The -18 dBFS rule -- myth or reality?

[clepsydrae]

Here also is Jim Williams from Audio Upgrades with some strong words on the subject (from the tapeop thread I started in 2012):

Originally Posted by Jim Williams

Contrary to previous statements here, well designed analog transistor based audio gear doesn't perform best at zero dbu. It does best at a higher level, the THD specs show that. That also helps lift the signal further from the residual noise floor giving a better S/N ratio.

An examination of a THD vs amplitude sweep on a decent analyzer shows decreasing THD with increasing levels, all the way up to the clipping point. My analog console does .0005% THD+noise at +27 dbu, it will degrade at lower levels mostly due to noise inclusion.

The only benefit of operating an analog mixing system at lower levels is you will get more noise.

Don't know how to square that with the PreSonus THD+N spec which seems to indicate that THD+N is <0.008 at +4dBu and .5 at +14dBu.

Oh wait, yes I do, he addresses that. :-)

Originally Posted by Jim Williams

Exceptions are the rule. Most modern opamp based audio gear shows lowering THD with increasing levels. Some don't follow that rule, mostly due to sloppy design. If the opamps do this and then the circuit doesn't, then that's in the domain of the designer. One example is the Toft ATB console. It does about .05% at +4, then goes up to 1/2% at +20 dbu, not very good and users complain about it.

Also:

Originally Posted by Jim Williams

With a wide dynamic operating area, levels are not important anymore. Only the residual background noise will vary. Here I hear no difference in +4 or +12 levels, but any noise is reduced at those higher operating levels.

[Reno.thestraws]

Sorry for bad english...

At first, there's an important thing to understand :
The difference between PEAK and RMS level.

The RMS level is something that try to display how loud the sound Feel. It's an average level based on X msec (mostly between 300 and 500).
The PEAK level is displayin how loud the sound IS at a "precise" moment.

Check out my video about recording levels

http://youtu.be/-SuWj023b2E

if you go at 18 min, I set up an example where I play a white noise through a VU meter

What will you see?

the VU dancing around 0VU

Then I add just a tiny crackle in this white noise

What will you see?

The VU still dancing around 0VU

But if you look at the PEAK meter of REAPER, it clips!

What does it mean?

that means that the little crackle was too short to by interpreted by the VU meter that calculate a average level and not a PEAK LEVEL

Now let's forget about digital and DAWs for a couple of minute and let's go back to the 70's where engineer recroded stuff on tape.

When the engineer recorded stuff, they only looked AT ONE METER, the VU meter.

They didn't care about the peak value of their material. Why? because they knew that the tapes could hadle it.

So when they were recording a snare drum, they turn up the preamp until the VU needle reach 0VU (1.23Volt = +4dbu)

THEY Knew that they were peak above that level but they didn't care because the natural compression of the tape took care of it.


Now let's back In our digital

If we record the same snare drum with the same preamp at 0VU plug in a ADC instead of tape machine, what it will happened?

depending on the alignement of the ADC, the 0VU (So the RMS level) will reach a level between -21 and -10 dbfs.

So no problem in appareance BUT, In digital recording, the peaks are not compressed AT ALL, they go straigh to the ADC and with instrument with high crest factor (Dynamic range, aka difference between peak and RMS level) such as snare, those peak can clip.

That means that you need to care about a second level now, the peak one.

SO, we can think that when record to digital, we just have to care about peak level.

yes but now.

It depend of the crest factor. If you're recording a snare drum (high crest factor), you'll know that your peak will be 14 or 16 db above the RMS level, so IF you're not clipping, you'll be 99% sure that your RMS level will be around the 0VU

If you're recording a SYNTH PAD now with a dynamic range of 4db, what will happen?

if your peak levels is at 0dbfs, that means tha your RMS level (VU) will be at -4dbfs but wait...

if my ADC is align at -18dbfs = 0VU = +4dbu = 1.23 volt, a level of -4dbfs RMS means that your preamp is 14db over the sweet spot. In short, it distord like crazy, in other term, the VU meter on your preamp will be at +14VU (the needle stay at max on the right)


If you still follow me, that means that in digital domain you have to check two levels

1) do not clip the peak value
2) make sure you're not overload the analog component that are placed BEFORE THE ADC). How can I know the this?
it's simple, open the user manual of your converter and search for the reference level.

exemple on a firestudio project

Reference Level for 0dBFS +10 dBu

that means that +4dbu = 0VU = -6dbfs

on a RME fireface UCX
Reference = Lo gain
0dbfs = +19 dBu
Headroom = 15 db

That means that + 4dbu = 0Vu = -15dbfs


So, with the presonus firestudio project, you nedd to be sure to don't exceed -6dbfs RMS to not overload the analog component on the way in

with the fireface, you can't exceed - 15 dbfs RMS to not overload your preamp!





Let's go back to our snare drum

If I record the snare drum that have a crest factor of 16 db with the presnonus? what it will happen?

My peak leavel is at 0dbfs, so my peak level is at -16 dbFS

since the presonus is align to -6dbfs, that means that my preamp is at -10 VU... is this a problem? no but you loose 10 db of SNR in your preamp

now let's record with the fireface

peak : 0dbfs
RMS : -16
alignement = -15
Preamp = -1VU...

in addition to that, there's the 24 bit

24 bit can record a dynamix range of 144 db (from -144 to 0dbfs)

THERE'S NO REAL LIFE SITUATION when you'll need such an dynamic range.

So why don't let a little headroom?

If you're recording at -18dbfs RMS that means that you'll never overload your preamp and you'll probably never clip (exept with signal that have a crest factor greater than 18db that is very rare)

Again, sorry for bad english, I'd prefer to explain all these technical stuff in french

[clepsydrae]

Merci pour la réponse -- dis moi si quelque chose ici n'est pas clair et je peux essayer de traduire en français -- désolé - ça fait longtemps que je n'ai pas pratiqué.

Originally Posted by Reno.thestraws

2) make sure you're not overload the analog component that are placed BEFORE THE ADC).

But the question being asked over and over is: does "overloading", by which I understand you to mean recording hot but not clipping, actually matter? No one has demonstrated that, that I can find. Jim Williams and RME themselves have said it's a myth (quoted just above). Sound on Sound effectively said it's not true, 14 years ago (quoted just above).

Can you please record a snare or whatever sound you think will most clearly demonstrate exhibit this "overloading" by taking one signal, sending to two inputs with one at -18 dBFS RMS average levels and one peaking as close to 0 dBFS as you can get without clipping, normalize them, and post the demonstration files so we can see if we can tell any difference?

If you don't think that test will produce an audible difference, what would be necessary to hear an audible difference due to driving the input stage of an ADC close to its max?

How can I know the this?
it's simple, open the user manual of your converter and search for the reference level.

The reference level has to do with aligning conveniently with the analog world, and, if you are one of the people who find that recording at lower levels is good for your workflow, with obtaining lower levels in your DAW. As far as I know, it's not about obtaining optimal sound quality, and no manual I yet know of makes that claim.

RME says that you can't degrade the sound of their interfaces by driving the input close to its max, and several manuals referenced in that other thread also recommend recording hotter than the nominal level.

24 bit can record a dynamix range of 144 db (from -144 to 0dbfs)

THERE'S NO REAL LIFE SITUATION when you'll need such an dynamic range.

So why don't let a little headroom?

- Because recording at lower levels on the way in introduces more noise in the analog stage before the 24 bit dynamic range is even relevant. That's not important to me, as the added noise going in to my device at -18 dBFS RMS average is quiet, but most people, espcially those that bother calibrating the voltage levels of their gear, do care about adding noise for no good reason.

- Also because Jim Williams says that THD is worse for most analog input stages at lower levels, and better the higher you go. (Also quoted just above.) I have no idea if this is actually a relevant amount of distortion. I'm certainly not going to advocate against -18 dBFS RMS average input level because of this potential. But it's a possible reason someone might have.

THD+N for my particular device is apparently worse at higher levels, according to spec, but apparently not too much worse, because the test clips I posted and seek responses to (so far, none...) seem to me to be indistinguishable from each other.

[Reno.thestraws]

But the question being asked over and over is: does "overloading", by which I understand you to mean recording hot but not clipping, actually matter? No one has demonstrated that, that I can find. Jim Williams and RME themselves have said it's a myth (quoted just above). Sound on Sound effectively said it's not true, 14 years ago (quoted just above).

It's impossible do make general rules. It depends on the quality of the preamp. If RME preamp don't suffer to be pushed hard, is it always true with an art pro channel or a berhinger?

The -18 dbfs rule is just a guide line for home studist that generally don't understand NOThINg about gain staging and have low quality pReamp. If you know what you're doing, ok, make things like you want but you can't advice on a public forum to record at -0,1 dbfs peak.

And it's also a matter of confort and ease of doing things?

Do you ever try to record a live drummer with peak levels peaking at the top of the meter? You always habe to check your meter and compensate the preamp gain to not clip!

If you let a headroom, you can forget the meters and focus on music


Just a last words about RME. Did you ever try the autogain feature of the brand new fireface ucx or ufx? When it's on, the preamp automatically adjust to have peak at -6 dbfs peak

Why the hell do they let 6 db of headroom if it's not useful?

Because, when the reference is 4dbu, the 0VU is at -9dbfs. So by adjusting the preamp levels to not exceed -6 dbfs, there's 99% chance to never overload the preamp and to never clip. (exeption if you're recording signal with less than 3 db of DR, which is near impossible in a real life situation)

[clepsydrae]

Originally Posted by Reno.thestraws

It's impossible do make general rules.

Not trying to be snarky, but didn't you just make two general rules about this issue?:

Originally Posted by Reno.thestraws

1) do not clip the peak value
2) make sure you're not overload the analog component that are placed BEFORE THE ADC

The first one I agree with.

As for the second, lots of people say that recording at -18 dBFS RMS average levels is an important thing to do for the sake of audio quality -- it is this general rule that I will object to, if it is wrong. It sounds like you were saying basically the same thing: "in short, it will distort like crazy".

The info from Jim Williams and Sound on Sound and others implies that you can in fact make a different general rule: "for the majority of modern ADC equipment, recording hotter is a bit better when you're only talking about audio quality of the recorded signal."

It depends on the quality of the preamp. If RME preamp don't suffer to be pushed hard, is it always true with an art pro channel or a berhinger?

Hey, great question, sounds familiar. :-) Maybe someone will do some tests and post examples. :-)

I doubt that it matters, personally, but I'm just trying to learn more.

If you know what you're doing, ok, make things like you want but you can't advice on a public forum to record at -0,1 dbfs peak.

If someone cares a lot about noise and THD, and if they aren't worried about calibrated level matching with analog outboard gear, then, based on what I've learned so far, I will indeed advise them to record as hot as possible without clipping.

Why the hell do they let 6 db of headroom if it's not useful?

I never said or implied that headroom isn't useful, especially for an auto-gain feature. Obviously you need to be careful to not clip. I'm not familiar with that RME feature, but I presume they left 6dB "just in case". There just doesn't seem to be any reason to use more headroom than you actually need. It is, according to Sound on Sound, "totally unnecessary", and RME would "tend to call such claims voodoo".

Without anyone posting any audio examples of the problems that hot recording will cause, this is all I have to go on.

[drtedtan]

Originally Posted by clepsydrae

No need to be insulting. The idea, as was expressed in the previous linked thread, is that the level should be such that an RMS meter, such as that found on Reaper's master fader, would peak at -18 dBFS. RMS meters aren't calculating a single RMS value over the entire audio program, but work on a moving window, so they can indeed have a peak.

I was a bit abrupt in my communication, but did not mean for it to be insulting. Several people had explained that RMS and peak value are completely different things, so I was trying to avoid people continuing to talk "around" one another, seemingly talking about the same thing, but not doing so in reality.

[karbomusic]

Most profession preamps have both input and output knobs so the level you drive the preamp at has - absolutely nothing at all - to do with the level hitting the converter... unless you leave the output at unity.

Purely academic question:

Since most users here (I presume) don't have the feature pictured above how do they compensate for that? I think all clepsy is saying is that they never need to. Is it safe to assume that the only reason that feature exists (pre/post) is in order to drive the preamp without clipping the converter? Again purely academic.

My conversation was mostly about being nominal, aka keeping analog unity until the last moment for known good troubleshooting purposes which at this point is totally irrelevant.

[Lawrence]

Originally Posted by karbomusic

Purely academic question:

Since most users here (I presume) don't have the feature pictured above how do they compensate for that? I think all clepsy is saying is that they never need to. Is it safe to assume that the only reason that feature exists (pre/post) is in order to drive the preamp without clipping the converter? Again purely academic.

My conversation was mostly about being nominal, aka keeping analog unity until the last moment for known good troubleshooting purposes which at this point is totally irrelevant.

Sorry K, I had deleted that post because it seemed a little off topic in hindsight.

Short answer: They can't, because in those all in one devices the preamps are hardwired to the converters and those affordable designs don't really lend themselves to inserting analog gain/trim after the pre and before the converter.

My point there was that - with few exceptions - full time profession level studios do not use all in one devices. There are of course exceptions to that, where some smaller project rooms are running USB & FW devices and similar but for the most part converters are completely separate from sources and the inputs are on a patchbay or something. For example: My MOTU 2408 is all converters. It has no preamps on it. It records whatever levels I put into it's analog or digital inputs. The only actual amps on the thing (afaict) are the main output and headphone amps.

Here's what I mean below about outboard preamps, like on my old DBX 576. It has a Drive and a Level (output +/- 15dB) control so that you can drive the preamp input circuit and/or the tube circuit and still control the final output level. One thing has nothing to do with the other.

It also has something else you don't get on the all in one devices... a pad. Note also that the VU metering is switchable from input, insert and output... which is something else you don't typically get on inexpensive all in one devices, insert + return jacks.



In a perfect world a newer design like the above would have both a VU meter like the above and also have a horizontal peak meter on the output just beneath it, so you'd never really need to look at the daw meter to see the true peak signal, if you need to see it.

But yeah, most of the chatter about this stuff is realted to all in one devices where those things are hardwired together. My rule of thumb for that - along with just listening to see what sounds good, I have an 44VSL with crappy meters - is to leave my peaks around -10 max or so... and go a little hotter on transient signals where the peak and RMS gap is much wider.

[Bristol Posse]

Originally Posted by clepsydrae

As for the second, lots of people say that recording at -18 dBFS RMS average levels is an important thing to do for the sake of audio quality --

It's a myth if you do not know how your converters are calibrated

As far as I can see it came about several years ago when a lot of pro converters were calibrated so that line level 0VU (+4dBu) came out on the DAW at around -20dBFS to -18dBFS as the max input level (0dBFS) for the converter was +24dBu to +22dBu

This meant that if you had the meters dancing around 0VU on the analog side of things then the DAW meteres would be dancing around -20dBFS to -18dBFS (RMS) on the digital side of things

Line level was and is important as it is generally where most analog gear is designed to operate at it's most linear fashion with the best S/N ratio and you generally want to minimize noise, THD and other non linearities except where needed as a deliberate effect

Today prosumer and pro level interfaces are calibrated all over the place however.

I've seen some where the max input level is +10dBu. so in this case 0VU (+4dBU) would show up in the DAW as -6dBFS. If you shot for an RMS level of -18dBFS on this converter you would be nowhere near to line level and could run foul of poor S/N ratio

I've seen some prosumer devices (some of the Scarletts have different fixed reference values for in and out IIRC as well as RME where you can calibrate input and output separately if you want or need to) where the max input and max output levels are not the same so If you get the input gainstaging right, you may be feeding a completely different voltage on the output side of things. Probably not a big deal unless you are trying to integrate analog hardware

Then there are some high end standalone converters that can go beyond +30dBu as a max input level. shooting for an RMS level of -18dBFS with these would put you well above line level and could mean that you are pushing the gain stages ahead of the converters into their headroom and getting increased THD.
If that's what you want/need for a track, that's all well and good but it's a question of intent. Getting THD on purpose for effect rather than smearing it across the entire mix because your pres feeding the converters are dep into their headroom to try and get some arbitrary number on a DAW meter are two very different things

The -18dBFS rule is still not a bad safety if you don't know how to measure and calibrate converters. It will give you headroom for large peaks in tracking as well as an easier time mixing ITB so you don't find yourself constantly having to pull down track and master faders so as not to clip your output stage.

It's also worth bearing in mind that a lot of popular emulation plugins use -18dBFS as a simulated 0VU and will start to add increased (and often cartoonishly overstated) THD and non linearities if you hit them harder than 0VU RMS. Many plugs are user calbratable too but default to -18dBFS as a default reference level

However -18dBFS (RMS) sounds better as a rule is completely meaningless without the context of how your conversion is calibrated and how good (in terms of linearity and noise)and how colorful (by design) or clean your analog front end and/or mixing chain is or even what emulation plugins you may be using

YMMV
Matt

[Lawrence]

Maybe I better explain the -10 rule of thumb thing so it's not misunderstood. Here's the thing...

- On signals with little or no transients, say... like sustained strings, the peak and RMS signals are generally near identical, so -10 peak is for all practical purposes -10 RMS. Like if you play a test tone the peak and RMS are the same level.

- On more transient signals the peak and RMS gap will be much larger so a -10 peak from a transient signal will have an RMS level that will be down closer to the converter calibration level. For example, a hard hit big snare peaking at -10 might indeed have an RMS level of -18 or so, or much less, depending on the RMS window and all that.

So using -10 max peaks in the daw when you don't know what levels to use is (imo, mmv) a decent rule of thumb that covers a good range of musical signals. That signal will likely not overdrive your preamp for transients and will also have a good S/N ratio for non-transient signals like strings, and leaves 10db of headroom for the unexpected.

[karbomusic]

It's a myth if you do not know how your converters are calibrated

Well the EBU standard is -18 RMS which should at least give someone a fighting chance. Of the handful, I've seen and tested, they feel within a couple dB of that range, hence the use of the term "range":

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
European & UK calibration for Post & Film is −18 dBFS = 0 VU

That being said, if it were me and I didn't have the data for my card, I'd consider the standard as my starting point since I have no other reference. Actually, I'd hit the line input at unity with a 1K sine wave using a voltmeter and compare that with the post conversion metering but that's irrelevant at the moment. Unless someone tells me differently I'd assume those standards work the same way RFCs do with computer protocols. Its the baseline manufacturers "should" go by so different hardware etc. can properly interoperate. I'd feel less comfortable statistically to ignore the standard when faced with a card minus the spec.

For the record, I have no idea of the GearSlutz and other -18 threads you guys are discussing, I don't read those forums for painfully obvious reasons and my advice is to never even go there.

So using -10 max peaks in the daw when you don't know what levels to use is (imo, mmv) a decent rule of thumb that covers a good range of musical signals.

I'm thinking that and -18 RMS are going to live fairly close to each other. IOW, -10 peaks should likely have RMS values somewhere below that (remember a DI'd electrick guitar kicked all this off, so I'm not speaking of a synth pad) which keeps us on the same page.

[Bristol Posse]

Originally Posted by karbomusic

Well the EBU standard is -18 RMS which should at least give someone a fighting chance.

in my world EBU standard for finished program material and "Sonically better" for level setting at tracking/mixing do not equal the same thing

YMMV LOL no offence etc

[karbomusic]

Originally Posted by Bristol Posse

in my world EBU standard for finished program material and "Sonically better" for level setting at tracking/mixing do not equal the same thing

YMMV LOL no offence etc

Notice, I never said a thing about sonically better or finished program material. I thought I was quoting the conversion/alignment spec. for incoming signals?

[Bristol Posse]

Originally Posted by karbomusic

Notice, I never said a thing about sonically better.

isn't that the whole point of the thread though?

[Lawrence]

Originally Posted by karbomusic

I'm thinking that and -18 RMS are going to live fairly close to each other. IOW, -10 peaks should likely have RMS values somewhere below that (remember a DI'd electrick guitar kicked all this off, so I'm not speaking of a synth pad) which keeps us on the same page.

Well yeah, that was my point that signal content varies, that a sustained signal at -18 RMS is a literal analog 0 (given that calibration) but a transient snare isn't, the peak will be much, much higher than analog 0 if the snare's RMS is at the analog 0 point, so some understanding of how it all works together helps.

Most good recording engineers I know (mmv as usual) don't stress over that stuff so much. What they do is simply monitor the converter output - before - recording and if they like what they hear they record it. The exception to that (obsessing over levels as a practical matter) is maybe being pressed for time (not having enough time for a good setup) and/or being in a high pressure situation and leaving a bit more room for the unexpected by leaving lots of headroom to not ruin an otherwise great recording by clipping something where the orchestra might be rented @ $5000 an hour.

[karbomusic]

Originally Posted by Bristol Posse

isn't that the whole point of the thread though?

If a person had a problem, I might suggest they reset to what the specs say then work from there and end up where they need to. I think I'm way too simple minded for all the experts here though; I don't mean that in a condescending way. And I'm off topic concerning a previous question. It's all good though, no worries.

[karbomusic]

Well yeah, that was my point that signal content varies, that a sustained signal at -18 RMS is a literal analog 0 (given that calibration) but a transient snare isn't.

True, I'm certainly guilty of having my hands on signal flow for so many decades, I assume everyone automatically takes all this crap into account and already knows a sine wave has the near identical RMS and peak values and a snare is anything but that.

[Reno.thestraws]

Hey, great question, sounds familiar. :-) Maybe someone will do some tests and post examples. :-)

On ONE track, there's no big difference because if you loose 10db of SNR it very hard to hear (exept with good monitors)

But if you're recording a "2014 standard" production with

18 drums tracks
4 bass tracks
16 layers of guitars
8 layers of vocals
8 layers og BGC
8 synth tracks

and constantly bringing noise up. You'll face 3 big problems when mixing

1) NOISE
2) Fader alignement. You have to trim down all your fader (or items) by 12 db just to avoid clipping on the 2buss... before start mixing.
3) Volume inconsistance. If you recorded your snare at -1dbfs and your PAD at -1dbfs that means that the RMS level of the snare will be around -12 dbfs (or lower) and the RMS level of the PAD will be at -5 Dbfs

So IF you want to HEAR them at the same volume in your mix, your PAD fader will be ~ 7db lower than the snare one which is very Unnatural when mixing. If I raise two faders together, I expect to have same volume on both tracks. IF not, the mixer is useless. I can let all my faders at unity and gain staging with plug ins volume output

[Lawrence]

S/N ratios in digital only change in post with dsp processes (like compression). What you record is what you'll always have in the recording. Any decent converter should not be adding noise, any problematic noise will come from pre-amps and similar, not from inline converter circuits.

If you have dedicated converters you can easily test that by simply recording with nothing plugged into it and gaining up the resulting audio file to where you can see and measure the noise level, if any. The noise level, if any, should be far below the noise floor of your preamps or similar, making it irrelevant.

You can do the same with preamps. Plug a tone generator into your preamp and gain the signal up so that the input in your daw matches your converter level, -18, whatever. Unplug the tone genrator and record a track and do the same, gain up the clip and measure the noise. Duplicate that track 23 times and bounce the master bus and measure the accumulated noise level at unity gain.

That's the key thing with "engineering" audio, "testing", doing - your own - testing, finding out what "is", not letting others tell you what should be or what they think might be on a system that you own that they've never even seen or operated.

It's ... odd to me... mmv... how on the net so many recording enthusiasts stress over stuff they can't even hear and (in their context) probably doesn't even matter, while completely overlooking stuff that actually does matter, like testing to make certain that their headphones are in phase with their output and speakers.

[Cosmic]

Its bollix.

I recorded a song DRUNK OFF MY FACE last night.

Levels were ALL OVER THE GAF.(never clipping the front end though)

I could barely focus on the screen.

Worked out fine..they do have these little trim pots on nearly every pluging I've ever seen.Just start twiddling them..you'll be grand.

YMMV

[Reno.thestraws]

If you have dedicated converters you can easily test that by simply recording with nothing plugged into it and gaining up the resulting audio file to where you can see and measure the noise level, if any. The noise level, if any, should be far below the noise floor of your preamps or similar, making it irrelevant.

I know. And As I said before, even the SNR of a good preamp is bigger of 99% of reality

Exemple : you record a Snare drum in a decent room

What's the noise floor of the room?
a Super quiet 5 stars recording room is ~ 30db SPL

A super HARD hit on the snare will be at ~ 115 db SPL

So with 85db of dynamic range, you get enough room to record it properly

Knowing that, what's the need to record at 0dbfs in your DAW?

even if you leave 24db of headroom, you still have 120db of dynamic range and so 25 db more than you really need. (with 24 bit converters of course)

And I don't even talk about Direct guitar recording (where all the discussion start) when the pick up of the guitar have a SNR of 70 db max... in a perfect world)

[brainwreck]

Whatever works for you, according to your ears, I say. I'm all for running little experiments (I have done lots of them to satisfy my own curiosity), but people should be doing this for themselves on their own gear, if there is any personal concern on incoming levels. There is nothing complicated or difficult about it. Set Reaper's master meters to RMS, record some stuff at different levels, and listen. Use what you hear as a personal guideline.

Then again, if you like to geek out on specs and abide by rules, for whatever reasons, do so.

[clepsydrae]

Bristol/Matt, your post is clear and in accord with my current understanding/suspicions except for this paragraph:

Originally Posted by Bristol Posse

Then there are some high end standalone converters that can go beyond +30dBu as a max input level. shooting for an RMS level of -18dBFS with these would put you well above line level and could mean that you are pushing the gain stages ahead of the converters into their headroom and getting increased THD.
If that's what you want/need for a track, that's all well and good but it's a question of intent. Getting THD on purpose for effect rather than smearing it across the entire mix because your pres feeding the converters are dep into their headroom to try and get some arbitrary number on a DAW meter are two very different things

Assuming that by "high end converters" you're talking about contemporary hardware, my understanding is that hitting the analog stage harder both gets you less THD (according to Jim Williams, I have no personal idea) and otherwise makes no difference to audio/tonal quality. (except less noise, if that's relevant to you). That paragraph is exactly the issue I want to probe: is "pushing the converters into their headroom" going to do anything perceptible in 99% of real-world situations with <15-year-old gear, or is it a worry that we should let go of?

However -18dBFS (RMS) sounds better as a rule is completely meaningless without the context of how your conversion is calibrated and how good (in terms of linearity and noise)and how colorful (by design) or clean your analog front end and/or mixing chain is

Sure - my impression is that the linearity/noise/colorfulness of basically all contempoary ADCs mean that the "rule" is wrong in terms of audio quality/THD/etc.

Originally Posted by Lawrence

Like if you play a test tone the peak and RMS are the same level.

Sorry to not pick, but just in case anyone is confused by this: AFAIK the RMS will be ~71% of a sine test tone, [see later in thread -- Bristol explains that a 3dB offset is traditional in the audio sense of "RMS" metering] and AFAICT the meter in Reaper at least reflects this as well (unless it's displaying with a "display offset", which it does by 14dB by default):

Originally Posted by Lawrence

So using -10 max peaks in the daw when you don't know what levels to use is (imo, mmv) a decent rule of thumb that covers a good range of musical signals.

...souds good to me...

Originally Posted by Lawrence

That signal will likely not overdrive your preamp for transients

...if you're talking about clipping, sounds good. If not, you know my skepticism.

Originally Posted by Reno.thestraws

On ONE track, there's no big difference because if you loose 10db of SNR it very hard to hear (exept with good monitors)

Not sure I follow: you get better SNR by recording hotter.

For example, I just tried an 8k test tone, because the article I linked to earlier said such tones were easiest to hear the THD from. I recorded it at -0.1 dBFS and at -18 dBFS RMS, and normalized the results.

[update: Reno makes the good point that because of how I ran the test, the SNR went down at the output; thus the noise of the ADC was commingled with the noise of the output, once the signals were normalized. See the improved test coming later in the thread.]

Here's the -0.1 dBFS peak result:



...note the smaller 16k THD spike visible in the spectrum. Here's the -18 dBFS RMS version:



(If anything, the THD looks worse in -18, which is what Jim Williams says to expect in most gear, but I didn't actually measure that. I certainly couldn't hear a difference.) What is clear is that the noise floor is louder in the -18 dBFS RMS average level version, although it's still very, very low.

Re: the levels issue you raise: yes, I have acknowledged many times already that for some workflows the level management is a good reason to argue for -18 dBFS RMS averages. I'm only concerned here about the sound quality issues.

Originally Posted by Reno.thestraws

Knowing that, what's the need to record at 0dbfs in your DAW?

Less noise, if you care about that, and less THD, if you care about that, if Jim Williams is right.

even if you leave 24db of headroom, you still have 120db of dynamic range and so 25 db more than you really need. (with 24 bit converters of course)

The dynamic range the 24bit digital domain is not relevant to the dynamic range of the analog part of the ADC input, which is far less than what 24bit can allow.

Recording hotter is about the dynamic range available in the analog stage.

When we talk about recording at "-0.5 dBFS" we're using a value from the digital domain as shorthand to refer to the signal level on the analog side, which varies according to the converter's alignment. The point is that recording "hot" (however you decribe it, but meaning "close to the specified maximum of the analog input stage") results in less noise from the analog stage. Whether your converters are 24bit or 8bit, that noise will be present in the digital signal, so it makes sense to record hotter to avoid noise, if you care about noise at the levels in question.

I'm not saying I do, or that anyone should. But you don't lose SNR by recording hotter, in this context.

And, more to the point, you don't gain anything in terms of audio quality by recording quieter, as far as I can tell, so far.

[clepsydrae]

Originally Posted by brainwreck

Set Reaper's master meters to RMS, record some stuff at different levels, and listen.

I would add: "if you hear a difference, render the results and do blind ABX testing, or don't bother with the experiment, because any subtle differences will be swamped by psychological bias."

Also, note that Reaper's RMS meters default to displaying with a 14dB offset. Meaning that what reads as "-18" is actually -32 dBFS RMS unless you change the offset to 0!

I wonder what percentage of people using Reaper's RMS metering know that. :-) I certainly didn't until I looked into this stuff.

[brainwreck]

Originally Posted by clepsydrae

I would add: "if you hear a difference, render the results and do blind ABX testing, or don't bother with the experiment, because any subtle differences will be swamped by psychological bias."

Also, note that Reaper's RMS meters default to displaying with a 14dB offset. Meaning that what reads as "-18" is actually -32 dBFS RMS unless you change the offset to 0!

I wonder what percentage of people using Reaper's RMS metering know that. :-) I certainly didn't until I looked into this stuff.

I think ABX testing is overrated, personally. Context matters. I can elaborate, if you like.

My offset is set to 3. I guess I should have remembered that earlier on. What's 3db among friends?

[Bristol Posse]

Originally Posted by clepsydrae

Bristol/Matt, your post is clear and in accord with my current understanding/suspicions except for this paragraph:



Assuming that by "high end converters" you're talking about contemporary hardware, my understanding is that hitting the analog stage harder both gets you less THD (according to Jim Williams, I have no personal idea) and otherwise makes no difference to audio/tonal quality. (except less noise, if that's relevant to you). That paragraph is exactly the issue I want to probe: is "pushing the converters into their headroom" going to do anything perceptible in 99% of real-world situations with <15-year-old gear, or is it a worry that we should let go of?

Depends on the hardware. Clearly, I'm not using whatever Jim Williams is using. For example my Chandler Preamps and Neve processors definitely (and by design) add more THD/non linear behaviour as they get pushed harder. It's not just opamps and electronically balanced gear out there. many of my pieces have transformers for balancing which saturate according to signal level and I also have some tube gear that is definitely getting more saturated as the levels are increased beyond it's nominal operating spec

I think it probably matters less in all in one interfaces which (theoretically at least) should have the pre and the converter designed to match and integrate properly and are usually all about "clean gain". but many "high end" converters like Burl, Lynx etc do not have a built in preamp stage, you need something else to get the signal to line level outside of the converter and depending on what you choose, higher levels could give you enormous changes.

The converter itself will make a minimal, miniscule difference to the THD, Noise and crosstalk etc because even without pres there is still an analog front end ahead of the A/D or after the D/A but it's more or less irrelevant compared to the Pres, compressors, EQs etc that come before or after it. Well unless there is a serious flaw in the converter

I'm not using high end conversion just RME, but I'm not using the built in "clean" pres either most of the time. I bypass those and go line in straight to the converters from external pres and signal processing.

I think it really boils down to knowing your gear and how it responds to get what you want

It'd be easy for me to prove to anyone that a Chandler pre pushed hard into a neve portico 500 series module pushed hard can give you so much THD that it almost sounds like a guitar distortion pedal. I can do it all without coming anywhere near to clipping the RME converters or right at 0.01dBFS if I so chose.

But that doesn't prove anything about the "-18dBFS Rule" nor does it disprove anything that Jim Williams has posted about his gear and his results as I'm using a completely different analog, A/D, D/A, Mixing chain than he is

the only tests that matter are on your own gear in your unique cicumstances (unless someone happens to have exactly the same chain doing the exact same thing you do).

If you can't do that the specs will give you a guideline and leaving some overs and mixing headroom isn't a bad idea so you don't have to be concerned about clipping or constantly fighting the 0dBFS output ceiling when mixing

[clepsydrae]

Originally Posted by Bristol Posse

but many "high end" converters do not have a built in preamp stage, you need something else to get the signal to line level outside of the converter

Oh, sorry, I misread that paragraph I quoted before... by "gain stages ahead of the converters", you meant separate pieces of gear, not the analog stage of the converter itself, and with your other comment about "pres feeding the converters ... deep into their headroom" I somehow overlooked "pres" and interpreted that sentence in the same mistaken way. I understand what you mean, now; sounds sensible.

I do still feel that tests are valuable if they might establish a clear pattern across a class of gear.

[clepsydrae]

Originally Posted by brainwreck

I think ABX testing is overrated, personally. Context matters. I can elaborate, if you like.

I am curious, but let's hold off on that issue for now or start another thread, since this one is already plenty convoluted. :-)

[Lawrence]

...if you're talking about clipping, sounds good. If not, you know my skepticism

Yes, I was talking about maybe clipping the preamp and/or overdriving it unfavorably. The context here is people doing home recording who, for the most part, have...

1. Less than stellar monitors.
2. Less than stellar isolation to hear what's being recorded.
3. Lower cost preamps with less headroom.

It's in these situations that some only really hear "issues" during playback, partly because of item #2. In a studio with good isolation you can always hear exactly what's being recorded, and only that, so if you record a signal that drove a preamp to an unfavorable distortion state, it's because you weren't paying attention, not because you couldn't actually hear it before you hit the record button because you were actually hearing it + the live source signal at the same time.

So basically I was saying... "If you really have no clue about all that stuff, we were all newbies at some point, set all of your levels feeding your little mobile device to peak at -10, max, not literal. Use -10 as the 0 point and just record."

For people who know how to use their gear already... they really don't need our advice.

[clepsydrae]

Originally Posted by Lawrence

Yes, I was talking about maybe clipping the preamp and/or overdriving it unfavorably.

Hmmm, well, it's the latter that I'm on about... In terms of not-ancient ADCs made by companies that anyone has heard of, I'm not convinced that there is such a thing as "overdriving unfavorably" if you're peaking at the analog input such that the digital signal exhibits peaks just under 0 dBFS.

If there is a contemporary ADC exhibiting this effect, I'd love to hear the audio samples demonstrating it, presented with whatever listening conditions are required to be able to hear that difference.

RME says that their devices do not exhibit this, Sound on Sound implies that no such devices do, and Jim Williams believes that virtually no such devices do, and no one on the internet that I can find has posted an audio example from one that does or cited a manufacturer that advises recording quiet for audio's sake, and several major ADC manufacturers use language that implies that recording hotter than -18 dBFS RMS averages is advised. Amazingly, even unsupported claims of specific devices that exhibit this are nearly non-existent. Everyone says "yes it's real thing, but it depends on your gear, test for yourself" without any elaboration of how they know that to be true.

This is the pattern evolving in my eyes, but i'm absolutely willing to be convinced otherwise.

[Bristol Posse]

Originally Posted by clepsydrae

Hmmm, well, it's the latter that I'm on about... In terms of not-ancient ADCs made by companies that anyone has heard of, I'm not convinced that there is such a thing as "overdriving unfavorably" if you're peaking at the analog input such that the digital signal exhibits peaks just under 0 dBFS.

If there is a contemporary ADC exhibiting this effect, I'd love to hear the audio samples demonstrating it, presented with whatever listening conditions are required to be able to hear that difference.

RME says that their devices do not exhibit this, Sound on Sound implies that no such devices do, and Jim Williams believes that virtually no such devices do, and no one on the internet that I can find has posted an audio example from one that does or cited a manufacturer that advises recording quiet for audio's sake, and several major ADC manufacturers use language that implies that recording hotter than -18 dBFS RMS averages is advised. Amazingly, even unsupported claims of specific devices that exhibit this are nearly non-existent. Everyone says "yes it's real thing, but it depends on your gear, test for yourself" without any elaboration of how they know that to be true.

This is the pattern evolving in my eyes, but i'm absolutely willing to be convinced otherwise.

I believe I'm starting to think long similar lines in regards to all in one interfaces. Every manual I have ever read for such devices suggests getting the clip light to show and then backing off just a touch. My RME manuals do tell you how to calibrate to various levels but that is mostly to deal with integrating various hardware options with various reference levels or to create more headroom to deal with more dynamic sources . They do not suggest recording at lower levels

Using my outboard gear I could theoretically get at any DAW level I like at the converter by simply using a final trim control after the final signal processor and ahead of the converter. But since that is another gain stage or pad and I like to use the bare minimum I can in between the source and the converter I don't want to do that.
So in my case using outboard pres and processors (often designed for color when pushed) most of my signals end up at line level(ish) for clean and slightly higher for pushed.
But again where this shows up in the DAW is rarely around -18dBFS RMS. Usually more like -15 to -12dBFS RMS

I think the -18 thing has just been latched onto for the couple of reasons I stated earlier
It was originally posited, years ago,that this is around where many pro level stand alone converters (ie no onboard pres) were calibrated. and since these were fed by outboard gear, Line level signals were the norm

It's where emulation plugins arbitrarily set 0VU, perhaps because of the pervasiveness of the -18dBFS rule. So having levels already set here means you don't have to re gainstage (an add another gainstage/pad) when using these popular plugins (like tape or console emulations that are supposed to be strapped accross every track in a mix)

[Bristol Posse]

Originally Posted by brainwreck

I think ABX testing is overrated, personally. Context matters. I can elaborate, if you like.

My offset is set to 3. I guess I should have remembered that earlier on. What's 3db among friends?

I'm with you here

Whilst I am 100% certain that ABX testing works to scientifically establish ability to discern if there is a difference between two things, I have no desire to second guess every decision I ever make or have ever made in a largely subjective field where isolating individual things in test scenarios rarely mirrors real life usage.

At some point you have to start trusting your decision making skills, even if they are prone to expectation bias and may not be based on pure scientific tests

plus I know everything and I'm always right. just ask my wife

[Reno.thestraws]

For example, I just tried an 8k test tone, because the article I linked to earlier said such tones were easiest to hear the THD from. I recorded it at -0.1 dBFS and at -18 dBFS RMS, and normalized the results.

What's the level of the tone before hitting the preamp?

If you run the test tone from your daw to a line output and plug it into a input, then you bring an amplified signal. In short you don't test the inout but the output...

You need to feed your preeamp with a signal that have a microphone level so something like 1,5 mV and low impedance

And test tone at 8k is irrelevant too because the two zone where the converters are the most sensible is 2500hz and below 100

[clepsydrae]

Originally Posted by Reno.thestraws

And test tone at 8k is irrelevant too because the two zone where the converters are the most sensible is 2500hz and below 100

I ran the 8k tone because I was curious about the THD+N issue, and the article I linked to above said that THD+N is much more audible to humans at higher frequencies -- I'm not wise in the ways of THD testing and am not trying to make claims about it. I posted the graphs to make the point about the noise floor.

What's the level of the tone before hitting the preamp?

I don't know, but assuming the presonus spec sheet is accurate, the hotter signal would be just under +14dBu.

If you run the test tone from your daw to a line output and plug it into a input, then you bring an amplified signal. In short you don't test the inout but the output...

Yes, there is amplification at the output. But that amplification didn't change between the two tests: I lowered the signal to -18dBFS RMS only in the DAW, so the noise floor, the impedance, etc, were exactly the same between the two tests, so the difference should relate to the noise floor from the ADC input stage. [update: Reno makes the good point that the SNR went down at the output; thus the noise of the ADC was commingled with the noise of the output once normalized. See the improved test coming later in the thread.]

So, the only other explanation I can imagine is that the output stage is generating more THD and noise when it's driven more quietly by the -18 signal. Maybe, I'm not qualified to say, but that seems unlikely regarding the noise.

But whether the THD and noise came from the output stage or the input stage, it sure doesn't look like the input stage distorted more or got noisier when both stages were driven hotter, unless some truly remarkable stuff is happening in that connection to mask that effect.

It's not really a debated question whether noise will be worse when recording quieter... that's basic gain staging. I'm sure there are rare exceptions out there somewhere, but I don't believe ADC input stages are likely to be those exceptions.

BTW, Lavry and Apogee have both responded... I'm posting next about what they said, and they address the noise issue as well.

[Reno.thestraws]

I lowered the signal to -18dBFS RMS only in the DAW, so the noise floor, the impedance, etc, were exactly the same between the two tests, so the difference should relate to the noise floor from the ADC input stage.

Yes but you trim down the preamp between the two takes?


Your preamp wat at let's say 10 o clock for the take at -18 and at 2 o clock for the take at 0dbfs?

[clepsydrae]

Originally Posted by Reno.thestraws

Yes but you trim down the preamp between the two takes?

Your preamp wat at let's say 10 o clock for the take at -18 and at 2 o clock for the take at 0dbfs?

No -- it's a single stereo output from my interface going to two separate inputs which have no gain knobs -- fixed gain on those inputs. The difference in signal level is achieved in the DAW on the way out, not with any gain knobs. There is a gain knob on the output, but it remains constant between the two takes.

It's possible that the two inputs have slightly different characteristics, but I've swapped which signal goes to which input with no change in results.

[clepsydrae]

Originally Posted by clepsydrae

No -- it's a single stereo output from my interface going to two separate inputs which have no gain knobs -- fixed gain on those inputs. The difference in signal level is achieved in the DAW on the way out, not with any gain knobs. There is a gain knob on the output, but it remains constant between the two takes.

To clarify -- there are not two takes -- the interface plays both signals at the same time, with different levels in each. Whatever differences might be there in the output/input channels is controlled by changing the routing. (out1->in1 and out2->in2 changes to out1->in2 and out2->in1).

[Reno.thestraws]

Originally Posted by clepsydrae

No -- it's a single stereo output from my interface going to two separate inputs which have no gain knobs -- fixed gain on those inputs. The difference in signal level is achieved in the DAW on the way out, not with any gain knobs. There is a gain knob on the output, but it remains constant between the two takes.

It's possible that the two inputs have slightly different characteristics, but I've swapped which signal goes to which input with no change in results.

So, You didn't test the input, but the output. That means taht you REDUCE the volume of the output (and so the SNR) before reinjecting.

If i Understand the whole point of this thread, It's the opposite that we discuss.

Try these experience

-Play a 8k tone on your speaker (not too loud)
-Place a SM57 or a similar dynamic microphone if front of your speaker (DO NOT MONITOR THE INPUT or you'll get larsen)
-Plug the SM57 on a preamp
- create a new track in reaper with the input set to your SM57 (Do not monitor input!)
- Set the preamp to achieve a level of -0.5dbfs in REAPER on the track input and record
- now ,mute the track you just recorded and create a new one
- set the preamp to achieve a level of -18dbrms in REAPER on the track input and record
- Normalize the two recorded track and make your analysis again

[clepsydrae]

Lavry Engineering and Apogee have both responded to my questions.

I sent Lavry a slightly different version of the email, so I'll include that at the end of this post. Apogee got the email I posted earlier in this thread. Kudos to both for getting back so quickly -- I'm really impressed with RME, Lavry, and Apogee for responding with such care and speed. A rare thing these days. I'll certainly consider them first for my next purchases. (If I ever get back to my life from this thread.)

Again, both are responding only to my email, not to anything else in this thread or the internet, etc.

The Lavry summary:

- affirms value of setting the output stage of analog gear earlier in the signal chain to the optimum levels and adjusting alignment of ADC to match that, since those output stages are typically more variable than the ADC input stage

- recommends recording at -1 to -0.5 dBFS peaks

- says recording quieter won't matter, ignoring the noise floor, but that if you "under-record by 20 dB" (presumably meaning peaks at ~-21 dBFS), you will probably hear a "minor degradation" requiring good monitoring/ear training/talent to hear.

- key quote: "setting the input level so the signal peaks never reach -6dBFS with our AD converters is not likely to have much of a down-side; but it probably will not result in a "cleaner" recording than one made with the input level set so the loudest peaks just reach -1 to -0.5dB."

Apogee summary:

- affirms value of huge headroom for avoiding clipping and affirms that recording quieter benefits the overall studio gainstaging

- says a "high-quality" converter "should have essentially linear performance throughout its dynamic range".

- key quote: "a signal recorded with a peak at -18dBFS and the same signal recorded peaking at -0.5dBFS ... then normalized in a DAW should ideally be tonally identical, with the -18dBFS signal having a 17.5dB higher noise floor."

They seem to have dropped the "RMS" off the "-18", but their answer seems pretty clear.

They provide Burl converters as an example of converters that change in tone up the scale by distorting more, because they are designed to do so. I checked the Burl website, and indeed, their B2 is designed to impart a tonality. Amusingly, though, even their site includes the quote "don’t be afraid to run the B2 Bomber hot, it only sounds better." :-)

Full answer from Lavry:

Originally Posted by Lavry Engineering Technical Support

First, you need to be clear on the difference between the RMS level and the peak level.

It is entirely possible for an analog signal with an RMS level of -18dBFS to also have signal peaks that "exceed" 0dBFS. For example, a "live" source without limiting or compression could easily do this.

It is also important to match the capability of the analog source to the input of the AD using the input level adjustment of the AD converter. In some cases, even though the manufacturers of the analog gear claim a peak output level capability of say +18dBu, the distortion increases dramatically below this level or the noise increases dramatically when the output is turned all the way up. In this case setting the input adjustment of the AD so that the analog source never gets near +18dBu on the signal peaks may produce the same "cleaner" results that under-recording with the AD converter set so that +18dBu = 0dBFS, without the need to under-record. This is why we recommend not worrying about specific dB levels and setting your analog source to its optimum output level (where is sounds the best), then adjusting the input of the AD converter so the signal peaks reach ~-1 to -0.5 dBFS. If you want to employ clipping, you would set the AD input to a higher level so the analog source does not "run out of gas" before the converter clips.

In theory (ignoring the noise floor), with a completely "linear" AD converter, it would not matter whether you recorded with the signal peaks reaching -0.5dBFS or -12dBFS if you can make up the gain in the digital domain. But most commercially available AD converters are less than ideal.

There are many places in a real-world AD converter where distortions can be introduced, including overload of the analog input stage or intermediate stages between the input and the sample and hold circuit at the input of the actual AD converter (typically an IC).

Our gear is capable of accepting analog signals right up to the rated maximum input level without distortion, so it is not likely that a relatively small difference in input level (< 6dB) would result in an audible difference. On the other hand, unless you are using one of our Gold AD converters, if you under-record by 20 dB, and raise the level digitally 20 dB, you probably will hear a minor degradation with the lower level recording.

It is hard to say how this difference would "sound" with our AD converters, aside from an apparent slight loss of over-all quality. It would require a good monitoring system and some natural ability plus ear-training to discriminate the difference. In the case of other brands of AD converter the audibility of the difference will vary greatly, depending on the manufacturer.

So if you are not using clipping as a means of dynamic range control, setting the input level so the signal peaks never reach -6dBFS with our AD converters is not likely to have much of a down-side; but it probably will not result is a "cleaner" recording than one made with the input level set so the loudest peaks just reach -1 to -0.5dB.

And from Apogee:

Originally Posted by Apogee

Our perspective on the matter is that a high-quality converter design like an Apogee should have, for practical purposes, essentially linear performance throughout its dynamic range. When recording at 24-bit with the excellent performance of our modern converter designs (Symphony I/O’s A/D has 120dB of dynamic range), the noise floor is so far down that even if your peaks are hitting at -18dBFS, you still have 102dB of dynamic range to work with - more than actually ends up on most commercial music products. The theory goes that it’s better to be a little closer to the noise floor and have a huge safety net of “headroom” than to go into digital distortion, which makes a lot of sense in most applications.

As further protection from overs, Apogee converters have long featured analog FET-based limiters before the A/D converters (“Apogee Soft Limit”) to help “catch” any stray transients. This is sometimes used creatively to add a bit of “analog” character with more rounded transients.

To your question, a signal recorded with a peak at -18dBFS and the same signal recorded peaking at -0.5dBFS (using no Soft Limit or similar functionality) then normalized in a DAW should ideally be tonally identical, with the -18dBFS signal having a 17.5dB higher noise floor. With most modern types of music, you probably couldn't notice the difference once the track is mixed and mastered.

Then things get more complicated when you get into circuits that significantly increase in distortion as they reach full scale (Burl converters, for example) - in that case, the signal recorded at a lower should be tonally different when level matched. To what extent depends on the specific design.

Sorry for the long email, but I hope that all makes sense! In short, we think the concept that recording digitally at lower levels (and not all the way up to full scale) is typically beneficial to overall gainstaging is sound advice for most applications.

The email that I sent Lavry (worded slightly different than the one to Apogee):

Originally Posted by me

Hi there -- question about Lavry interfaces, if you have the time:

There is a claim made often on the internet that many ADC converters sound "better" in various ways if gainstaging is such that signals coming in to the ADC result in a digital signal with RMS levels at about -18 dBFS. This is specifically claimed in contrast to recording "hot" signals that result in peaks just under 0 dBFS.

There are many sensible reasons for the recommendation, including matching analog gear voltages, convenience of workflow once in the DAW, and so forth.

I'm curious only about the audio quality aspect of the claim: would your gear sound any different if a 24-bit signal was recorded at -18 dBFS RMS vs. recorded at -0.5 dBFS peaks, provided that no clipping has occurred, and ignoring the noise floor, once normalized in the DAW?

If there would be a difference besides the noise floor, under what circumstances would this difference be discernible? (e.g. high-end monitoring? expert listeners? certain types of audio or instruments? etc.)

Thanks a lot for your time!

[Reno.thestraws]

Originally Posted by clepsydrae

To clarify -- there are not two takes -- the interface plays both signals at the same time, with different levels in each. Whatever differences might be there in the output/input channels is controlled by changing the routing. (out1->in1 and out2->in2 changes to out1->in2 and out2->in1).

SO, you reduce the output from 24 bit to 21 bit and it's totally normal that you have such results...

It's not what's on purpose here (if I understand eveything)