Drew,

I understand what you're saying.  But your way of mixing promotes a "no rules apply here" type of mixing, which is why we have this Loudness War mess today.  It'd be a different story if the music industry had some type of loudness stardard like the film industry does.  We wouldn't have producers pushing the ME's to bring levels higher than before because they would have to obey those standards.  And this is what this article, as well as the K-System, promotes.

So careful mixing ITB, regardless of how much headroom we get with a 64 bit engine, pays off at the end of the day.  You'll get consistent mixes that translate better volume-wise because you're working with certain standards.  What's the point of mixing hot if, at the end of the day, I have to bring levels down anyways?  It just makes no sense.  It is so much easier to work on mixes where you don't have to keep lowering levels everytime you apply an FX, or add another track.  The goal is not to pretend to work in an analog mixer, but to produce consistent mixes that sound good.  That's the point of the article.


Take care!
 
 
 

Jose7822


Drew,

I understand what you're saying.  But your way of mixing promotes a "no rules apply here" type of mixing, which is why we have this Loudness War mess today.  It'd be a different story if the music industry had some type of loudness stardard like the film industry does.  We wouldn't have producers pushing the ME's to bring levels higher than before because they would have to obey those standards.  And this is what this article, as well as the K-System, promotes.

So careful mixing ITB, regardless of how much headroom we get with a 64 bit engine, pays off at the end of the day.  You'll get consistent mixes that translate better volume-wise because you're working with certain standards.  What's the point of mixing hot if, at the end of the day, I have to bring levels down anyways?  It just makes no sense.  It is so much easier to work on mixes where you don't have to keep lowering levels everytime you apply an FX, or add another track.  The goal is not to pretend to work in an analog mixer, but to produce consistent mixes that sound good.  That's the point of the article.


Take care!

 Jose,

Just to be clear, I'm not advocating running things hot. The point is, non-linear plugins aside, there is absolutely no difference in a DAW between trimming channels down and boosting the output vs. running channels hot and reducing the output. Technically, it really doesn't (and can't) make it any more or less loud (and thus make more or less sense) either way. What's important in terms of the loudness wars is not overcompressing/limiting/maximizing and leaving some headroom at the output. 

In terms of plugins, at the track level you would always start below 0dB anyway (assuming you came from an ADC and don't boost things), so you shouldn't run into any trouble with most plugins. The only ones that will be affected are non-linear plugs designed to add "color" or saturation. If you didn't want color or saturation, you probably wouldn't be using that plugin. And you absolutely have to use your ears to decide how much color you want.

The only place you really need to be careful is if you send several hot tracks to a sub bus and then place a non-linear plugin at the bus level.

I don't have a problem with the idea that we want to use techniques to produce consistant mixes that sound good, and also don't want to produce overly loud, overcompressed mixes (+++). My problem is that the method argued here doesn't make any sense to me. People are claiming it will give you better results, but I have yet to see what I think is a valid reason why that should be so. I mean, if you argued it maximized your fader/meter resolution or something, I might buy that, but talking about maintaining headroom in the mix bus where you have hundreds (or thousands) of dB's of headroom doesn't make any sense. On the other hand, K-System makes sense to me, and Katz argues it well.

And I don't advocate "no rules" mixing, but rather just a few simple rules that make sense to me:
1. Don't repeatedly compress things, or apply more than a few dB of gain reduction at a time.
2. Don't ever apply more than a few dB of gain reduction at the master bus.
3. Leave several dB's of headroom at the output.
4. Don't overprocess/oversaturate things. If you add harmonics to everything, you end up with mud.
5. Cut, don't boost (unless absolutely necessary).
6. Use your ears.
7. And as for this discussion: if one track clips during mixing for some reason, don't waste time reducing each and every track (trying to maintain the balance between tracks); just turn down at the bus level so it doesn't clip there.

ITB gain staging is not on the list, only because it doesn't make any sense to me. You have plenty of headroom and S/N ratio, so what's the point?. Because we need to do it in the analog world we should do it ITB too, even though we don't need to?


In terms of the loudness wars, my deepest darkest fear is that it isn't really an education problem, but rather that people might (gasp) actually like that sound.

Peace,
drewfx

It's not just about final "volume" at the main outs.  It's about the character of the sound, which we want to present at its optimum.

It's not true that you can change the gainstaging ITB for a given signal, and it will be equally good-sounding just as long as it doesn't clip. 

Some people need to spend less time on "theories" and start twisting knobs and listening.  It becomes quite apparent when you do.

Jose7822


Everything you've said assumes no one will be using plugins that color the sound when in fact that is the one thing we all strive to use (that magical plugin that will make our songs sound "analog").  You're also dismissing the possibility of using External Inserts, for example.

No, I basically have said you have to set levels appropriate to each plugin when the plugin is non-linear (i.e. when the plugin sound varies with level). The problem is, you have no way of knowing what appropriate "gain staging" is for a given plugin without using your ears. As I have said, many plugins that add color have input level/drive/etc. controls at the plugin in order to set optimum levels (external inserts in Sonar are like this - they have Send/Return level controls). For the plugins that don't, you need to use your ears to set the trims to appropriate levels to get the proper amount of color. Setting your trims generically to -6dB (or whatever) might give you the proper level on occaision, but more likely it will be either too low or too hot in a given situation. If we're using a "color" plugin, we probably aren't looking to send the plugin a low level to keep it in its linear range - we want it in the non-linear range (that's why the programmers wrote it to be non-linear, and thus give it "color"). The point is, there is no way of determining what the proper level is without listening. You can't just subtract an arbitrary number from 0dB and say "that will always be perfect for all plugins in all situations, because that's the number that works best on a great sounding analog SSL console".

  Another example is when bouncing softsynths where having high levels may make them clip. 

Agreed. Anytime you leave the DAW, whether via sound card or bouncing to disk (if at 16 or 24 bits), you need to make sure you don't clip. But what does this have to do with setting the trims to an arbitrary level?

It's like Mike said, why make it harder than it should be?  Just because we can doesn't make it right.

Why set trims lower than they need to be? Just because we can? Because we need to in the analog world, where otherwise we will run out of headroom? You seem to be missing my point here, which is that gain staging is just not the same ITB as in analog. In the analog world, proper gain staging is based on balancing S/N ratio and headroom. You set levels high to maximize S/N, but not too high or you'll run out of headroom. There is a definite "optimium level" in analog. ITB you just set things so they don't clip when leaving the DAW (whether via sound card or writing to file), or where a given plugin sounds best. You don't really have to worry about S/N or headroom. There is no "optimium level" in a DAW's floating point mix bus. And the "optimium level" for a color plugin differs for each plugin (and possibly each situation). Using analog gain staging techniques ITB won't make it sound any more "analog".

I'm not trying to be contentious here. I just think the author of the gearslutz article probably mistakenly believed (as many people do) that DAW's do things at 24bits internally, and gave advice based on that flawed premise. Unfortunately, many people with a background in analog audio (even some circuit designing geniuses), don't understand all the details of digital. Many myths exist because sometimes people jump to conclusions without really understanding the intricacies of how digital works. And when someone with great credentials in the analog world makes a statement about how things "work" in digital, it's taken as authoritative even if it's technically wrong.

drewfx

Drew,

Here's the deal, and the point I'm trying to get across.  I understand what you're saying about the amazing headroom we have inside a single and double precision floating point engine.  Yeah, it is great to have the ability to go past 0dB FS and not clip, but my argument is that inconsistent gain staging ITB is what makes our mixes inconsistent (hence why I brought up the K-System).  It also makes mixing harder because we have to constantly lower volumes ITB to get everything sounding clean wether on the output or inside a plugin's UI.  And let's not forget the headroom needed before going into the mastering stage.

Sure, we can just keep lowering things until we meet the desired output level, but why make things harder?  Isn't it better when things just fall in the right spot without having to constantly lower the faders?  Granted this takes into account that proper gain staging was conducted before going ITB.  But you're giving yourself enough headroom to work with without worrying about clipping anywhere.  If you're monitoring using the K-System, you can just keep speaker levels as they are because you have enough headroom.  So basically, the levels almost take care of themselves if you leave the appropriate headroom all throughout.  And this is the point I was trying to get accross.  This way you get consistent mixes that are easier to work with.


Take care!

Mike,

But the answer is indeed simple. 

You just need to understand what the meters in the analog world translate to when in the digital world.  As you know, when your analog gear reads 0dBVU, it means that the signal passing through is at its nominal level (or the level at which it provides the best S/N ratio).  This yeilds a clean recording with no coloration.  Pushing these levels a little ususally gives the sound a nice color.  But you have to realize that, by doing this, you're trading optimal S/N ratio with coloration.

Your signal chain should be set up in a way that 0dBVU = +4dBu = -18dBFS (more or less).  How much exactly is based on the amount of headroom built into your converters when it is being fed a +4dBu signal (reading your audio interface's manual should provide you with the correct number).  But we also need to consider how much headroom was built into your preamps and any other analog gear used in the signal chain.  Otherwise, you might end up pushing your converters too much and negatively altering the S/N there.  This can easily happen when people record too hot.  Believe it or not, too hot can be as low as -6dBFS PEAK.  And this is not even considering intersample peaks which can be as high as 6dB.  So you may be already clipping even though your meters show you're not.

Optimizing bitness when recording in digital sounds like the logical way to go about this.  But it doesn't consider any of the above mentioned points like S/N ratio and intersample peaks.  Or even the fact that you might be stressing your converters by feeding them a hot signal.  So you're basically Limiting on input.  With 24 bit, there's no need to push things at all.  We have an amazing noise floor that goes all the way down to -144dB, so why not take advantage of that?  Say you're peaking at -12dBFS, that's still gives you 132dB to work with.  OK, you're right, those are theoretical numbers.  I'll give you some real numbers and use my interface as an example.  The FF400 records about 18.6 bits of clean audio before there is noise in the signal (when set to 24 bits).  That's a noise floor of around 112dB.  If we set our recordings to peak at -12dBFS then that still gives us 100dB to work with, and we're only trading off 2 bits. 

So it is a trade off.  You trade off a few bits to get cleaner recordings that'll sound better and avoid clipping.  At least to me that makes sense.


Take care! :-)

"This yields a clean recording with no coloration.  Pushing these levels a little ususally gives the sound a nice color.  But you have to realize that, by doing this, you're trading optimal S/N ratio with coloration."


We have had this very conversation before.

With all due respects, I absolutely disagree with your premise that professional preamps are introducing distortion as you describe.

Let's assume 0vU equals +4dBu. The preamps may begin to seem "colorful" at +18dBu and they don't clip until about +26dBu.

It is certainly possible for any of us to manipulate the tone in a multi stage preamp. I CAN get Lots of color way below 0dBu if I wish. Drive the input. Saturate the transformer. Open uop the first gain stage. etc.

That does not mean a pro preamp can't easily put out clean power beyond +4dBu. They can and do.

With all due respects, your premise relies on

1) A idealized myth that at 0vU a audio circuit displays "flat" (implying uncolored, "honest" reproduction) response. The circuits' are by strict definition are ALWAYS colored, they are never flat. Maybe sort of. No. Not really. Having said that... the point is a good preamp doesn't suddenly change it's color characteristics just because you opened it's output a few dB.

2) The operator is incompetent with regards to gain staging before and inside the preamp. This can be very hard to define. Maybe the operator and the musician want it to sound distorted.?

and

3) The audio equipment has been been sourced from the "industrial grade" segment of 1970's production audio> Back then the 0vU = +4dBu rating was often skewed by the fact that the circuits only had about 6dB of headroom. Meaning that it clips at +10dBu rather than the +26dBu that is so common in pro gear. 16 extra DB is a "whole bunch".

You might enjoy investigating power supplies and how compromises in power distribution in mass produced prosumer gear is the number one source of grainy nasty sound.

If there's enough power in the preamp then both the best of the old, and the latest stuff  today runs with loads of headroom.

If there are analog headroom problems that need improvement it is in the AD/DA appliances. Heck, that's why the big rack mounted two channel units sound great. They have a proper power supply.

Recording at what I consider suitable levels ( I prefer that to saying "hot" like you guys do ) is not about S::N. That would be silly. What noise?

For me it's about data. It's about having more data to run through filters and its about archiving data for use with anticipated improvements in delivery mediums.

If you think you are recording too hot... then you are. Don't record to hot. Record at a suitable level... what ever that means :-)

It seems simple.

Peace and LOVE!
mike

I have asked for clarification about intersample peaks on many occasions.

The way I understand them is that they exist for exactly less than 1 sample. So that means... if I could hear at 44.1kHz I might be able to notice a intersample peak... if I was really paying attention.

It's all academic... because I don't get near digital clipping in routine... I'm just curious.

Here's an illustration I made a while ago. I'm still asking for suggestions for corrections etc. if I have the concept wrong.

The term intersample peak sounds so big time until you start thinking... gee, what's a portion of one sample sound like?



best,
mike

Hey Mike,
I'm a little confused about intersample peaks as well. I kind of assumed they are a form of overshoot in the analog side of some D/A converters and the way they smooth the staircase nature of digital in to an anolog-ish output via filtering. From what I've read, it's something that doesn't impact all D/A circuits, but it potentially could impact some of the less expensive or older circuits. I have seen some of the explanations go into flaws in the design of the filtering methods used in some D/As that create the problem. I've also heard that well designed D/As should have enough headroom so that these inter-sample peaks don't necessarily cause a problem.

I guess reagardless of the cause, leaving a bit of headroom in the mix isn't going to be perceived advesely by most listeners, and might make some listeners have a better listening experience depending on their playback equipment.

Jose when you say:

"As you know, when your analog gear reads 0dBVU, it means that the signal passing through is at its nominal level (or the level at which it provides the best S/N ratio).  This yeilds a clean recording with no coloration.  Pushing these levels a little ususally gives the sound a nice color.  But you have to realize that, by doing this, you're trading optimal S/N ratio with coloration."

What are you saying?

I'm saying you've got a beefy range of output where the basic "color" tonality of the circuit remains stable and that the range extends well beyond the nominal levels you suggest (I think that what's you're saying) as a limit.

As far as S::N in the preamp... it may be a fact that using the preamp to produce more gain creates some noise. I find that the levels are imperceptable or at least neglible when working with most sources.

If indeed the track is turned down during the mix then it becomes even less of a problem.

If however the noise is ambience from the enviroment heard by the mic... then having more data to work with FFT noise reduction is advantageous and once again you will probably mix down the levels so both the ambient noise and the artifacts from filtering (all filtering actually... EQ, comp etc.) will be minimized.

Maybe I didn't understand what you said.

I'll go read about intersample peaks. the last time I did I concluded it was problem on SSL desks designed in the 1990's and rarely encountered elsewhere. And I asked the same question I asked above but never found anyone who would entertain it.

How long does an intersample peak exist? Is it some portion of a sample?

best regards,
mike

mike_mccue


...

How long IS an intersample peak?

best,
mike

An Intersample Peak is a point - no length.

Mike,

Read the whole thing.  All your questions are answered in there.

Hey Mike,

The article also explains why these Intersample Peaks occur.  You'll find this at the bottom paragraph on page 4 and ends at the top paragraph on page 5.  Basically, what they're saying is that the way digital sampling works doesn't provide the whole picture of the waveform because samples are being taken every 1/44,100 of a second, assuming we're recording at 44.1KHz.  This is why we get a staircase-looking waveform when we zoom in on an audio clip inside of Sonar.  The rest of the waveform is being filled in at the output stage when it goes through the D/A process.  So this means that a waveform can be shown as legal (below clipping) inside our DAWs, yet be a different story when the actual waveform is reproduced in the analog world.

What you see in the picture is not a squared wave (nor one that suffered limiting).  It is the digital representation of a complete waveform.  Perhaps this is what's confusing you.  That's what the Nyquist Theorem explains, which is why a lot of people still believe that recording at higher sampling rates gives them more resolution because there are more sample points available to reconstruct a waveform.  This is simply not true.  But I digress.

I guess this could be easily explained with images, but I honestly don't feel like posting them :-P  However, if my explanation was not enough, then I'll try harder.  Just let me know as I don't mind doing it.


Take care!

Jose, this is mildly exasperating.

Does this mean you choose not to acknowledge that not only have I read the bottom paragraph on page 4 but that I quoted it and commented with the response "OK, so that's all sort of obvious" in a previous post.

You seem to think I can not comprehend what they are saying :-)

That... I don't get. :-)


Think about this for a long time.... it's an intersample peak... the result of ambiguity... or it's an over.

If the samples are at 44.1kHz. What peaks can sneak thru the aliasing?

I'd love for you to try drawing an intersample peak.

When you do you will be confronted with the very thoughts I had to deal with when I considered how to create my illustration.

It's one thing to think *I'll draw up a waveform and it will make sense* It's another thing when you try to draw it to scale with a sinusoidal curve and realize you either have to get the peak to fit between two samples or you have an obvious over in the subsequent sample.

And that is my point.

If it's an over it's an over. No big deal... at least it's not some hard to understand concept.

That very first sample might have become an intersample peak. And it existed for the length of the sample.

Please take a moment to draw.

I was taught many years ago that Albrecht Drurer once said "to draw is to see".

And what is up with all the sticking tongue out stuff? It's bad enough that you seem to think I can't read. :-)

Mike,

The first thing you talk about is the Anti-aliasing filter which has nothing to do with Intersample Peaks.  There's also no need for a Limiter to have Intersample Peaks.  On the contrary, Limiters like Ozone have built-in features to prevent them.  All that's needed is a hot enough signal, that's it.  This signal won't show as an over unless you have something like the SSL meter to tell ya.  The DAW meters won't show you Intersample Peaks because it uses a limited amount of samples to represent a waveform.  The complete waveform is only heard, not seen which is what makes Intersample Peaks dangerous.  Actually, what makes them dangerous is how they can affect playback on certain systems.  Lastly, Intersample Peaks are only an issue during Tracking and Mastering.