This is such a basic question, but I have never really gotten a solid answer to it. I have been mixing in Cubase for a long time, and have found that if a channel goes over 0 db, it never has mattered if the output bus was turned down to compensate. This has lead me to the habit of mixing everything to taste and then pulling down the master fader if things are clipping. I feel like this is probably bad practice, but sonically it seems perfectly fine. Can anyone shed any light on this for me?
Sounds like what you are doing is using the master fader as a master volume rather than avoiding frequency build ups and finding a place for everything.
I used to do that too. Started with the drums and brought them up to where they were great alone but you start adding things and you start to get overs at the master bus.
I’m not sure if this is also true but by pulling down the level at the master you are also losing dynamic range by reducing the your bit depth because everything is loud is now hitting within the top 30 percent of the range.
I have switched to starting with the track faders lower than I would have before so the meters are rarely anywhere close to 0db and increased the volume on my monitors to compensate. It seems to have opened up my mixes and they are easier to do as well.
I did this after steve lukather mentioned when they record things and the meters are hitting right in the center rather than the top things seem to mix themselves.
Then again I could be completely wrong but its what I do now.
Hi hikarateboy! I am, indeed using it as a master volume. 
I know on an analog console going to a physical medium (tape) this would be a total no no, but it seems like a digital daw is different. I guess that the thing is that I don’t really feel like I hear any artifacts problems with it, per se. It seems to me that if one has a mix that is peaking at -.1db on the mix bus that it is using the full bit depth available, so the fader level should be irrelevant if it is at 0 or -5. I don’t know if that is correct at all, though. I guess that I wonder if a mixer channel peaking at +4 db with the master fader turned tdown by 4 db is the same as a channel peaking at 0db with the master set at 0db.
Hmmm, well I wonder if the lack of responses means that a lot less people know the answer to this than I might have guessed. Surely someone out there knows the actual mechanics of headroom inside Cubase?
Surely, if read in the forums a bit, you may find there are a lot of people who do and who posted their knowledge about this several times already…
My pennysworth, based on what I have been advised by better engineers than me and my own experience.
Moving your master fader is cheating.
- Start a mix so that the kit is around the -6db mark, which will give you the headroom to work with.
- If you are getting sudden peaks this is probably to do with two things fighting for the same frequency band, which you must stop happening as all it does is take up headroom without contributing to the sound. E.g. in the rhythm section your kick, snare and bass need to be working in separate spaces.
- It is often (even usually) better to cut than to boost.
- Don’t use eq and compression unless you have to.
I think headroom is less critical than it used to be as there just isn’t the noise floor you had with tape but I don’t think that’s any excuse for getting sloppy. A good set of keywords to start with in your search would be “gain structure”, a subject I am not qualified to talk about, I’m afraid.
Yeah, i’s been covered over and over but it defies final clarification.
??? Inwhich way it’s “cheating”? If it wasn’t ment to be used, it would’n be there in the first place, right? In my opinion it’s much less “cheating” than for example multitrack recording or close-micing. If you don’t want to “cheat”, you better just use stereo microphone and cut directly to vinyl 
Back to the original topic. In floating-point DAW like Cubase levels do matter only in:
- Input
- Output
- File being recorded (in case of using fixed-point file format)
- Some plugins, which have “sweet spot”. This includes “vintage gear emulatiors” and compressors/limiters, which have limited treshold setting range.
In any other places in your DAW’s virtual signal path you have 100s of dBs of dynamic range/headroom.
I think headroom is less critical than it used to be as there just isn’t the noise floor you had with tape but I don’t think that’s any excuse for getting sloppy. A good set of keywords to start with in your search would be “gain structure”, a subject I am not qualified to talk about, I’m afraid.
“Gain structuring” or “Gain staging” indeed was/is essential with analog equipment (and older versions of unnamed “industry standard” DAW). But with Cubase there’s no reason why you couldn’t “get sloppy”.
Internally Cubase uses 32-bit Float. The mantissa is 24 bits, which is what the meter is based on. So when a channel goes over 0 dB it means that the mantissa was exceeded and you will lose some resolution, but there will not be hard clipping on that channel.
However you still need to be careful. If you export that channel to 16 or 24 bit fixed the result will be clipped. I believe you could also run into issues with certain plugins clipping (i.e. 32-bit float in converted to 24-bit fixed at input of the plugin could cause clipping).
Ron
First off, thank you all for your input on this. It would seem that, unless I am reading one of them incorrectly, Ron and Jarno are slightly at odds in that Ron suggests that there is a potential loss of resolution for a channel, (unless he means that this only applies to directly exporting that track or leaving the master fader at 0 and exporting.) It seems that Jarno is indicating that a channel can go much hotter than 0 db as long as the actual output is pulled back down to compensate (which it kind of seems like to me.)
I also kind of agree with Jarno about the “cheating” idea. There is a big discrepancy between old school recording techniques necessitated by older technology and what is necessary with new technology. I think that people often get confused about why certain things were done in the past. I certainly understand the sentiment in this case becasue many novice mixers suffer from “fader creep” as a result of arguing channels, but if one has a mix that is tight and balanced but it is hitting the main bus 2db over 0 and you dont’ wish to strap a mix bus limiter across the final mix, I hear little problem with pulling down the master 2db. I don’t see how this would be any different sonically than grabbing all the faders as a group and notching them down until you are 2 db quieter. I really would like to know if I am wrong about this, though, because at these kind of level shifts, potential degradation could be slight enough that I wouldn’t notice unless I did an A/B comparison between the 2 approaches. I could be doing something less desirable to my mixes and not even knowing it.
Please understand, I am not stating that I am right. I just am stating what my ears seem to communicate to me. I appreciate Ron mentioning “mantissa”: a term I have never heard. What exactly is that in relation to Cubase? (I just googled it and the result was greek to this old english major. 
I believe Jarno and I are saying the same thing. You CAN exceed the zero dB mark and not clip (on a channel). But there is a slight loss in resolution that you will likely never hear. You WILL get clipping if you try to directly export that 32-bit floating channel to 16 or 24 bit fixed.
Take a look at this
and scroll down to the diagram that says 32-bit ploating point data.
where is says “fraction”, that is what I was refering to as the mantissa. It is 23-bits + there is 1 sign bit = 24-bits. So if the exponent is always zero, you have 24-bits for your audio data (equivalent to 24-bits fixed from a dynamic range point of view). If you exceed the 24-bits, then you need to have an exponent. Here is a simplictic example. I’m going to ignore the sign bit but I think the point will get across. Assume you have a floating point representation in which the mantissa is only 3 bits, but you have 2 bits for an exponent.
Exponent
Mantissa 2^0 2^1 2^2 2^3
000 0 0 0 0
001 1 2 4 8
010 2 4 8 16
011 3 6 12 24
100 4 8 16 32
101 5 10 20 40
110 6 12 24 48
111 7 14 28 56
If this were a 3 bit fixed sysem, it would clip above 7. but in a float system, it would go to 8, but there is no 9. The next value available is 10. So we have lost some resolution. In example the sample values are:
0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28 etc.
Ron
Spot on.
Nice one! (and true)
As long as you don´t have any plugins in the master channel that react to level thresholds or that nare not 32 Bit float…
You can simply try yourself:
-Take an audio file - duplicate it,
-add two groups (they simulate your Master output bus)
-route one track to group one, the other to group two.
-Raise level on audio track 1 by 10dB (for example with event volume) lower the group fader of this audio track´s group by the same amount
-Leave the other as is
-switch polarity on one of the two groups
-If both signals are identical after the groups, they will cancel, and you get -inf dB at the (real) master out.
(-If you don´t get -inf at the master out, you probably did something wrong)
Thanks Ron! Well, I think a small piece of my brain just melted out of my ear after reading your post, but it generally made sense. I don’t have to read binary on a regular basis. Anyway, I appreciate your taking the time. A couple of thoughts:
- So if I use Cubase 64, is the mantissa 48?
and…
- I just did as thinkingcap suggested. perfect null!! The boosted track was peaking at +24 db. (I used a gain plugin) I think you could take a non clipped audio file and boost it by 50db and pull the master bus down 50db and have 0 loss of resolution. The fact that it worked, though seems to negate the idea of any kind of loss of resolution whatsoever. Is there a flaw in this experiment that thinkingcap and I are missing? (Thanks thinkingcap for the great test idea! Not sure why it didn’t occur to me to try.)
No. 32-bit float is used internally for both 32-bit and 64-bit. It has more to do with thge addressing being 64-bit vs 32-bit.
I’ll need to look into this one.
Ron
You’ll likely need a more accurate meter to check for a true null and Cubase doesn’t include any metering plugs for some odd reason that I’ve yet to put my finger on.
Cubase’s master meter is not measuring down to the floor of the entire scale (afaik, few, if any, do), more likely it’s metering down to -70 or something. If you can set it to -144 or something (dunno) you might see remnants, or not. In either case it’s irrelevant because you can’t hear it anyway and it’s far below the noise floor of any analog part of your system.
If you doubt the ability of your DAW to meter something that’s actually there, insert a dither plug somewhere and look at what the meter shows. Usually nothing… unless you can lower the floor of the meter to measure a much lower point. These things cause random debates because various testers often don’t actually know what they’re looking at, in context. A meter null isn’t an actual technical perfect null unless the meter is metering the entire scale. It’s a “practical” null because you can’t hear anything. Anyway…
Bottom line… it’s all 100% irrelevant (and way below what you can hear anyway) in floating point… for everyone but Clark Kent.
At any rate, if you do decide to get really anal and test some of that stuff it’s probably a good idea to measure the tools that you’re using to measure the other stuff, so you know exactly what you’re looking at. Measuring where Cubase’s master meter stops metering would be a good start. As an example, one of my hosts master meter measures down to -60, and then stops metering … which means it cannot prove a perfect null… only that no remnants (if any) exceed -60.
But it does have metering plugs whose floor can be set to -144.
Did this ages ago, but peaked close to +100dBfs and also threw in Cubase’s EQ processing boosted and unboosted signal in parallel. Result? Not null, but the difference signal was way below -120 dBfs. Good enough for me.
Exactly. Rounding errors in 32-bit floating-point audio happen at 144dB below current instant signal level. Completely inaudible. Even for those who can hear the difference between $5 000 and $50 000 speaker cable.
Great info here guys!! I Googled some about gain structuring and I’ve had a headache ever since.
I bet you did! By googling you’ll find “the truth” found in context of unnamed “industry standard” DAW. But hey … it must be The Whole Truth And Nothing But The Truth, because that’s The Industry Standard
And for a reference: I mix in 32-bit fixed-point domain (TASCAM DM-4800) and never have to worry about gain-staging, because I have 24dB headroom above 0dBfs (yes, that’s a huge headroom). Now … think about 32-bit floating-point engine with 1500dB dynamic range … 1500dB? … can you even think about it? It’s something like comparing sound of the air molecule hitting your desktop to the atomic bomb blasting off at the same place.
if you’re using analogue emulation plugins then you still need to use proper gain staging.
MC