by Paul Frindle - I

Передовой инженер цифровой записи проясняет мозг пользователям.

Originally Posted by Paul Frindle

Ok here goes, my second (somewhat attenuated and bleary) attempt to answer these very good points. After dinner and a half bottle of wine with notepad at the ready - cos I'll be logged out again when I finish. Let's hope I don't bugger this one up :-)

You are making some very good points here that deserve a proper answer, because they hit right at the heart of the matter for most people - thanks for raising this.



In fact many of the most creative engineers I worked around in the 70's and 80's knew absolutely zilch about what was going on technically in their systems. But this did not stop them creating whole new fashions of production and defining new eras in our art. These people are my friends and I have always had the greatest respect for them. Suggesting that we should be deprived of their contribution based on their lack of technical knowledge and being 'clueless' would have been a crime to our art and a terrible loss to our industry. No block diagram was needed - and no block diagram would have done anything other than put these people off doing what they did best - which was to create art.

I totally agree that what we have now is a zoo of confusion, largely caused by a combination of aggressive consumer style marketing and a naivety that has plagued this new technology from the outset, which somehow started on the wrong foot and has never been able to admit it. I agree entirely that the industry is awash with misinformation and confusion. Much of this is generated by marketing initiatives and people 're-employing' words and sound-bites in a totally inappropriate manner (the stock in trade of consumer marketing).

But since these concepts are so compelling (because that is what they are designed to be) to anyone armed with simple common sense and just an honest desire to learn stuff, they are incredibly difficult to unseat - even if they are totally inappropriate and quite misleading. You only have to witness just how many posts are still running on about 'resolution', as though a digital audio signal were something like a PC monitor or a digital photo, to see how such ideas stick so effectively. :-(

Added to this, we have developed a production style based on maximum loudness which has evolved at the behest of yet more industry marketing perceptions, that is almost guaranteed to produce 2 dimensional results lacking in dynamic range and punch - and operating levels based on clipping only which will almost certainly result in errors.

No wonder the users are confused - hell, despite being in this industry as a designer for 35 years or so - I am confused! Each time someone asks a question about this stuff I am forced to think it through and work it out again for myself. What chance does a typical user have - knowing this stuff is not even their profession?!

Ok, to illustrate the point let's just go through your mentioned situation facing users about whether or not a fader can be used to recover from overs, on say a typical professional Pro Tools rig. This is what happens:

1. The signal comes into PT from an ADC as fixed point data. Fixed point can only represent a max data value of (nearly) +/-1.0 - but this is ok because and ADC can only produce max data of +/-1.0 - so no clipping here..

2. If there are no plugs in the channel and the signal gets straight to the mix buss; the PT processor has a 24bit native data width and a 48bit accumulator with 8 bits of overflow - so no trouble summing here either. You can add things at your heart's content up to very high levels and bring it back down with the master fader without clipping.

Earlier systems however needed to break out back into 24bt domain to communicate with other processors - when the mixer got bigger than what would fit in one IC, and so there could be clipping under some bussing and routing conditions for data getting bigger than +/-1.0. But we'll ignore that for now... It's history.

3. If you insert an RTAS plug-in in the channel, it runs on the host processor (your MAC) in floating point (because all general purpose processors are FP). The data is converted to float and processed in float - and converted back to 24bit fixed into the PT mixer. So clipping can occur here for values greater than 0dBfs. Can you recover from this by reducing the fader? No.

4. If you follow the RTAS plug-in with another RTAS plug they will both run in the host floating point processor and can pass floating point between them. So no clipping happens out of the first plug however large the data gets - and providing the second plug brings the signal back down to below 0dBfs it will be ok. Can you recover this with a fader? Yes.

5. However if you place a TDM plug-in in the channel, data passing in and out of it is fixed point 24bits - because that's the TDM buss format. If the process (say an EQ) increases the level it will clip - regardless of whether the next plug is RTAS - and even if the next plug reduces the level below 0dBfs. Can you recover from this with the faders? No..

6. If you pass data into an external helper processor (like Powercore) the data will be passed at 24bit fixed point (because of the interfaces) and can clip - regardless of whether the external processor is running float or not. Can you recover from this? No.

If your pass your data to any digital external outboard gear the same will happen because the interfaces are 24bit fixed point.

7. Ok so you've got your rough mix together understood all this stuff and got all the red lights out. Are you out of trouble? Not at all. Your meters are reading sample values only - and not signal at all..

You decide to put an EQ into a channel somewhere. If this is an oversampling EQ design it partially decodes intersample peaks (those that exist at higher signal rates between samples that approach max). Now parts of this track can peak at +3dB (usually, but can be higher) - even though it's set to flat and no loudness increase has occurred - you haven't even started to EQ yet!

If the this is an RTAS plug and the next one is also and RTAS plug that reduces the level again you are ok - you can recover from it.

If the plug is TDM you have had it. If the plug drives anything external whether fixed or float, you've had it. Can you recover from this? No - the only thing to do is to reduce the gain of the track into the plug and re-do your whole mix to compensate for the lower contribution of this track :-(

8. Of course other things can result in bigger than 0dBFS samples. Looking at the simple EQ, boosting any freq range will produce bigger values that may clip. You can indeed have multiple bands some in boost and some in cut that give a valid signal out. An RTAS plug should cope, but what if the plug is TDM fixed point? Are the values clipping internally? Have they got it right inside there, what is the order of the sections? DO you really know and how can you tell? Are you sure you can recover from it with a fader down line? No.

Even cutting can produce larger peak values (up to +6dBFS for heavily saturated sounds) as you are changing the waveform shape. Can you recover from this without destroying your balance? No..

9. Ok so you have avoided all this stuff by operating in RTAS or host only - no TDM and no outboard digital processors or add-on cards in sight:-). The occasional red light doesn't matter? Surely you safe now? Not a bit of it :-(

You introduce a compressor (no matter if TDM or RTAS) - a process that has to have a knowledge of real world signal level in order to function. If there is any deliberate non-linearity in this process it will be based on some operating level (most usually flat out as this IS currently the only operating level for digital audio).
If you hit something like this with signal at +10dBFS it may sound like hell :-( Can you recover from this without destroying you balance - or accepting inappropriate sound qualities from the plug-in? No..

10. Ok so finally you've understood all the above and all the possible pitfalls and you've got you final mix together, no red lights are on and it sounds great.

You stick in your fave buss comp/limiter to get the required level and punch and you are done - yes? No not a bit of it :-(

Your meters are still reading sample data values and not signal itself. The comp has bought up parts of the sound that was quiet in the recording adding some nice sounding distortion and the limiter has rounded off some of the peaks producing a greater number of them at higher values.

You are now back in intersample peak land - when your data hits your DAC it is reconstructed into signal and in parts produces levels up to +3dBfs (even as much as +5dBfs for very heavy program with lots of HF). What will your DAC do? What will the user's DAC do? What will happen if it's coded MP3, AAC or another lossy coding format? Will it clip, will it fold over with 'splats'? How can you know if you are hearing what the customer will hear?

I could go on and on but this is getting tedious. Do I think I have covered all major combinations? No not in the least - how can I know all possible combinations of these things - I am only a designer of the stuff which I have designed?

So what about the users - how are they ever to know? How can they ever be other than 'clueless' by definition.

Do you think a block diagram of anything less than a whole book would help you out of this?

So whilst people are struggling away with all this stuff wondering this and fearing that, another engineer has simply given up even trying to understand this shit. He hasn't the time or inclination to figure out what combination of this or that, in this or that order, under these or those conditions, may or may not produce an error he might or might not hear - which his customers may or may not suffer from.

He's put aside the dire fears about the much-dreaded concept of 'resolution' - and simply trims the gain down at the head of every track - to produce something like the headroom analogue system enjoyed (for good reason) - and gets on the with the business of producing art - which is after all, what he's paid to do :-)

This is the issue IMVHO :-)