Is that +62dBFS headroom a really really bad idea?

[aquashack]

I recently logged on to this forum intending to ask some newbie questions to do with gain levels, metering and clipping in Live. I held off as I figured it would be sensible to check out past threads and see what I could find that was relevant.

So this past week I did a crash course thanks to this forum (and Gearslutz). Turns out this was a superb learning experience and I think I gleaned a whole lot more information looking through the archives than I would've from just asking one or two isolated questions. Along the way I read some great articles by Tarekith and over 100 pages of a Gearslutz thread on analogue versus digital and a whole lot on the history of mixing consoles that has given me a better understanding of the hows and whys of modern setups.
So my original questions were covered but I'm looking for clarification on the following statements...

1. The Full Scale to which dBFS refers is the full scale of available bit depth and has nothing to do with any VU meter scales or fader scales. (zero dB on a VU is presumably whatever it has been calibrated to).

2. Any digital signal above 0dBFS will be subject to some harsh clipping, BUT, thanks to the magic of floating point numbers that Live employs, we get 62dB ABOVE zero of headroom on individual channels. I don't fully understand the ins and outs of this but I take it that it's effectively "shifting" the zero point up by as much as 62dB when required (similar to the windowing system in NICAM sound?)

3. The Master channel works at fixed-point 24 bits and so signal can never exceed 0dBFS (0VU) without clipping.

4. Plugins should be treated as real world effects and should be properly gain staged at (a suggested) -12dBFS or lower. Is this because some plugins use fixed-point scales?

So if we're looking to gain-stage each insert-effect along the chain to a sensible level (say -12dBFS) and the final master channel should be kept below 0dBFS then I'm wondering when you would/should ever make use of that 62dB of extra headroom. Surely it's just asking for trouble and some sloppy gain-staging to offer this. Seems to me it allows users to be careless with signal levels and that's all.

[CooCooCaCha]

I recently logged on to this forum intending to ask some newbie questions to do with gain levels, metering and clipping in Live. I held off as I figured it would be sensible to check out past threads and see what I could find that was relevant.

So this past week I did a crash course thanks to this forum (and Gearslutz). Turns out this was a superb learning experience and I think I gleaned a whole lot more information looking through the archives than I would've from just asking one or two isolated questions. Along the way I read some great articles by Tarekith and over 100 pages of a Gearslutz thread on analogue versus digital and a whole lot on the history of mixing consoles that has given me a better understanding of the hows and whys of modern setups.
So my original questions were covered but I'm looking for clarification on the following statements...

1. The Full Scale to which dBFS refers is the full scale of available bit depth and has nothing to do with any VU meter scales or fader scales. (zero dB on a VU is presumably whatever it has been calibrated to).

2. Any digital signal above 0dBFS will be subject to some harsh clipping, BUT, thanks to the magic of floating point numbers that Live employs, we get 62dB ABOVE zero of headroom on individual channels. I don't fully understand the ins and outs of this but I take it that it's effectively "shifting" the zero point up by as much as 62dB when required (similar to the windowing system in NICAM sound?)

3. The Master channel works at fixed-point 24 bits and so signal can never exceed 0dBFS (0VU) without clipping.

4. Plugins should be treated as real world effects and should be properly gain staged at (a suggested) -12dBFS or lower. Is this because some plugins use fixed-point scales?

So if we're looking to gain-stage each insert-effect along the chain to a sensible level (say -12dBFS) and the final master channel should be kept below 0dBFS then I'm wondering when you would/should ever make use of that 62dB of extra headroom. Surely it's just asking for trouble and some sloppy gain-staging to offer this. Seems to me it allows users to be careless with signal levels and that's all.

First a little info on how numbers work in computers and computer audio.

FLOATING POINT:
A floating point number is actually represented by two numbers in a computer. Think of it as being in scientific notation. 1024 in floating point would be 1.024 x 10^3. 10240000 would be 1.024 x 10^7. Notice how similar these two numbers look? The only thing that changed is the exponent. This is why you have so much headroom in floating point audio, all that really changes is the exponent. Now, the exponent can only be so big which is why there is a limit like you said (64dB).

INTEGERS:
On the other hand integers are only one number and do not have decimal places. So with integers you cannot have a number like 1024.41. The digits to the right of the decimal would either be removed or rounded to form a whole number.

DIGITAL AUDIO:
Digital audio can be either floating point or fixed point. With floating point 1 and -1 are considered maximum volume, however since these are floating point numbers you can make them pretty much as loud as you want. However, there are two things to keep in mind.

1. Plugins might distort when fed a signal that is too loud. This isn't really a problem with the numbers themselves but the plugin itself. A compressor, for example, generally have thresholds below 0 so if you feed it a really loud signal it will compress the hell out of it.

2. Your soundcard probably uses integer numbers.

Why is number 2 a problem? Well unlike floating point numbers there is a hard limit with integers. Let's say you have an 8 bit number. That number can represent numbers 0 through 255. Well in audio 0 is usually silence like we'd expect and 255 would be maximum volume. If you have a sound that hits 255 then if you try to make it louder, it simply won't do it, or it'll start clipping. This is because with 8 bit numbers you can't go any higher than 255, you'd need more bits to do that. Keep in mind 255 isn't always maximum loudness. a 9 bit number has a maximum value of 511, so if you were using 9 bit numbers your maximum volume would be 511. If you were to convert from 9bit audio to 8bit audio, 511 would become 255.

In audio, converting from floating point to integer is easy. You just take the max integer number (let's use an 8 bit number, so 255) and multiply it by the floating point. So for example 1.0 * 255 = 255.

So onto your questions:
1. In your DAW (lets assume ableton), 0dB refers to the max volume (which would be -1 or 1 in floating point). If your channel fader reads +1dB then that would be 1 decibel above the max volume.

2. Hopefully I explained this above. For reference we are dealing with floating point numbers here.

3. Yes, this is because the master audio is converted to fixed point before it goes to your soundcard.

4. This isn't necessarily true. Maybe if it's a plugin that emulates hardward this would be true but not necessarily for plugins that are made to sound "digital" or "clean". This is a question that should be answered on a plugin-by-plugin basis.

Overall, I think you're trying to apply old school mixing console concepts to the digital domain. These concepts do not necessarily translate properly. For example, an analog mixing console has no concept of floating point so gain-staging would be very relevant in this case. However, it would not apply to Ableton because of how it handles digital audio.

Hopefully this makes sense.

[Tone Deft]

not quite.

outside of Live, in hardware, with DACs and ADCs and AES receivers/transmitters, your soundcard etc audio is transmitted as PCM which is in a 24 bit 2s-compliment format.

the OP should read Tarekith's guides a little closer, maybe Tarekith can take some time to write a few words.



I believe I use the headroom in the same way some people use compressors. run tracks HOT to get full rich deep dynamic range.

editeded for typos.

[aquashack]

Hi CooCooCaCha and thanks for that thorough explanation and also for replying so quickly. All makes sense. I'm finding that there's a lot of fundamentals that need to be fully understood before diving into any of this stuff.

Having come into mixing in the digital world and never having had any experience with analogue consoles and the way they work I figured that reading up a little history and technique might give me a better understanding and perhaps better working practices in Ableton. But yes, you're right these things don't translate exactly and I need to make the necessary adjustments.

[aquashack]

not quite.

outside of Live, in hardware, with DACs and ADCs and AES receivers/transmitters, your soundcard etc audio is transmitted as PCM which is in a 24 bit 2s-compliment format.

the OP should read Tarekith's guides a little closer, maybe Tarekith can take some time to write a few words.



I believe I use the headroom in the same way some people use compressors. run tracks HOT to get full rich deep dynamic range range on tracks.

Tone Deft - revisiting Tarekith's guides next as per your suggestion. So much knowledge to take in and these things always require several reads to sink in properly.

[CooCooCaCha]

not quite.

outside of Live, in hardware, with DACs and ADCs and AES receivers/transmitters, your soundcard etc audio is transmitted as PCM which is in a 24 bit 2s-compliment format.

the OP should read Tarekith's guides a little closer, maybe Tarekith can take some time to write a few words.



I believe I use the headroom in the same way some people use compressors. run tracks HOT to get full rich deep dynamic range.

editeded for typos.

PCM is a series of integer values so I'm not sure what you're getting at. Sure I left off some more technical details but there was no need since it wasn't relevant to what aquashack wanted to know.

[lunabass]

4. Plugins should be treated as real world effects and should be properly gain staged at (a suggested) -12dBFS or lower. Is this because some plugins use fixed-point scales?

i find this point quite interesting and it's one i'm seeing quoted a lot yet every test i've done on any of my hardware emulating plug-ins shows absolutely no benefit at running the input of the plug-in at -12dbfs (0VU).

it's a really easy (blind a/b) test to run using racks. i've tried waves renaissance eq & comp, ssl and a few others with absolutely no difference noticed in any listening tests.

can someone point me to some audio tests that highlight the difference?

i do believe in proper gain staging between plug-ins and i never allow my plug-ins to receive audio at 0dbfs or higher but i cant help but think people are getting themselves confused with hardware. i mean do you really think a plug-in manufacturer is going to allow their plug-ins to perform poorly or worse at input levels between -12dbfs & 0dbfs?

[mharris]

Remember that a large proportion of Live's functionality is aimed at performing live. In this situation the live performer doesn't really have time to spend ages gain staging everything that they add to the performance.

They only have to keep an eye on the master channel and not have to worry about what's going on under the covers. This assumes that only built-in devices are used and no gain sensitive plugins.

[theophilus]

i find this point quite interesting and it's one i'm seeing quoted a lot yet every test i've done on any of my hardware emulating plug-ins shows absolutely no benefit at running the input of the plug-in at -12dbfs (0VU).

i do believe in proper gain staging between plug-ins and i never allow my plug-ins to receive audio at 0dbfs or higher but i cant help but think people are getting themselves confused with hardware. i mean do you really think a plug-in manufacturer is going to allow their plug-ins to perform poorly or worse at input levels between -12dbfs & 0dbfs?

this is an ADC issue IIRC, not necessarily a mixing issue. more about recording @-12dbfs. recording at 12db down gives you a few bits for your compressors, eqs, etc. to eat into without burying the needle.

whether the plugin responds to the input level or not depends a lot on what the plugin is trying to do.
most real HW devices have a 'reference level' which you need to be at, or close, for the hw device to do its best.
sometimes, this is just a HW protection feature (i.e. you overload the input causing distortion if you go too high).
a basic delay, for instance, doesn't do anything but delay your audio - it shouldn't really care what the input level is.

but imagine, for instance, a tube preamp (or any tube gain stage). tubes are nonlinear - up to a certain level, they are (fairly) clean, and then they start compressing, then overdriving. set to the same gain, the output will vary based in the input level, as the input level * gain will determine where you are in that curve at the output. a very low level won't be amplified out of the clean range... a much higher input level may go all the way into overdrive.

if you have a guitar amp sim, and you want it to respond with your guitar same as the HW, you need to know where the plugin treats its reference. Typically, guitars are at -20db or lower, but you may want to turn the gain up in your preamp for better SNR or something. Let's say the guitar amp expects a guitar signal at -20db and the simulation is designed so when the input is at -20, the amp stages are in the same parts of the curve that they would be in HW. if you send it a -12db signal, that's equivalent to putting a clean boost pedal in front of the amp. It's safe to do that, but it is a different sound. OTOH, if the amp sim wants -12, and you send in -20, that's like turning down the volume on your guitar. so the input gain is important to get right.

[Angstrom]

Bearing in mind a lot of the technical stuff can become seem occult to the newcomer, my advice is :

The magical headroom is a safety net, not a foundation. Try to keep all your meters out of the red, try to keep everything in the safe zone, and all will be well. Live does have a lot of headroom in its internal paths, but a plugin or an External Instrument could still screw you over. Pretend the magical headroom isn't there. It's your fairy godmother, looking out for you just in case you need a wish granting.

[Tone Deft]

not quite.

outside of Live, in hardware, with DACs and ADCs and AES receivers/transmitters, your soundcard etc audio is transmitted as PCM which is in a 24 bit 2s-compliment format.

the OP should read Tarekith's guides a little closer, maybe Tarekith can take some time to write a few words.



I believe I use the headroom in the same way some people use compressors. run tracks HOT to get full rich deep dynamic range.

editeded for typos.

PCM is a series of integer values so I'm not sure what you're getting at. Sure I left off some more technical details but there was no need since it wasn't relevant to what aquashack wanted to know.

I know you don't get what I'm talking about, go read.
http://www.cirrus.com/en/applications/a ... APP25.html

the twos complement values are normalized to whatever output standard you're measuring to, which can be integer values or numbers between -1 and 1.

[CooCooCaCha]

I know you don't get what I'm talking about, go read.
http://www.cirrus.com/en/applications/a ... APP25.html

the twos complement values are normalized to whatever output standard you're measuring to, which can be integer values or numbers between -1 and 1.

Two's compliment numbers are integers so converting them to integers doesn't really make much sense unless you're talking about another integer format. Of course they can be converted to whatever.

I really don't see how you're information really helps the OP's understanding. He's clearly just started learning how audio works in the digital domain so overcomplicating it is counterproductive.

[justin]

Here's my attempt at clearing things up:

dBFS is only relevant within digital audio systems, with 0 being the absolute maximum. Anything beyond 0 will lead to clipping as the DAC on your sound card cannot go beyond it.

Be wary of saying "X" dBVU (and other analog scales) is equal to "Y" dBFS. dBVU is a reference level, and in a pro audio setting, -18dBFS is often used as the 0dbVU calibration level (note this is usually done by sound card manufacturers). Another thing to bear in mind is that different areas of audio production (post / broadcast / music) use different calibration levels!!!

I'm not so sure about the inner workings of Live, but i would suggest the mixer uses additional bits to increase the headroom (32-bit). I presume this is to make the mixer more flexible, and to an extent forgiving... Not sure what happens with the final mix bus (master fader), and i'm guessing it would be determined by the sound card settings (ie. 44.1 Khz / 24 bit).

I'm also not too sure on the idea that Live would switch processing from floating to fixed point processing. I would have thought that it was all 32 floating point. Things might differ a bit with plugins, but i would still expect the majority to be floating point.

In terms of gain-staging, its good practice to pay attention to levels as they go through the plugin chain. A compressor plugin can only work if there are dynamics within the threshold band. However, you may find some of them clip nicely when pushed so YMMV and remember to trust your ears

[aquashack]

4. Plugins should be treated as real world effects and should be properly gain staged at (a suggested) -12dBFS or lower. Is this because some plugins use fixed-point scales?

i find this point quite interesting and it's one i'm seeing quoted a lot yet every test i've done on any of my hardware emulating plug-ins shows absolutely no benefit at running the input of the plug-in at -12dbfs (0VU).

it's a really easy (blind a/b) test to run using racks. i've tried waves renaissance eq & comp, ssl and a few others with absolutely no difference noticed in any listening tests.

can someone point me to some audio tests that highlight the difference?

i do believe in proper gain staging between plug-ins and i never allow my plug-ins to receive audio at 0dbfs or higher but i cant help but think people are getting themselves confused with hardware. i mean do you really think a plug-in manufacturer is going to allow their plug-ins to perform poorly or worse at input levels between -12dbfs & 0dbfs?

You're right, it's all hear-say as far as I'm aware. The -12dBFS is something I heard a while back. The suggestion being that there's some indiscernible noise/rubbish that appears in each plugin above this input level and the overall effect for the (possibly dozens) of plugins leads to something that does become noticeable (if not entirely tangible).

I believe the actual -12dBFS figure was mooted because Ableton themselves set the Simpler default o/p level to -12dB but beyond this I'm not sure if there's any solid proof.

[aquashack]

Be wary of saying "X" dBVU (and other analog scales) is equal to "Y" dBFS. dBVU is a reference level, and in a pro audio setting, -18dBFS is often used as the 0dbVU calibration level (note this is usually done by sound card manufacturers). Another thing to bear in mind is that different areas of audio production (post / broadcast / music) use different calibration levels!!!

Terrific stuff - thanks for the comments. I'd noticed that -18dBFS = 0dBVU was used in a few places as a default calibration (Sonimus Satson for instance has this as standard but also provides -14dBFS and -20dBFS as options). It took me a while to work out exacty what that meant to me as a user. But I see it basically lets me gain-stage to my choice of hotter or cooler signals and see that signal represented as 0dBVU on the plugin meter.

My questions in this thread actually all started when I bought a copy of Satson and was trying to understand what kind of levels I should be feeding it. Initially I initially inferred from the user guide that I should whack in close to 0dbFS or more per channel. I even thought that full-scale meant the position of the VU needle to the far right (I know, pretty dumb but there you go!). Eventually I sussed it all out.