limited bandwidth ( 0=0.) shits me up the paint.

Gimmie electrons and tape for the last go.

Tape = instant dither.

[bunnerabb]

That is to say, if I'm maxed on the ITB 2 buss, but I'm coming out of the converters into a desk, channel per channel, then I can use the channel gains on the desk to set my gain scheduling and pans, no? Mo better?

It is to me. ITB is a pain in the ass at mixdown, IMHO.

I can do it but never as fast or as lucidly as I can with a console.

I do NOT like working with weights on my feet.

[Bob Olhsson]

Usually, unless I hear obvious distortion, I say to hell with the little red lights. If it sounds good then I'm happy...Unfortunately that's analog thinking in the digital world!

Digital clipping can be pretty hard to hear especially when it's only one element within the context of a mix. The problem is that it generates nonsense numbers rather than numbers that represent the sound of clipping. It remains innocuous until you start using the nonsense numbers to calculate a change in sound. The result is that classic digital glare that everybody hates.

A lot of us have learned the hard way that it's better to error on the low side than on the high side when it comes to digital levels. This is compounded by the fact that all but high-end digital gear suffers from wimpy power supplies that can't handle the dynamic demands of music when it is run close to the top of the analog stage's range.

Years ago one of the first things an engineer had to learn was where the sweet spot for levels was in the signal path that you wanted to use. The way we checked this out was by running tones through the console and each piece of outboard gear. We turned the tone up listening for where each piece of gear clipped and mapped out the optimum place to insert each piece of gear and the appropriate gain settings for each situation.

As I said, digital clipping can be more subtle to hear but this is a really valuable homework exercise for anybody who is trying to do serious work using a DAW and plug ins.

No digital code is written with the idea of being used outside of range. You can get away with more in a floating point environment but there is no reason to assume you don't need to worry about levels and gain structure.

[malice]

No digital code is written with the idea of being used outside of range. You can get away with more in a floating point environment but there is no reason to assume you don't need to worry about levels and gain structure.

Guys : read this again. If you don't get it : read it again.

You have to be carefull in digital. You can mess things up if you don't understand what you're doing.

Even if I feel more relaxed with floating point regarding headroom, I always make sure I don't clip something in the process.

Be carefull with that.

malice

[eagan]

...... recently started to wonder what happens when one of the busses clips that is attenuated downstream, so overall level should't exceed 0.
Like buss clips by 2 dB occasionally and is rooted to the next buss with say -4 dB gain.
With almost all modern DAW utilising floating-point math should I beware of internal clipping if end level do not exceed zero?

I copied the above and highlighted a bit to something that shouldn't sneak by.

Like Bob pointed out, once something clips, in digital, you have garbage numbers for amplitude values for any samples that exceeded max value (clipped). One that happens, that's trash, and it doesn't matter whether or not the levels of that audio are attentuated downstream. It's already damaged.

It's easy to appreciate the reasoning in why Bunnerabb does things the way he does, because basically, what it comes down to with audio as data in the digital domain is that you don't want to reach clipping level ever at any time at any point anywhere along the way, because once it happens, you're fucked, and there's no fixing it anywhere along the way downstream, short of going in with an editor and finding and removing those glitched pieces of audio (which is hardly a real fix).


JLE

[imagineaudio]

So what about returning the DAW outputs of a full mix back to the inputs and clipping the input channels on your DAW for more volume? I admit I've never done this (more level is best left to the ME), but have heard some AE's do this to get a reference copy hot enough for the client?

[otek]

So what about returning the DAW outputs of a full mix back to the inputs and clipping the input channels on your DAW for more volume?

You mean basically recording the mix back into your daw and hitting the converters harder? I hear some people are doing that, but it would depend completely on the sound of your converters (and that of the limiting circuitry built into some of them).

[malice]

So what about returning the DAW outputs of a full mix back to the inputs and clipping the input channels on your DAW for more volume? I admit I've never done this (more level is best left to the ME), but have heard some AE's do this to get a reference copy hot enough for the client?

It's a well known trick.

Except : it is usually better to clip your DA converters and record back a clean signal.

my 2 cents

malice

[Fulcrum]

I'd think that the technique is OK as long as you're not getting overs going back into the ADC/sound card; in other words, in mixing as well as tracking, you still need to watch out when you're dancing with 0db digital. Malice, Otek, am I correct?

[Pimp-X]

I keep levels in mix as hot as practical. Sometimes there are overs, but that generally doesn't matter - but your milage WILL vary depending on your mix engine's design.

In my experience it is good practice to keep input levels to aux effects such as reverbs as hot as possible - they seem to get quite grainy if you're not using all the bits you can on input.

[chrisj]

Either way if you're resampling you're putting in at least one anti-aliasing filter. Clip out of DAC- reconstruction filter, maybe the DAC handles the level better than your ADC. Clip by coming into the ADC too hot- anti-aliasing filter, same deal: maybe it handles it better. Either way you have a built-in brickwall filter... clipping in the DAW you do not, and further processing might do seriously ugly things with the clipped samples. As Bob says, garbage data.

It's garbage data because it doesn't even necessarily follow the rules for what samples can be in a bandlimited sampling system... it is possible for samples to have the value of 'WRONG!'. Just being a recorded number isn't enough, the sound is represented by combinations of samples.

[myrtlebacker]

Even if I feel more relaxed with floating point regarding headroom, I always make sure I don't clip something in the process.


I think an optimal DAW would use floating point unsaturated internally until it's time to commit to output and in that case it shouldn't matter how hard you drive a channel before the final mix to the converters (then it suddenly matters :)).

On the other hand... plugins - even those bundled in the shrinkwrap - might not be so well behaved or require a different representation internally, in which case they might be saturating your samples (clipping) or truncating (creating "random" noise) on input.

This is so vendor specific, it's going to be trial and error. The nice thing, it's all digital, so you can fairly easy find out what's happening by pushing a sine wave through and analyzing the resultant waveform.

[meLoCo_go]

I think an optimal DAW would use floating point unsaturated internally until it's time to commit to output and in that case it shouldn't matter how hard you drive a channel before the final mix to the converters (then it suddenly matters :)).

On the other hand... plugins - even those bundled in the shrinkwrap - might not be so well behaved or require a different representation internally, in which case they might be saturating your samples (clipping) or truncating (creating "random" noise) on input.


Umm...
I think this sums it all nice.
So, make be this would be a good place to share your experiences - how the specific DAW handle overshoots, how the plug-ins behave? Anyone will? Otek?))

[robot gigante]

The nice thing, it's all digital, so you can fairly easy find out what's happening by pushing a sine wave through and analyzing the resultant waveform.


If you have the time to do that... it's a lot easier to gain or trim things down before you start mixing enough so that there's no chance in hell to clip the plugins at all, if you get stuff that's tracked too hot.

Do that and gainstaging within a DAW becomes a lot easier.

I mean, forget how the DAW may handle overshoots, just make sure there isn't a chance for any to happen to begin with and then you can concentrate on the more important act of actually mixing, right?

[imagineaudio]

I just tried to type a question and after erasing and retyping, like, 3 times, I give up. I have a general idea of the theroy, but not enough for my questions to make sense. there's something about this subject that just makes my brain twist up inside my skull.

I just keep everything low in the DAW, and if I'm getting a little noise, so what...

Maybe it's time for me to spend a bit of time with a tone generator, an analyzer, and a strong cup of coffee.



Edited to add:

I remember reading a post by otek at the other place where he mentioned that DAW's have a bit of analog modeling built into the mixer. Though I could be mistaken, wouldn't be the first time. How would this play into gain staging?

[myrtlebacker]

I mean, forget how the DAW may handle overshoots, just make sure there isn't a chance for any to happen to begin with and then you can concentrate on the more important act of actually mixing, right?

It wouldn't do it for me. I'd like to know how my tools perform even in borderline cases, so I can do the gainstaging with some confidence and in case I inadvertantly or advertantly overshoot, I know what happens. I can also more easily pinpoint problems.
With knowing I get confidence, with not-knowing I get confused :)

[Bob Olhsson]

...With knowing I get confidence, with not-knowing I get confusedThis is true. On the other hand some testing of low levels will demonstrate that it's lots less of a problem than most people think so you really can simply drop your levels and not be distracted by worrying about it.

The problem is that the folks who created the DAWs didn't know the first thing about audio gear or audio production. If they had, we wouldn't even be talking about digital clipping or dither.

[robot gigante]

It wouldn't do it for me. I'd like to know how my tools perform even in borderline cases, so I can do the gainstaging with some confidence and in case I inadvertantly or advertantly overshoot, I know what happens. I can also more easily pinpoint problems.
With knowing I get confidence, with not-knowing I get confused :)

I understand. I guess to me this kind of testing hampers my workflow when I'm mixing so I'd rather just drop things down to a level where I don't have to even think about it.

[Jimsmack]

Would it be safe to mention Paul Frindle's advice on this issue that can be found elsewhere? I would paraphrase but for the sake of not fucking it up I won't... Google away!

[Skwaidu]

This is true. On the other hand some testing of low levels will demonstrate that it's lots less of a problem than most people think so you really can simply drop your levels and not be distracted by worrying about it.


Very true.

[GeeWhoLeeo]

Very true.


very verissimo true
keeping your levels down to -6 in every channel AND the mix bus should be totally safe.
correct me if i'm wrong but i think this has a lot to do with snare and kick beats (or any other poweful transient)that aren't measured very well in the digital meters: i mean if those meters tell ya that the the snare is clipping at -3 there's probably a quick clipping of the sound that exceeds -3 and goes up to 0, or even more.
that thing makes the daw screw and all that shit happens, with the sound field collapsing.
this is what i understood in theory, and in practice keeping my levels down has helped a lot.

[imagineaudio]

i mean if those meters tell ya that the the snare is clipping at -3 there's probably a quick clipping of the sound that exceeds -3 and goes up to 0, or even more.


Now, maybe I've been wrong, but aren't the meters in most DAWs peak meters by default? I understood that they aren't actually reading level per say, but samples.

Any time I've analyzed the properties of a wav file bounced down from a mix, I've never had loudest sample peak louder than my master meter indicated. For instance bouncing down a mix that peaks at -3 when analyzed in sound forge will show the loudest sample hits -3 not -1 or -0.3, etc...

?

[Skwaidu]

Now, maybe I've been wrong, but aren't the meters in most DAWs peak meters by default? I understood that they aren't actually reading level per say, but samples.



You have it right. At least in Pro Tools I do trust my meters, and have not yet once been fooled by them. However, as I'm constantly dancing just below the digital ceiling on my 2 buss I have on some occasions just missed a clipping plug-in or what have you... :lol:

[Brendo]

I thought the PT meters always blew their loads a touch early for some reason, as a "safety"?

[Skwaidu]

I thought the PT meters always blew their loads a touch early for some reason, as a "safety"?

Not as far as I know... ADATs did do that. Do a test? :P

[Pimp-X]

Agreeing with Skwaidu. Peak is peak.

[meLoCo_go]

Agreeing with Skwaidu. Peak is peak.
Other than that inter-sample clipping...
However you must have pretty much Hi-Freq information and be unlucky with phase)

[Pimp-X]

If only I could bring myself to give a rats about samples that don't exist, eh? :)

[Skwaidu]

I thought the PT meters always blew their loads a touch early for some reason, as a "safety"?

Hmm. I gotta add that I guess this notion(which to my knowledge is false) might come from the fact that a digital clip might not be audible or bad sounding on transient material, to a point... Hence some folks use mild clipping instead of lookahead limiters for "punch".

But anyway this applies to all digital, not PT only...

[Skwaidu]

correct me if i'm wrong but i think this has a lot to do with snare and kick beats (or any other poweful transient)that aren't measured very well in the digital meters: i mean if those meters tell ya that the the snare is clipping at -3 there's probably a quick clipping of the sound that exceeds -3 and goes up to 0, or even more.
that thing makes the daw screw and all that shit happens, with the sound field collapsing.
this is what i understood in theory, and in practice keeping my levels down has helped a lot.

Are you summing ITB or analog? Anyway, I'm guessing that the explanation to your findings would actually lie in the behaviour of your converters...

I don't *really* buy the notion that the PT buss would sound vastly different depending on level. At least I haven't really noticed anything like this. It's just math, ainnit?

(But it's a totally different game when you start adding shit, both analog and digital, to the 2... Suddenly gain staging becomes much more important.)

But yeah, I don't like to intentionally clip things. Better to err on the side of safety, like Bob and Malice have pointed out.

[meLoCo_go]

If only I could bring myself to give a rats about samples that don't exist, eh? :)

Dunno, probably the older DAC would crap out... Modern ones are oversampled and can tolerate stuff in-between, um it's getting too geeky)

[Skwaidu]

Now for 48 fixed, like in PT, I don't know ...

malice

Ok, I had to to do a quick test just to reassure myself.

I fed a full scale 1kHz sinewave to a PT bus(bus 1) from an aux track(A). It hit the red, until I backed it 0.1 db. There you go Brendo! Back to FS, I created a master fader(M) for this bus(1) and sent it to another aux track(B), then one more bus(2) and aux track(C) with PAZ on it so I could see as well as hear the clipping.

After this I raised the 1st sine wave- channels(A) output to +12 db. Clipping, audible and visible. I lowered the master fader(M) of the 1st internal PT buss(1) the same amount, back to clean. I returned this Master Fader to zero and lowered the *second* aux track(B, input bus 1- output bus 2) the same amount, and it was still clipping but lower in volume.

I don't know how PT LE or other floating systems would handle this but this is exactly like I thought PT HD would handle it. A channel output *has* heardoom over 0, but if it is not brought to the correct "scale" with a master fader, a cascaded input(Or the D/A converters) *will* clip. A channel output has *as much* headroom over it's meter's zero as it's corresponding output's master fader is lowered, though max for a single channel without any summing is +12 db(The mixer's maximum volume value). Shit, am I making sense at all? :lol:

[Brendo]

(del)

[volthause]

Ok, I had to to do a quick test just to reassure myself.

.....

Shit, am I making sense at all? :lol:

you lost me at "Ok."

:lol:

[Skwaidu]

(del)

? :)




Volt, yeah... I lost myself just a tad before that. ;)

[Skwaidu]

You have it right. At least in Pro Tools I do trust my meters, and have not yet once been fooled by them. However, as I'm constantly dancing just below the digital ceiling on my 2 buss I have on some occasions just missed a clipping plug-in or what have you... :lol:

Btw, my test did indicate one new thing to me, which is of no consequence to me regarding the way I have been doing things but interesting nevertheless...

A clip indication on a channel output, in the case of raising the fader above unity, doesn't mean that audio is getting clipped, if that output's corresponding master fader(if even present) is lowered by a larger amount than the channel is raised above unity... The clip red lights up but the audio isn't clipping.

However the lowered master fader helps nothing if the channel fader is at unity and clipping happens at a plug-in output...

[Bob Olhsson]

Lots of plug ins clip.

[GeeWhoLeeo]

Are you summing ITB or analog? Anyway, I'm guessing that the explanation to your findings would actually lie in the behaviour of your converters...

I don't *really* buy the notion that the PT buss would sound vastly different depending on level. At least I haven't really noticed anything like this. It's just math, ainnit?

(But it's a totally different game when you start adding shit, both analog and digital, to the 2... Suddenly gain staging becomes much more important.)

But yeah, I don't like to intentionally clip things. Better to err on the side of safety, like Bob and Malice have pointed out.

i'm mixing both analog(soundcraft ts12) and digital (pro tools LE with swissonic converters).
the console has clearly a sweet spot, push levels beyond that point and it sounds like a flushing toilet.
same for digital.
i don't know, i feel the soundscape getting really 2dimensional if i'm close to 0db in the masterbuss.
it's like something in the kick and snare get lost, not a pleasant feeling at all.
i don't know if i'm hallucinating. must be the wine ola left here (just kidding Ola, the wine is still safe, locked and hidden in a secret location :lol: )

[GeeWhoLeeo]

Lots of plug ins clip.
amen!
some are a pain in the arse, like URS plugs.. very delicate.

[myrtlebacker]

Just a thought:

If I have a track that hit's the "red light" and then add for instance a reverb plugin, this will necessarily increase the volume of the sample. Original Sound + Echo -> Louder. I won't notice it though, because the "red light" is already on...

Now the guy who writes such a reverb plugin, he could try to estimate a very conservative amount of dynamic room he needs and attenuate the volume on input accordingly, so that there won't be overflow. I am pretty sure his users would not approve of this, because the track would become much quieter.

As an alternative, he could saturate the overflow, so that it distorts somewhat gracefully. This sounds preferable, but it depends on the plugin environment. Maybe the processed data is fed through a secondary plugin, which actually turns down the volume. Then there would be no problem and the saturation would be unneccesary and harmful.

Most likely plugin authors just ignore the problem. You get clipping (ugly sounds) and it's the AE's job of figuring out, when things go wrong.

P.S. Just looking at the "blinken lights" will not help, to see if an overflow occured inside the plugin. The sample might have been scaled internally to overflow (clipped or saturated) and then scaled back down to "green level"

[Brendo]

? :)
I typed a question but then you answered it in your next post, which was above mine. Can't delete posts on here, so, (del) it was.
Now the guy who writes such a reverb plugin, he could try to estimate a very conservative amount of dynamic room he needs and attenuate the volume on input accordingly, so that there won't be overflow. I am pretty sure his users would not approve of this, because the track would become much quieter.


digi's d-verb starts with input att. at -4dB...

[Skwaidu]

Lots of plug ins clip.

Yeah, in my experience, almost all of them can easily clip if used carelessly. Wise gain staging needs to be remembered, at every step in a mix...



I typed a question but then you answered it in your next post, which was above mine. Can't delete posts on here, so, (del) it was.


Ah. :)


digi's d-verb starts with input att. at -4dB...

...Though I *never* have a reason not to crank it to 0 in normal use... IMHO more of a PITA than anything else. (D-Verb is actually great for some stuff! :Wink: )

[Brendo]

Dverb is the only reverb I have, and nobody's complained yet! Shits all over every reverb included with Logic Express (which I run for MIDI)...

[bunnerabb]

Lots of plug ins clip.

Oh.. I don't know. I'll plug in a reverb or two, on the aux busses, some comps on the submasters, comps on the channels and nothign clips for me... ah. oh.. wait...


That's hardware.

Sorry.

:Wink:

[Empty Planet]

If you really want to get into this subject, you might want to consider Paul Frindle's reply on the fifth page of this (http://recforums.prosoundweb.com/index.php/mv/msg/4918/0/64/0/) thread.

It's not easy reading, necessarily, and the solution is merely to mix so that one's faders stay below -6db till you hit the master, but the point is well taken that when it comes to plugins, we seem to be rather credulous toward plugin makers and how they choose to define and measure clipping.

His test is interesting, to say the least.

Cheers all.


:)

[otek]

Dverb is the only reverb I have, and nobody's complained yet! Shits all over every reverb included with Logic Express (which I run for MIDI)...

Yeah, unfortunately with Logic Express you are missing out on the only good reverb Logic ever developed, the Space Designer.

The older Logic reverbs are pretty weak.

The Space Designer on the other hand, sounds fantastic.

[juergen]

Dverb is the only reverb I have, and nobody's complained yet! Shits all over every reverb included with Logic Express (which I run for MIDI)...

The only potentially usable setting i could find in Logic's older verbs was a slightly modified version of the default setting of the platinum verb...

Everything else sounds funny.

I'm with Otek on the Space Designer.

Speaking of plug-in clipping - I've noticed a slight change in sound just by activating some of Logic's plug-ins (i do most of my work in Logic), as if they were trying to "enhance" the sound in some way...but i haven't done any serious testing.

[Mikebuzz]

Terry Manning and Paul go over this in depth over at the PSW Whatever works forum , Lot's of good info and ideas regarding digital levels.
I dont know how to use reverb so I'll stay out of that one ????

LAter
Buzz