Quote: |
I am extremely intrigued by your comments. I am trying to reconcile your recommendation with a statement made by digidesign engineering. They claim that it is impossible to clip the internal mix bus. With the maximum number of tracks possible in Pro Tools, all with coherent signals at +6 db on each fader, it is impossible to clip the internal mix bus. Yes it is possible to clip the output, but not the internal mix bus. To avoid clipping the output, they say you can simply lower the gain on the master fader however much needed to avoid the problem. Your statement and digidesign engineering's statement may be completely unrelated, so forgive me if I am linking them together inappropriately. Can you comment?
|
Yes they are right - the mix summing has sufficient range to add a great many channels at full level without saturating. But I am not talking about a summing issue as such. I am talking about what may occur in the signal domain if you add lots of hot contributions together where either clipping or significant processing may occur within individual tracks - and you try to modulate at full output from the mix using sample value metering supplied with the DAW. And of course comparing this with a chain that reconstructs every channel before you mix them - i.e. OTB etc..
Quote: |
Your experiment recommends never peaking above -6dbr, even after any final limiting. Are you saying it's impossible to ever bring a final mix up to 0dbr without adversely affecting the sound quality?
|
No - and that's the important point, it IS possible to have a full scale sample value signal that is legal.
For instance in my experiment, had the noise generator (set at -6dBr) been followed by a brickwall filter (like a DAC) it would have given a legal SIGNAL output at full level.
Similarly if an ADC (if properly designed with a good filter) is driven to it's max output modulation, this is also a legal signal that a good DAC (with a good filter) will decode correctly. (However this does rather expose those guy's who propose that the ADCs and DACs should have minimal filters etc..)
SO if you use the system as a straight recorder (with well designed ADCs and DACs) it shouldn't be possible to produce an error of this kind and the sample value metering - although not perfect - will give an acceptable indication of peak programme level.
Quote: |
If I put a limiter on a master fader in Pro Tools, the digital summing has already occurred at 48 bits then been dithered to 24 bits before it even hits the limiter plug-in. If I were to sum my mix, never peaking above -6dbr at any stage before hitting the limiter plug-in, then bringing the final level up to 0dbr using the gain on the limiter, would this negate your experiment?
|
This is an interesting question - and complex to answer in one go because there are 3 situations working together.
Firstly - providing that all the contributions in your mix were entirely legal at every stage, raising the gain to peak levels at the mix output should not result in an error in itself. But in practice it's risky since any contribution that gets processed after the ADC recording that introduces phase shifts, non-linearities or accentuates distortions that existed in the recording at upper mids or HF could result in a reconstruction error at higher contribution gains. In other words raising the levels of a 'troubled signal' may push it into the reconstruction error zone, where previously it was admissible.
Imagine for instance a loudish instrument with rich HF percussive harmonic content that for one reason or another only just reached peak values at the output of your ADC in record. You then EQ it a bit (perhaps rolling off the HF a bit) changing the relative phases of the freqs in the spectrum, noticing that the peak sample value level has dropped a dB or so, you increase the gain to max once more. The drop in peak sample value resulting from a slight re-arrangement of the phase of the freqs - may still have resulted in an almost flat out signal when reconstructed - before you added the gain - now it could overload even though no red light is on.
Secondly we need to consider HOW the limiter acts on the signal. For instance it is possible (even likely) for the fast peak limitting which is popular today to produce it's own illegal signal by modulating the sample values quickly. Imagine for instance a pure sinewave that has had it's peaks reshaped by the fast action of the limiter - this is harmonic distortion which could result in an illegal signal during reconstruction if freq are high enough, despite never producing full level sample values. There are ways of preventing this but many applications do not include them.
A third and more interesting thing to think about is how the limiter sidechain will respond to the levels of the signal. If we go back to my experiment with the noise genny and the filter; we can see that filtering the noise samples produces nearly twice the peak sample value for the same apparent signal level. Now if you follow this with a limiter or compressor that has a sidechain that measures sample values (i.e. acting like your meters) how will it respond in either case if set with a threshold of -6dBr? Well with the filter out it won't compress or limit at all since no values get bigger than -6dBr. However with the filter in it will compress and reduce the reconstructed output to -6dBr again. I.e. it will compress by 6dBr - and the audible result of this will be to drop the signal level by 6dB. The presence of the filter severely reduces the total loudness you can obtain from the limited signal (and this is another story for another day). Try it - it's a real eye opener
Quote: |
How could a DAW application be designed to eliminate the problem you point out? Would some kind of an internal reconstruction filter after every track and process be required? Is the problem apparent only in DAW's or does it show up in any digital mixer?
|
There are several possible ways one might arrange things to avoid this problem within the design of the whole system (particularly if the production culture was more sensible). But we should appreciate by now that making a mixing app of great quality is not simply a question of providing something that 'adds samples together correctly'. And of course this fact blatantly exposes the fallacy of those who try to compare the quality of mixer apps by setting them up identically and taking the results of one away from the other and measuring the differences!!
As with all pro-audio design at least 50% of the design effort is about how you present the information to the user and the nuances of control you give him over it. And of course the cost of the system is important wrt the quality the user expects to obtain and the resilience of the system under duress and not forgetting how it performs within the popular production culture of the times. This subject is too big to discuss here in great detail, but it can be seen that there are largely hidden performance issues regarding digital systems driven to full metered levels that exist at multiple levels, from the quality of the ADCS and DACs (filtering in this case), headroom within the application at the interfaces, plug-in process quality, metering style etc etc.. It is likely that some combinations of system may sound different from others when faced with high sample levels. It is definitely likely that the users' CD players will vary in the artefacts they produce - and this is perhaps the most worrying aspect of it all. In tests I have done most popular CDs produce reconstruction errors at around 2dB or so somewhere in the duration of the production - and paradoxically these are not necessarily the loudest or harshest styles of music. Many of the worst I have are actually (intentionally) clean crisp sounding jazz style CDs. The most often offending programme includes percussion - cymbals, bells, tamborines etc and highly Eq'd (and intentionally clipped) female vocals. Remember that intentionally clipping sounds to produce bite, attack and 'definition' is commonplace in certain quarters of artistic production. How much more clean and crisp would they have actually sounded if mixed and mastered just 3dB less hot
The simplest practical advice right now (with the kit you are currently compelled to use - and if your paymasters will let you) is to think of sample value levels in the green section of the meter as always legitimate (i.e. repeatable at destination). Those in the yellow area (-6dBr and -3dBr) are most probably ok, but caution should be taken as they're definitely big enough and may just cause reconstruction errors if clipped or intentionally distorted in the digital domain or digitally recorded from an artificial source. Levels between -3dBr and 0dBr are dangerous and those that actually reach the red light are almost certainly broken signals and are very likely to degrade in various ways at the destination DAC - in both your's and the end user's!! And above all - don't assume that any meter anywhere within the system indicates a legitimate signal by dint of it not hitting a red light
And to get back to the original subject and my original reason for posting - be aware that by mixing OTB in analogue and encoding to digits via an ADC afterwards - you are removing the possibility of making reconstruction overs in your master. This is very significant within an industry environment where everyone is currently aiming for absolute max loudness and modulation.