Headroom for mastering

[chris_shape_drummer]

What are you guys generally leaving for headroom in your mixes? lately ive found -6 on the main from the hardest snare hits to be my sweet spot. Gives me plenty of room for the clip and limiter on my mastering chain. I master pretty loud, usually -6 to -4 LUFS depending on heaviness of the song, so i try to make sure no distortion gets introduced during mastering by getting things compressed right and transients under control.

 

[Craig Anderton]

I've just submitted a tip for the PreSonus blog on gain-staging. It's more nuanced than leaving X amount of headroom. It also depends on whether you're mastering while mixing by using bus processors, or treating mastering as a separate process.

I do the latter, for many reasons. So, my first priority is having signals hit the mixer at an average -18 dBFS level with peaks between -6 and -12 dB. The input trim controls are great for adjusting the incoming levels. (If the peaks exceed -6, I'll dial back the average enough to keep the peaks under -6 dB.) Studio One's VU meter can scale so that -18 dBFS registers as 0 VU, which makes reading the level easier than looking at the low end of the scale. I try to maintain -18 dBFS average through any effects chains.

This allows mixing in a comfortable way in terms of fader travel, avoiding clipping, keeping the output level close to 0, and having signals that hit the sweet spot for amp sims and processors that emulate analog devices. Then the mastering process does the heavy lifting for levels/loudness/tone.

Keep your eye on the PreSonus blog for the upcoming gain-staging posts. There are two parts. The first is about recording and audio engine resolution. The second covers gain-staging.

 

[chris_shape_drummer]

I've just submitted a tip for the PreSonus blog on gain-staging. It's more nuanced than leaving X amount of headroom. It also depends on whether you're mastering while mixing by using bus processors, or treating mastering as a separate process.

I do the latter, for many reasons. So, my first priority is having signals hit the mixer at an average -18 dBFS level with peaks between -6 and -12 dB. The input trim controls are great for adjusting the incoming levels. (If the peaks exceed -6, I'll dial back the average enough to keep the peaks under -6 dB.) Studio One's VU meter can scale so that -18 dBFS registers as 0 VU, which makes reading the level easier than looking at the low end of the scale. I try to maintain -18 dBFS average through any effects chains.

This allows mixing in a comfortable way in terms of fader travel, avoiding clipping, keeping the output level close to 0, and having signals that hit the sweet spot for amp sims and processors that emulate analog devices. Then the mastering process does the heavy lifting for levels/loudness/tone.

Keep your eye on the PreSonus blog for the upcoming gain-staging posts. There are two parts. The first is about recording and audio engine resolution. The second covers gain-staging.

This is awesome info!

 

[Craig Anderton]

If you're interested in the origin story, -18 dBFS is the same level as +4 dBu, like consoles and outboard gear use. In other words, if a signal is coming into Studio One at -18 dB and you diverted it to an analog mixer, it would read +4 dBu. In the analog world, there was typically 20 dB or so of headroom about 0 VU. That same kind of headroom is "built in" to -18 dBFS.

FWIW Waves explicitly states when an analog emulations were was modeled with a reference of -18 dBFS. Line 6 Helix also expects that kind of input level, as do other plugins. Regarding Studio One, I highly recommend that the level hitting Ampire is -18 dBFS. You'll find the controls have a better feel, and the sim will be less likely to overdistort.

 

[SwitchBack]

If you're interested in the origin story, -18 dBFS is the same level as +4 dBu, like consoles and outboard gear use. In other words, if a signal is coming into Studio One at -18 dB and you diverted it to an analog mixer, it would read +4 dBu. In the analog world, there was typically 20 dB or so of headroom about 0 VU. That same kind of headroom is "built in" to -18 dBFS.

As a working proposition that's fine, but technically speaking it's apples to oranges.

Standard analog VU meters are calibrated 0 VU=+4dBu with a scale ending at +3 VU. There is no reference to peaks or headroom. For digital realm VU meters and plug-ins there is no standardised calibration level for 0 VU. -18dBFS is a popular setting yet most meters have a few settings to choose from. But saying "-18dBFS is the same level as +4dBu" is like saying "32°F is the same level as 5cm", even if it's so on your home thermometer

 

[Craig Anderton]

As a working proposition that's fine,

That was my intention. I was under the impression that digital VU meters respect the Full Scale calibration that represents the maximum attainable signal, but that doesn't prevent them from changing the apparent scaling/calibration. For example, when using the PreSonus VU meter to measure the incoming levels to the mixer, I change the scale to -18 so that it reads 0 on the meter. It's much easier visually to set levels precisely at the higher end of the VU meter's scale. But still, the meter is receiving a -18 dBFS signal - I'm just altering how the meter chooses to interpret that signal. Yes?

I appreciate your clarifiying there's no equivalence to headroom between digital and analog devices. That often trips up people just getting into recording. To clarify further if needed for others, digital has a fixed, unalterable maximum level it can accept (Full Scale). The maximum signal level analog devices can accept depends on the circuit design. Furthermore, because distortion often increases linearly over a range of levels (e.g., tape) the maximum level with analog devices may also depend on what's considered an "acceptable" amount of distortion.

 

[SwitchBack]

And I was hoping that we could leave it at 'working proposition'

The thing is that dB is always relative to something. Even when the 'somethings' for two different dB meters have no relationship at all the meters will still read in dB, which to some suggest that a fixed relationship is there. With dB scales the key question is always: "relative to what?". 0 dBu = 0.7746 Volts, so dBu is relative to Volts. 0 dBFS is relative to 'full scale' whatever digital value that may be. Different entities, with no universal fixed relationship. You may establish one (i.e. calibrate your converter), until you move the gain knob on the interface and it's different again. So the working proposition of 'go for -18dBFS' will do, but the relationship to dBu's is moot

 

[Craig Anderton]

Fair enough. I don’t disagree with any technical points you raise. The purpose of including the “origin story” is so that those who haven’t spent time in analog studios and want to optimize plugin performance will know how to interpret statements like UA saying that “the majority of the UAD plugs are modeled to operate at a nominal level of –18 dBFS,” why the Waves CLA-76, CLA-2, and CLA-3A explicitly document treating -18 dBFS as 0 VU, and why the Blenheim Sound reVUe, Klangheim VUMT Delxue, PSP Audioware TripleMeter, Waves VU meter, and many other reference meterfs have default calibrations where -18 dBFS equals what was 0 VU on analog consoles. I thought it would be helpful if people understood that those numbers weren’t pulled out of the air, but have a historical precedent that bridged the analog and digital worlds. One of the most common questions I get at seminars is “why do companies recommend levels of -18 dBFS when DAWs have such a huge dynamic range?"

 

[SwitchBack]

... One of the most common questions I get at seminars is “why do companies recommend levels of -18 dBFS when DAWs have such a huge dynamic range?"

Excellent points and examples. And my answer to the question would be: "Because the original units these plug-ins are modeled after or inspired by do not have that huge dynamic range, and they work as they do because of it." The -18dBFS was arbitrary at first but it has become a de-facto standard, as has +4dBu in the past. It provides a reference level for the plug-in's noise floor/non-linearity (much lower) and clipping/saturation (significantly higher) to give it its character. Yes, in that sense -18dBFS (plug-in) and +4dBu (analog unit) should produce the same result. But between AD and DA there are no volts so there can't be dBu's either. If you want -18dBFS on the inside to correspond with +4dBu on the outside then you have to calibrate the line inputs and outputs on the interface.

Thinking of it I do wonder what's the exact value of 'full scale' inside the DAW. It can't be the maximum internal floating point value because that's a ridiculously big number. So is it the maximum fixed point value to/from the interface? Then that would make it converter precision dependent (16-bit/24-bit/...). Or did the industry agree on something, e.g. the number +8388608 (2^23 with sign bit)?

 

[Gerran]

Thinking of it I do wonder what's the exact value of 'full scale' inside the DAW. It can't be the maximum internal floating point value because that's a ridiculously big number. So is it the maximum fixed point value to/from the interface? Then that would make it converter precision dependent (16-bit/24-bit/...). Or did the industry agree on something, e.g. the number +8388608 (2^23 with sign bit)?

You got my inner researcher aroused...
Since binary integer representation range is asymmetrical, full scale is defined using the maximum positive value that can be represented. For example, 16-bit PCM audio is centered on the value 0, and can contain values from −32,768 to +32,767. A signal is at full-scale if it reaches from −32,767 to +32,767. (This means that −32,768, the lowest possible value, slightly exceeds full-scale.
According to: Wikipedia

 

[Davey]

It took me quite a while to accept that gain staging is still important in these days of floating point DAWs. I think it was a guitar Fx sim that finally convinced me. You had to hit it with a decent signal to get the right tone out of it.

Shame really. Before that I was completely happy in my "who gives a stuff about levels" bubble.

 

[Craig Anderton]

Thinking of it I do wonder what's the exact value of 'full scale' inside the DAW. It can't be the maximum internal floating point value because that's a ridiculously big number. So is it the maximum fixed point value to/from the interface?

This is a fun discussion I don't have an answer, but I have a theory. I think Full Scale is a leftover from the early days of 16-bit converters (which had more like 12 or 13 "real" bits) and 16-bit audio engines. 40 years ago, digital audio was like tape - you had to go for as high a level as possible to overcome the crappy sound that happened at lower levels (which dithering tried to hide). But the concept of "going into the red" or "as close to zero as possible" had no meaning with digital, so it needed a new name. I would have preferred that we adopted the Spinal Tap nomenclature of referring to the maximum attainable level as "11," but apparently people liked "Full Scale" better

Then evolution happened. We had 20-bit converters that gave 16 real bits, 24-bit audio engines, and 1 Gigabyte hard drives that could hold a CD's worth of material and cost only $2,000! But there were still audio interface limitations, and a 24-bit audio engine still wasn't perfect, so the full scale concept remained valid. However, as resolutions increased, you could leave some headroom because the resolution extended down to lower levels. You could get away with giving up a bit or two at higher levels, and the sound didn't suffer.

In today's world of 32-bit float, I think FS may be kind of like a vestigial organ but it's still in use because, with rare exceptions, audio interfaces don't break the 24-bit barrier. So, there's still a resolution bottleneck going into, and coming out of, the DAW where we need a term that says "don't go into the red because digital distortion really sux" in digitalspeak. At least that's my theory as to why Full Scale is still around.

 

[SwitchBack]

You got my inner researcher aroused...
Since binary integer representation range is asymmetrical, full scale is defined using the maximum positive value that can be represented. For example, 16-bit PCM audio is centered on the value 0, and can contain values from −32,768 to +32,767. A signal is at full-scale if it reaches from −32,767 to +32,767. (This means that −32,768, the lowest possible value, slightly exceeds full-scale.
According to: Wikipedia

Ah thanks, I can happily accept that. I should have thought of that myself, 16-bit PCM setting the standard

 

[SwitchBack]

This is a fun discussion I don't have an answer, but I have a theory. I think Full Scale is a leftover from the early days of 16-bit converters (which had more like 12 or 13 "real" bits) and 16-bit audio engines. 40 years ago, digital audio was like tape - you had to go for as high a level as possible to overcome the crappy sound that happened at lower levels (which dithering tried to hide). But the concept of "going into the red" or "as close to zero as possible" had no meaning with digital, so it needed a new name. I would have preferred that we adopted the Spinal Tap nomenclature of referring to the maximum attainable level as "11," but apparently people liked "Full Scale" better

Then evolution happened. We had 20-bit converters that gave 16 real bits, 24-bit audio engines, and 1 Gigabyte hard drives that could hold a CD's worth of material and cost only $2,000! But there were still audio interface limitations, and a 24-bit audio engine still wasn't perfect, so the full scale concept remained valid. However, as resolutions increased, you could leave some headroom because the resolution extended down to lower levels. You could get away with giving up a bit or two at higher levels, and the sound didn't suffer.

In today's world of 32-bit float, I think FS may be kind of like a vestigial organ but it's still in use because, with rare exceptions, audio interfaces don't break the 24-bit barrier. So, there's still a resolution bottleneck going into, and coming out of, the DAW where we need a term that says "don't go into the red because digital distortion really sux" in digitalspeak. At least that's my theory as to why Full Scale is still around.

Well, as with everything in digital, FS has to have a number on it. And per Gerran's reply I think we can accept that a signal is at full scale when it reaches from −32,767 to +32,767 (peak-to-peak). Floating point then adds the positions behind the 'binary point' to give that range full 24-bit precision (with 32-bit FP) or higher (64-bit FP). There are plug-ins (e.g. 'clean' EQ plug-ins) that happily go far above or below FS but why would you when the next one doesn't.

And your point about converters is absolutely true. Most 24-bit interfaces have converters with far fewer bits inside. Oversampling will add a few but look at the dynamic range (or SNR*) in the specs to learn the actual resolution of an interface. A perfect 24-bit interface will have a 144dB dynamic range (extending from FS downwards). I haven't seen one yet. Nor would I need one

Love the Spinal Tap reference. Classic!

 

[SwitchBack]

Something to add to the fun, dBFS for analog levels (Wikipedia):

• EBU R68 is used in most European countries, specifying +18 dBu at 0 dBFS.
• In Europe, the EBU recommend that −18 dBFS equates to the alignment level.
  • UK broadcasters, alignment level is taken as 0 dBu (PPM 4 or −4 VU)
  • The American SMPTE standard defines −20 dBFS as the alignment level.
• European and UK calibration for Post & Film[clarification needed] is −18 dBFS = 0 VU.
• US installations use +24 dBu for 0 dBFS.
• American and Australian Post: −20 dBFS = 0 VU = +4 dBu.
• In Japan, France, and some other countries, converters may be calibrated for +22 dBu at 0 dBFS.
• BBC specification: −18 dBFS = PPM "4" = 0 dBu
• German ARD and studio, PPM +6 dBu = −10 (−9) dBFS. +16 (+15) dBu = 0 dBFS. No VU.
• Belgium VRT: 0 dB (VRT ref.) = +6 dBu; −9 dBFS = 0 dB (VRT ref.); 0 dBFS = +15 dBu.

Between all those standards and conventions '+4dBu translates to -18dBFS' can be found, and a quite few others as well. End of the day you have to decide yourself on an alignment level and how you want to calibrate your converters.

 

[Craig Anderton]

As the old joke goes, "God must love standards, because He made so many of them." But let's loop back to the original question, which was about headroom in mixes. Given how many companies design their plugins for a nominal -18 dBFS input, that seems like the best choice for someone using Studio One. Also check whether peaks hit between -6 and -12 dB, and if the peaks are excessive, you can choose to dial back the average level.

 

[SwitchBack]

And program content, don't forget about program content. Oh well, if it was simple and easy we wouldn't still be writing about it.

 

[Stuart Welwood]

Great discussion with many useful comments.

As I understand it, you have been talking about the relationship between an amplitude as represented by a number digitally and an analog representation that is rooted in the history of the phone system.

I'd like to take it into the realm of SPL at the listening position. For me, it starts with a CD called "Sound Check" by Alan Parsons and Stephen Court. The first track has a 1kHz Ref. Tone @ -14 dB FS. It is played on a Tascam CD-RW900 Mk II plugged into a McIntosh C42 preamp set to a gain of -15dB. This provides a sound level of about 80 dB.

After ripping the CD on the PC, I dragged the first few tracks into a new S1 song. Here's the first track in solo:

The Studio 1824c has the bottom four meter bars lit. A high resolution digital meter connected to the 1824c's S/PDIF out shows -14 dB. All good.

Using the noise band playing at 85 dB SPL, I matched the CD playback to the SPL level from S1 by adjusting the big Main knob on the 1824c. I can now hear my music at the level that it would be coming off an actual CD.

 

[SwitchBack]

SPL is the final step, calibrating your amp/speaker combo to get a specific listening output level from your line (output) alignment level. Bob Katz wrote the book on that. Do note that a single sine wave puts all energy in one frequency where noise distributes its energy across the frequency spectrum. There's a non-interchangeable use for both.

Cool to see the imported WAV produce exactly -14dBFS, as advertised

 

[Stuart Welwood]

SPL is the final step, calibrating your amp/speaker combo to get a specific listening output level from your line (output) alignment level. Bob Katz wrote the book on that. Do note that a single sine wave puts all energy in one frequency where noise distributes its energy across the frequency spectrum. There's a non-interchangeable use for both.

Good point. I try never to use a sine for any type of measurement in the air. Maybe if I wanted to look for problems.

My purpose for matching the sonic output from the speakers was so that I would be working from a familiar reference. I essentially listen to everything through this system. Since the McIntosh preamp can be precisely set to -15 dB gain and the rest of the settings are never changed, this gives me a repeatable level. I'm trying to give my ears a solidly familiar environment in which to judge sound. I may be there.