dBVU, dBFS, noise, and YOU! (correct mixer gain)

[tarnationsauce2]

It’s a common misunderstanding that it is best to turn the mixer up until your DAW (soundcard) is recording waves that are close to peaking to the top of the waveform being recorded. THIS IS A BAD PRACTICE.
This comes up a lot around here, and I may be beating a dead horse for some of you. But hopefully some of you will find this informative.

First thing’s first:
dBVU is what an anolog mixer uses for reference, 0dBVU would be where the mixer works at it’s optimum. Noise is the least, while having enough output to drive the sound card.
dBFS is what a digital sound card uses for reference. 0dBFS is where the sound card will reach it’s maximum and if pushed beyond that it will badly distort. There is no “optimum” level in the digital realm per se, as long as you are above 16-bits of usage.

Key words here: 0dBVU = -18dBFS

What does that means to us? Well if a mixer works at it’s optimum at 0dBVU, then that means that your waves when recording should peak at approx -18dBFS.

But wait!! That means you are losing resolution because you’re not using all 24 or 32 bits for your audio!
Well, you see, 16-bit is pretty much the standard of what is acceptable for hi-fi sound in that the human ear cannot detect quantization errors.

A full 16-bit waveform has 65,536 samples or “starsteps”.
A full 24-bit waveform has 16,777,216 samples or “stairsteps”.
A 24-bit waveform at -18dBFS (or 0dBVU) obviously has more than enough (more than 16-bits), exact number I don’t know (can someone help with this?).

At first I really thought that was all bullshit. But it’s not.

I recently took some data.

These are noise floor values from my mixer (lower peak amplitude is better).

I used a 1kHz sine wave to set gain, then disabled it for the noise floor test.
========================
Mixer set so that my sound card was peaking close to 0dBFS. (wave file close to peaking)
========================
Left Right
Min Sample Value: -4.71 -4.79
Max Sample Value: 4.84 4.94
Peak Amplitude: -76.61 dB -76.43 dB
Possibly Clipped: 0 0
DC Offset: 0 0
Minimum RMS Power: -87.89 dB -88.12 dB
Maximum RMS Power: -87.51 dB -87.72 dB
Average RMS Power: -87.73 dB -87.89 dB
Total RMS Power: -87.73 dB -87.89 dB
Actual Bit Depth: 24 Bits 24 Bits

Using RMS Window of 50 ms

========================
Mixer set so that my sound card was peaking close to -18dBFS. (wave file at -18dBFS, pretty small looking)
========================
Left Right
Min Sample Value: -1.78 -1.8
Max Sample Value: 1.69 1.77
Peak Amplitude: -85.29 dB -85.18 dB
Possibly Clipped: 0 0
DC Offset: 0 0
Minimum RMS Power: -98.07 dB -98.48 dB
Maximum RMS Power: -97.42 dB -97.72 dB
Average RMS Power: -97.74 dB -98.12 dB
Total RMS Power: -97.74 dB -98.11 dB
Actual Bit Depth: 24 Bits 24 Bits

Using RMS Window of 50 ms

========================
Mixer set with gain/faders all the way down. (just for reference)
========================
Left Right
Min Sample Value: -.59 -.79
Max Sample Value: .56 .69
Peak Amplitude: -94.89 dB -92.35 dB
Possibly Clipped: 0 0
DC Offset: 0 0
Minimum RMS Power: -108.91 dB -107.01 dB
Maximum RMS Power: -107.62 dB -105.87 dB
Average RMS Power: -108.41 dB -106.56 dB
Total RMS Power: -108.41 dB -106.56 dB
Actual Bit Depth: 24 Bits 24 Bits

Using RMS Window of 50 ms
========================


Wow!
As you can see setting the mixer to 0dBVU / DAW to -18dBFS gave us a whopping 8.75 dB peak / 10.36dB RMS Less noise. A lot less.

 

[mshilarious]

Wow!
As you can see setting the mixer to 0dBVU / DAW to -18dBFS gave us a whopping 8.75 dB peak / 10.36dB RMS Less noise. A lot less.

Not that I disagree with your premise, but . . .

What happens to the noise floor when you mix the tracks? Let's say that you are only recording a single stereo track. Since you aren't going to release a CD with peaks at -18dBFS, when you increase the gain 18dB, you would have a higher noise floor than your 0dBFS example.

In fact, if you always work with peaks below 0VU, then your signal never rises above 1V. Any decent piece of gear should be able to crank out a few more volts than that, whereas the noise floor will be somewhat more fixed.

Surely there is a happy medium . . . in fact I think you have proven that with your gear the best level to work at is probably peak at -9dBFS or so. I would be interested to see normalized noise @ 0 and -9, and THD at the same levels.

Also, 0VU does not always equal -18dBFS. In my case, it's either -9 or -15dBFS, depending on the button I press on my converter. There is really no substitute for reading the converter's manual

 

[blueroommusic]

yeah...what he said.

although it is not a good idea to get "HOT" levels into your comp, it is a good idea to get good levels...-18 is too low, me thinks.

 

[SouthSIDE Glen]

Key words here: 0dBVU = -18dBFS

What does that means to us? Well if a mixer works at it’s optimum at 0dBVU, then that means that your waves when recording should peak at approx -18dBFS.
...
As you can see setting the mixer to 0dBVU / DAW to -18dBFS gave us a whopping 8.75 dB peak / 10.36dB RMS Less noise. A lot less.

This is a very difficult subject to broach, and you have made a good inroads into it with a very good post here. I'd just like to refine and comment on a couple of things you said above:

First, there is no actual coversion formula that says 0dBVU = -18dBFS, in fact there is no relation between dBVU and dBFS whatsoever. It's not like converting Celsius to Farenheit. they are entirely independant scales each designed for it's own purpose.

That said, there is an approximate consesus in the industry that a signal that measures 0dBVU at the input of an A/D converter (in reality meaning only that the signal at the A/D input is at the full rated line-in signal voltage) should come out the other end of the converter with somewhere in the -9dBFS to -18dBFS range. Typically most converters are calibrated somewhere in the -14 to -18 range, but there is that range of variance.

You were right in the second sentence to say "approx. -18dBFS" because the 18dBFS is not a hard and fast rule. It would be more correct to say "0dBVU should convert to somewhere around -14 to -18dBFS, give or take a couple of dB, depending on the calibration of the outputs of the analog device and of your A/D converters."

There are also other practical reasons than noise floor for having headroom in th "dBVU-to-dBFS conversion".

Not the least of these is the well known propensity for humans to push analog signal levels past 0dBVU, whether it be for tape saturation purposes, pushing the S/N ratio, or purposely saturating a preamp to get "that sound". Without the extra headroom when converting to the digital realm, analog overs would have no place to go.

I could keep going, but I'll leave some pizza on the plate for others .

Nice report, though.

G.

 

[tarnationsauce2]

What happens to the noise floor when you mix the tracks? Let's say that you are only recording a single stereo track. Since you aren't going to release a CD with peaks at -18dBFS, when you increase the gain 18dB, you would have a higher noise floor than your 0dBFS example.

Yeah I was thinking about that too... Hmm...
Perhaps I'll do some more noise floor measurements and find where the s/n ratio is best. It may actually be above 0dBVU.

This is a very difficult subject to broach, and you have made a good inroads into it with a very good post here. I'd just like to refine and comment on a couple of things you said above:

First, there is no actual coversion formula that says 0dBVU = -18dBFS, in fact there is no relation between dBVU and dBFS whatsoever. It's not like converting Celsius to Farenheit. they are entirely independant scales each designed for it's own purpose.

That said, there is an approximate consesus in the industry that a signal that measures 0dBVU at the input of an A/D converter (in reality meaning only that the signal at the A/D input is at the full rated line-in signal voltage) should come out the other end of the converter with somewhere in the -9dBFS to -18dBFS range. Typically most converters are calibrated somewhere in the -14 to -18 range, but there is that range of variance.

You were right in the second sentence to say "approx. -18dBFS" because the 18dBFS is not a hard and fast rule. It would be more correct to say "0dBVU should convert to somewhere around -14 to -18dBFS, give or take a couple of dB, depending on the calibration of the outputs of the analog device and of your A/D converters."

Yeah I knew that the -18dBFS isn't a set in stone rule. I should have mentioned that. But from what I've gathered -18dBVU is a good estimated place to start.

All good points.

 

[xstatic]

The general theory of this thread certainly holds true in my opinion. Like mentioned above, -18 is not a law like 0dbvu is. Also, analog was nt designed for 0dbvu to be your peak value either as far as I know. I always thought and was taught that it is designed to be your average level. There is a big difference here. with 0 dbvu as you peak level on a track, that could leave your average level around -10 dbvu for many different types of tracks. That is waaaay too low in my opinion. if however you raise things so that you average level is 0dbvu, than based on an average -18dbfs level, your new track would peak at -8 dbfs, or at +10 dbvu. Most analaog equipment is set up to go up to +22 or so without really audible distrotion (within spec). This is 18 db of output above the average +4 level. Basically, to play it safe, set your average level of a track to somewhere around 0 dbvu, and with most all equipment you will have good levels as well as good headroom.

 

[Mixxit12]

Feel free to keep going Glen. This has been a heavily debated topic around these parts.

I am also interested in how bringing your levels close to 0 DBFS affects the signal after you mixdown.

Thanks for all of the insight everyone!

Edited to say DBFS (sorry)

 

[Farview]

I am also interested in how bringing your levels close to 0 affects the signal after you mixdown.

Which 0? VU ro FS?

 

[Mixxit12]

FS. Sorry. I edited my earlier post.

 

[RAK]

Also, analog was nt designed for 0dbvu to be your peak value either as far as I know. I always thought and was taught that it is designed to be your average level.

VU (Volume Unit) meters show an average (RMS) voltage that is more representative of how the human ear hears.

I was a little confused why everyone is using dBVU. Technically speaking the correct abbreviation is dBu (or sometimes dBv). But I know we all know what the meaning was, I just like to get hung up on details

 

[Farview]

If you record close to 0dbfs, you are running your analog signal chain too hot. This will result in a pinched and possibly distorted sound. Add that together with a bunch of other tracks with the same small pinched sound, you end up compounding the problem.

If you are talking about taking a mix and bringing it up to 0dbfs in mastering, that's just what everyone does. Most ME's don't seem to like it, but the clients demand it.

 

[tarnationsauce2]

OK, so I set out to see exactly how much more noise is actually introduced by pushing the board higher than 0dBVU.

Here is my method:

I played a 1kHz tone into the board. Started with 0, +2, +4, +7, +10 dBVU on my board.

Here is a pic of what my waves looked like. You can see where I killed the incoming signal, and where I adjusted the board to the next dBVU setting. (NOTE: I completely unplugged the incoming signal so the sine wave generator would not introduce any self noise).

Then I normalized all the separate dBVU settings as well as the noise floor at the end of the corresponding waveform.

There is what it looks like now:

Next I collected amplitude data of all the noisefloors.

(lower is better)

0dBVU = -12.0dBFS
Peak Amplitude: -81.36 dB Noisefloor

+2dBVU = -9.6dBFS
Peak Amplitude: -83.56 dB Noisefloor

+4dBVU = -7.9dBFS
Peak Amplitude: -85.57 dB Noisefloor

+7dBVU = -4.9dBFS
Peak Amplitude: -87.79 dB Noisefloor

+10dBVU = -0.85dBFS
Peak Amplitude: -88.78 dB Noisefloor


HMMM….
This basically says everthing I wrote in the initial post is completely false (at least for my board and interface).

Can someone else take the time to do this test as well to see if they have the same (or similar) results?

In essence, it really is best to get as high to 0dBFS without clipping.

Hows that for full circle… twice. lol

 

[Farview]

Gain staging isn't only about signal to noise ratios, it's about running all the equipment in it's comfort zone. All of your analog equipment was designed to work most efficiently at line level. That is where all the specs were measured. That is the level that the designers intended for it to be run.

 

[mshilarious]

Can someone else take the time to do this test as well to see if they have the same (or similar) results?

No, you aren't done yet! Nobody ever said that you wouldn't maximize dynamic range by recording hot. That will be true of all but crap gear. However, there are two trade-offs: no headroom left for unexpected peaks. That's a good reason to turn it down, but not relevant in a controlled test.

The other reason (oft stated by Massive) is distortion caused by running your gear hot. This you haven't tested yet.

Go back to your normalized 1kHz test tones, and plot them on an FFT. See what you find for distortion (you should be able to see the noise floors there too).

 

[tarnationsauce2]

Gain staging isn't only about signal to noise ratios, it's about running all the equipment in it's comfort zone. All of your analog equipment was designed to work most efficiently at line level. That is where all the specs were measured. That is the level that the designers intended for it to be run.

I know. Gah! I really expected the noise floor to get worse.
I like measurable factual data.

Of course I'll trust decades of electrical engineering, and counless audio engineers.

I am going to see if I can measure THD accurately with Adobe Audition.

 

[mshilarious]

I know. Gah! I really expected the noise floor to get worse.
I like measurable factual data.

Of course I'll trust decades of electrical engineering, and counless audio engineers.

I am going to see if I can measure THD accurately with Adobe Audition.

If you can't, post your .mp3 & I'll do it for you

 

[Massive Master]

I think we're losing sight of the big picture here -

It really doesn't matter how you have your converters calibrated (in the big picture) if you're using your *front end* as it is designed to run. The converters, as important as they are, aren't creating the sound - They're supposed to accurately represent the sound digitally.

The analog chain - A preamp, outboard gear, a compressor, EQ, etc. - Those are the levels you really need to worry about. A good converter will faithfully digitize the signal coming out of a preamp. If that signal is clean and open and focused - or pinched and veiled and overdriven - is the issue here.

 

[SouthSIDE Glen]

Also, analog was nt designed for 0dbvu to be your peak value either as far as I know. I always thought and was taught that it is designed to be your average level.
...
Most analaog equipment is set up to go up to +22 or so without really audible distrotion (within spec). This is 18 db of output above the average +4 level. Basically, to play it safe, set your average level of a track to somewhere around 0 dbvu, and with most all equipment you will have good levels as well as good headroom.

The whole dBVU topic is really quite muddled. I think it's rather like quantum theory; there's about 7 people on this planet that *fully* understand it, and 6 of them are probably lying .

Nor do I claim to have a 100% understanding of every aspect of it (I am not one of the 7 ). A big part of that misunderstaing, I believe, is because the hardware manufacturers have played fast and loose with VU metering over the past half century or so. Just what the VU meters are calibrated to depends greatly on the purpose of their host machine and the whims of the designers. The theoretical/lab reference to 1v does not necessarily apply

0VU on a radio transmitter, 0VU on a tape recorder, and 0VU on a mixer channel strip do not necessarily all mean the exact same thing. On a radio transmitter, 0VU is usually calibrated to represent 100% modulation of the carrier frequency. On a analog tape recorder, 0VU can potentially refer to a couple of different things; typically either it's calibrated to equal a magnetic field on the tape head equal to the number of nanoWebers for which the reference tape for that deck is rated, or it can refer to either an input or an output signal equivalent to the rated line level I/O for that machine (typically either +4 or -10dBu.) On a mixer channel strip they can be measuring signal pre-fader or post-fader (depending upon the mixer and the settings), etc.

Add in the fact that there are "averaging VU meters" and "peak level VU meters", neither of which necessarily reflects a *true* average or peak, and that two different models of meters with obstensibly the same electrical specs and calibration method will have different physical "ballistics" (how fast and accurate the needle responds to the voltage in it's driver coil), and you have VU meter reasings from machine to machine that can be all over the map.

I prefer to keep it simple and avoid the headaches and follow the general rules that one should typically ride somewhere between -1 and +3 on a peak reading analog VU meter, depending upon the particular characters of the device and it's meters, which really need to be test-driven and learned first, and that after converting to the digital realm, my gain staging is close enough if that close-to-0dBVU peak is registering somewhere between -9 and -18dBFS (again depending upon how hard I'm pusing the analog and on the nature of my converters at the time.)

G.

 

[RAK]

The analog chain - A preamp, outboard gear, a compressor, EQ, etc. - Those are the levels you really need to worry about. A good converter will faithfully digitize the signal coming out of a preamp. If that signal is clean and open and focused - or pinched and veiled and overdriven - is the issue here.

On the other hand, digital clipping: bad, analog clipping: not necessarily as bad. (I know this is over simplifying). But you could look at in terms of needing to pay more attention to the digital signal in terms of clipping.

Gain staging isn't only about signal to noise ratios, it's about running all the equipment in it's comfort zone. All of your analog equipment was designed to work most efficiently at line level. That is where all the specs were measured. That is the level that the designers intended for it to be run.

Sounds good to me.

 

[apl]

It seems that VU has a time factor built into it, hence the term VU ballistics. dBFS has no time factor, it is an absolute limit of your recording chain.

It also seems that if you mixed together two signals that were both 0 dBFS by setting the faders to 0 on each channel and the master to 0, the result would be +3 dBFS and would sound poopy. However, your software can handle the interim overage and if you turn down the master fader to -3 dB it should be OK.