Q for Paul Frindle

[Paul Frindle]

Gotcha. Reconstruction errors are only on the D to A process. Thanks Paul for clearing things up, including the dither on the trim, I was wondering about that too.

Not necessarily. Some processes may have the equivalent of reconstruction within the applications themselves. They may therefore spit out larger value samples than those that went in - without any increase in volume or audible change in the programme!! This was one of the things I demonstrated on the PSW thread.

However since these increases in sample values should show up on the DAW meters you can reduce levels to avoid them as they are not hidden from you - like they may be in a DAC situation.

BUT - this could still mean that in very bad cases programme that would otherwise pass legitimately (to be reduced further downline in the mix) would now need to be reduced immediately to prevent clipping - maybe by as much as 3dB. In the really contrived demo I did on PSW using the PT oscillator set to white noise, the reduction required was nearly 6dB..

The offending processes are likely to be those that upsample the signal - do the effect - then downsample it again at the output. The downsampler is effectively a partial reconstruction filter - that's why it behaves that way. So even with the effect turned off, the up and down sampling can still produce this effect.. No Oxford plug-in employs upsampling/downsampling in the signal path you get to hear..

Is there a specific reason you shouldn't use the input on the first plugin at -6 if it's a double precision, high quality eq plug like sony or URS?

No not at all - reducing level in this way is perfectly legitimate also. It's just that you need to keep an eye on things if you start moving processes around later. For instance if you decide to put the compressor before the EQ etc. don't forget that the EQ was giving you your headroom :-)

Personally I prefer to deal with it as a separate matter so I don't have to worry about stuff like this when I'm trying to concentrate on sounds and artistic aspects of the mix.

[innesireinar]

Originally Posted by bob katz

My goodness. It's a moving target. At the first writing of that article, low resolution digital (ADATs) abounded, and there was no question that recording and mixing to analog tape via analog was the best solution. Today, 9624 converters abound. Are they better than analog tape? It's come to the point where it's more a matter of taste and flavor, since the flavors are all a bit more pleasant than the ADAT and DAT and 16-bit days.

The updated 2005 brought it up to 2005, for sure. I'd have to re-read it to see if there's anything I regret. I probably added a lot more "probablys" and added some caveats about all-digital processing with cheap plugins and the need to oversample non-linear processes and such. Was there anything about the 2005 version that struck you as out of date, inaccurate, or problematic?

BK

It started last week when I and a friend OM had one of the boring discussion analog/digital. After hours of discussion where everyone has remain in his position I promised to link him the Digido site. After two days he called me saying "your guru (you) is stating that analog is better" - And I "maybe this article is old" - "no, is updated 2005".
I went to read the article and I've only stated a discrepancy about your words between the PSW thread and the article, because on PSW you said that analog is far from perfect, works that you have received in your mastering studio done in digital, with the add of good analog outboard sounded better than those done mixed on the desk.
Therefore I'd like to have a your definitively point about that.

[innesireinar]

Originally Posted by Paul Frindle

Thats actually a deeper question than perhaps you mean to ask, as I have witnessed 'beliefs' of +ve for RTAS (as they have the fabled 'float math some people believe is somehow 'better') and -ve (because they are easier to make and can run on the cheaper PTLE).

But in honesty we have gone to great lengths to ensure that they are exactly and precisely the same as far as the user is concerned - even to the point of adding stuff to force them to be the same. The reason for this is that either something is working as intended or it's not; if the RTAS were different from the TDM - one of them would be wrong and I just couldn't live with that!
This would also be highly confusing and unfair to the user and not in the least the kind of uncertainty one would expect from a high end professional application used by people whose reputations are on the line. I couldn't live with that either :-(

Thank you. Clarified.

[Paul Frindle]

Originally Posted by innesireinar

It started last week when I and a friend OM had one of the boring discussion analog/digital. After hours of discussion where everyone has remain in his position I promised to link him the Digido site. After two days he called me saying "your guru (you) is stating that analog is better" - And I "maybe this article is old" - "no, is updated 2005".
I went to read the article and I've only stated a discrepancy about your words between the PSW thread and the article, because on PSW you said that analog is far from perfect, works that you have received in your mastering studio done in digital, with the add of good analog outboard sounded better than those done mixed on the desk.
Therefore I'd like to have a your definitively point about that.

I have a lot of sympathy with this kind of thing. The problem with books is that they persist in a world that is ever-changing. There's a further problem if one tries to write pragmatically about 'what is the case at this time' rather than 'what should or could be the case if things were done correctly in future'.

If Bob has written what he honestly believes is correct and pertinent at that time, that's all one can expect. Demanding more than this would be a little unfair IMHO. All writing involves an element of opinion - and opinions change quite naturally - and so they should because having an open mind is the essence of advance in the art!

The first time I witnessed digital audio working correctly (i.e. transparently from a sonic point of view) and understood the issues was after we'd had the honour of actually making the whole thing ourselves from first principles (at least a 7 year research and design effort). Very few people have had this luxury and I will probably not get it again myself! In this time we learnt a mountain of knowledge. If someone had asked me to write a book comparing analogue to digital in 1987 it would have been a very different story to what I would write in 1997!!

And much of what we write about even now is actually new knowledge - although we may have understood this stuff theoretically for ages, the full importance of it may have only surfaced in the past few years. And I can say from a personal perspective even as a designer - there are still issues raised by users (even on these forums) that often cause me to think again about something I thought I understood already :-) People are still discussing and researching issues where their importance is still contested and generally unknown at this time. We can never know everything and no one is infallible - anyone who thinks they are is IMVHO incapable of advancing things further!

Most readers want fairly understandable and simplified read that sets out stuff in a way that is useful at the time it's written. There's always a dilemma between providing up to the minute info which is useful to people right now - and being too general with fine technical and theoretical detail about 'what should be' that turns people off..

Every time I post on a forum I am aware of this balance - some people would like more detail, others would just like to cut to the chase and have a list of bullet points that make sense straight away.. For instance - even now all this discussion and hoo-ha about peak levels that mess people up at this time is something that shouldn't happen if systems were designed correctly. However, how many people would want to drag through pages of detail as to how I would design this problem out of a system that I would make (if given the chance), if that doesn't actually exist - and right now people are sitting in front of something that puts on red lights inexplicably or sounds strangely bad, despite showing no red lights at all?!

[innesireinar]

Reading the Bob's book I've discovered that the nowaday 24bit are not true 24 bit but 19/20. Is it an converters issue only? And if so, why conv. houses don't make converters with more bit in order to have a real 24 bit recording?
28bit? 32bit? It could be possible?

[Paul Frindle]

Originally Posted by innesireinar

Reading the Bob's book I've discovered that the nowaday 24bit are not true 24 bit but 19/20. Is it an converters issue only? And if so, why conv. houses don't make converters with more bit in order to have a real 24 bit recording?
28bit? 32bit? It could be possible?

24bits is the output wordlength, but the actual signal to noise performance is obviously less than that. Typically these days SNR values of around 110dB are possible - which is still very good.

Watch out for the published figures (they all of course use) that are 'A weighted', as this filters out the extreme HF and can show performance of as much as 10dBs better than the true flat response figures. IMVHO this is a con, pure and simple :-(

Making a converter with a true standing SNR of better than around 124dB is virtually impossible because of the noise limitations of components at room temperature. For this same reason it will never be possible to make a converter with a true 24bit SNR of 143dBs.

[innesireinar]

But improoving the "true" resolution not only for the SNR but for having a more dense grid of amplitude, for having more and smaller steps in the Y axis. As close as possible to analog wave.

[Paul Frindle]

Originally Posted by innesireinar

But improoving the "true" resolution not only for the SNR but for having a more dense grid of amplitude, for having more and smaller steps in the Y axis. As close as possible to analog wave.

Now that is one of the biggest misconceptions that plagues our whole industry and probably the cause of more misunderstanding than ANY other issue :-(

There is no such thing as resolution - it's a complete myth - please forget anyone ever used the word! The mathematical precision dictates the signal to noise ratio - NOT - the distortion or purity of the signal.

In order to make any progress in understanding how digital audio actually works this is the very first lesson people must learn :-)

[innesireinar]

Can you explain why my thesis doesn't work?
I've always thought that more steps in the two axsis= more close to analog

[Paul Frindle]

Originally Posted by innesireinar

Can you explain why my thesis doesn't work?
I've always thought that more steps in the two axsis= more close to analog

Oh heavens, you are coming up with some questions. I have thrashed this one out more times and in more threads than I could even hope remember. One such thread was the so called '96K' thread on PSW which even inspired one person to write a book! Let me try as briefly as possible:

1. We know that natural signals are not quantised in quantity or time. We humans chose to measure amounts in quantities (discrete numbers) and time in quantities - but this is a human convenience - it has nothing to do with the natural world. We invented 'counting' NOT nature.. Anything that we measure and count is therefore an approximation of reality - NOT the reality itself.. Something like a circle has an infinite number of 'points of measurement' around its circumference and no finite precision of math can completely accurately describe it.

2. Therefore the very act (and concept) of 'measurement' and 'counting' is a source of error - cos it can never be completely accurate - simply because it's discrete and therefore 'stepped' in a way that nature is not.

3. So whilst it is true that the more precision you use (and the smaller the steps become) the more accurately you will describe an event - no amount of counting to ever higher degrees of accuracy (short of an infinite) will give you a completely accurate representation of a natural event.. So increasing 'resolution' to banish distortion is a lost cause - as some will remain whatever you do and however much data you waste on it!

4. Ok - now if we go back to the situation of the audio signal, it's notionally a continuum (like the circle). But the quantisation steps we introduce because of our finite 'measurement precision' turn up as harmonics in the signal (unwanted freqs) that were not there in the original signal. In other words - what makes it impure is that these harmonics are caused by regular discrete measurement. If they were irregular the errors would be random noise only - and not harmonics. We would then consider it a 'pure signal'.

5. The way we avoid this harmonic error (due to regular measurement) is to turn it into random noise by adding statistically random values to the signal so that the quantisation value steps are no longer the same every time - i.e. the steps are blurred out. This process is called dithering.

6. What we are then left with is a signal that has no harmonic error (no unwanted tones - just the signal) with some added random noise due to the dither blurring the measurement steps.
This is the equivalent of an analogue signal passed through an unquantised system, that has a finite signal to noise ratio - i.e. just like the real world around us :-)

7. The amount of noise we get from this is proportional to the size of the steps we are blurring out with the dither. This is of course proportional to the math precision we are using to quantise the signal in the first place (our chosen accuracy of measurement). The less the bit width - the greater the noise required to blur the steps out completely.

So we can calculate the error content by knowing the step size and therefore how much noise we will end up with. So for instance for a 16 bit signal the total number of steps from end to end is 2^16 = 65536.

This means that each step is 1/65536 of the total.

If the total is called '0dB' then the error of the steps is 20*log(1/65536) = -96.33dB.

Smoothing out the steps with noise will generally cost us another 3dB, so the total SNR will be around 93dB total :-)

In other words a perfect signal (without any steps or distortion) with some random noise at -93dB below..


We can do the same thing for any bit width; For instance an 8 bit quantisation, it has 256 steps, each step is 1/256 wrt max 0dB.

So, 20 log(1/256) = -48.16dB. So when we dither it thats a perfect signal without steps or distortion with a noise floor of around -45dB below..


I've done some illustrations of this stuff as part of the limiter manual discussing different dither type options:

http://www.sonyoxford.co.uk/pub/plug...ail.htm#dither

Ok - so now where does the perception of 'resolution' come from? Well all day long people are looking at editing screens that show DISCRETE sample values - because they MUST do - because this is ALL that there is to see within data in your system. Everyone therefore assumes that the signal will end up sounding like it looks on the screen.

However these sample values are NOT signal until decoded - the way it looks is a totally invalid representation of what it will sound like.

There is no such thing as 'resolution' in the sound - the term has no theoretical validity. 'High Resolution Audio' is simply a marketing term - that would only make any sense if their systems were busted anyway:-(

I hope this helps :-)

BTW this does not deal with timing periods and sampling rates - thats another post..

[Nordenstam]

A quick addition to Frindles post;

The samples are not quantized along the time line, the X axis. The sample clock is steady. The signals are not moved back and forth to meet their nearest mathematical value. This realization helped me a lot in understanding digital!

The sample rate is an can/can't matter, black or white. It's either able to reproduce signal with full accuracy below half the sample rate, or not able to reproduce them, if the frequency content is above half the sample rate.


PS: Lavrys introduction to sampling at lavryengineering.com/support is a great read for anyone interested

[Paul Frindle]

Originally Posted by Lupo

A quick addition to Frindles post;

The samples are not quantized along the time line, the X axis. The sample clock is steady. The signals are not moved back and forth to meet their nearest mathematical value. This realization helped me a lot in understanding digital!

The sample rate is an can/can't matter, black or white. It's either able to reproduce signal with full accuracy below half the sample rate, or not able to reproduce them, if the frequency content is above half the sample rate.


PS: Lavrys introduction to sampling at lavryengineering.com/support is a great read for anyone interested

Yes that's right. The discrete sampling times do not cause errors because the reconstruction filter on your DAC removes the artefacts. For the signal that comes out of your DAC there is no quantisation of phase or time (providing it's working correctly and well designed).

However - this also means that the samples you see in your edit window are not an accurate reflection of the signal in the real world.. And this also means that you can make stuff in your edit window that will not be decoded the way it looks on the screen..

[bob katz]

WEll, I'm going to have to look at that "ever-changing" and evolving article again. The original purpose of it was to help people understand the tradeoffs between analog recording and "cheap digital". The problem is that today, in 2006, the performance of cheap digital is far better than the best digital was in (pick your date).

I just restored a Mitsibushi X-80 (48 kHz/16 bit) recording from 1973 (if I recall correctly). It sounded horrible, harsh, bright. I had to put it through so much analog gear and "sweetening" and manipulation to make it sound acceptable it's amazing that anyone ever liked this medium. And this was "state of the art digital" for that year.

I agree that the techniques and the outboard gear that was used to make the Mitsubishi recording had a lot to do with the sound that I heard. And that we had to learn to deal with a medium that was more linear than any analog tape we had ever dealt with. However, objectively speaking, the digital technology of that time was subject to measurable problems (jitter, monotonicity, low level resolution, phase shift, etc.) that you will not find in even the cheapest audo A/D converter made today.

Fast forward to today, and through the use of better chips and better technology, the bar has been raised.

That was the original point of my article, it keeps on flowing and I wager if I rewrite it today, it will be out of date again tomorrow. Should I pull the article from the website until I rewrite it?

BK

[Ashermusic]

Originally Posted by bob katz

shift, etc.)
That was the original point of my article, it keeps on flowing and I wager if I rewrite it today, it will be out of date again tomorrow. Should I pull the article from the website until I rewrite it?

BK

Yes.

[Bob Olhsson]

Originally Posted by bob katz

...I just restored a Mitsibushi X-80 (48 kHz/16 bit) recording from 1973 (if I recall correctly). It sounded horrible, harsh, bright....

In fact they were 15 bit however I've heard recordings made with them that didn't sound that way at all.

[innesireinar]

Originally Posted by bob katz

WEll, I'm going to have to look at that "ever-changing" and evolving article again. The original purpose of it was to help people understand the tradeoffs between analog recording and "cheap digital". The problem is that today, in 2006, the performance of cheap digital is far better than the best digital was in (pick your date).

I just restored a Mitsibushi X-80 (48 kHz/16 bit) recording from 1973 (if I recall correctly). It sounded horrible, harsh, bright. I had to put it through so much analog gear and "sweetening" and manipulation to make it sound acceptable it's amazing that anyone ever liked this medium. And this was "state of the art digital" for that year.

I agree that the techniques and the outboard gear that was used to make the Mitsubishi recording had a lot to do with the sound that I heard. And that we had to learn to deal with a medium that was more linear than any analog tape we had ever dealt with. However, objectively speaking, the digital technology of that time was subject to measurable problems (jitter, monotonicity, low level resolution, phase shift, etc.) that you will not find in even the cheapest audo A/D converter made today.

Fast forward to today, and through the use of better chips and better technology, the bar has been raised.

That was the original point of my article, it keeps on flowing and I wager if I rewrite it today, it will be out of date again tomorrow. Should I pull the article from the website until I rewrite it?

BK

It was only a request of your point of view about this thread since the article and the post on PSW are of the same age, but have two different point of view (IMHO).

In your book you wrote that a digital signal can't go higher than 0 once it has been recorded on a medium and meters detect overs by reading consecutive squared samples at 0. Does this rule work also for signal just converted but not yet stored on a medium? What does mean "headroom" in digital? If a digital signal can't go higher than 0 what the headroom stands for? Is there a place in a DAW where it's possible to have a signal higher than 0dB? Or is it a conversion issue only?

Thank you

[innesireinar]

Originally Posted by Paul Frindle

Oh heavens, you are coming up with some questions. I have thrashed this one out more times and in more threads than I could even hope remember. One such thread was the so called '96K' thread on PSW which even inspired one person to write a book! Let me try as briefly as possible:

1. We know that natural signals are not quantised in quantity or time. We humans chose to measure amounts in quantities (discrete numbers) and time in quantities - but this is a human convenience - it has nothing to do with the natural world. We invented 'counting' NOT nature.. Anything that we measure and count is therefore an approximation of reality - NOT the reality itself.. Something like a circle has an infinite number of 'points of measurement' around its circumference and no finite precision of math can completely accurately describe it.

2. Therefore the very act (and concept) of 'measurement' and 'counting' is a source of error - cos it can never be completely accurate - simply because it's discrete and therefore 'stepped' in a way that nature is not.

3. So whilst it is true that the more precision you use (and the smaller the steps become) the more accurately you will describe an event - no amount of counting to ever higher degrees of accuracy (short of an infinite) will give you a completely accurate representation of a natural event.. So increasing 'resolution' to banish distortion is a lost cause - as some will remain whatever you do and however much data you waste on it!

4. Ok - now if we go back to the situation of the audio signal, it's notionally a continuum (like the circle). But the quantisation steps we introduce because of our finite 'measurement precision' turn up as harmonics in the signal (unwanted freqs) that were not there in the original signal. In other words - what makes it impure is that these harmonics are caused by regular discrete measurement. If they were irregular the errors would be random noise only - and not harmonics. We would then consider it a 'pure signal'.

5. The way we avoid this harmonic error (due to regular measurement) is to turn it into random noise by adding statistically random values to the signal so that the quantisation value steps are no longer the same every time - i.e. the steps are blurred out. This process is called dithering.

6. What we are then left with is a signal that has no harmonic error (no unwanted tones - just the signal) with some added random noise due to the dither blurring the measurement steps.
This is the equivalent of an analogue signal passed through an unquantised system, that has a finite signal to noise ratio - i.e. just like the real world around us :-)

7. The amount of noise we get from this is proportional to the size of the steps we are blurring out with the dither. This is of course proportional to the math precision we are using to quantise the signal in the first place (our chosen accuracy of measurement). The less the bit width - the greater the noise required to blur the steps out completely.

So we can calculate the error content by knowing the step size and therefore how much noise we will end up with. So for instance for a 16 bit signal the total number of steps from end to end is 2^16 = 65536.

This means that each step is 1/65536 of the total.

If the total is called '0dB' then the error of the steps is 20*log(1/65536) = -96.33dB.

Smoothing out the steps with noise will generally cost us another 3dB, so the total SNR will be around 93dB total :-)

In other words a perfect signal (without any steps or distortion) with some random noise at -93dB below..


We can do the same thing for any bit width; For instance an 8 bit quantisation, it has 256 steps, each step is 1/256 wrt max 0dB.

So, 20 log(1/256) = -48.16dB. So when we dither it thats a perfect signal without steps or distortion with a noise floor of around -45dB below..


I've done some illustrations of this stuff as part of the limiter manual discussing different dither type options:

http://www.sonyoxford.co.uk/pub/plug...ail.htm#dither

Ok - so now where does the perception of 'resolution' come from? Well all day long people are looking at editing screens that show DISCRETE sample values - because they MUST do - because this is ALL that there is to see within data in your system. Everyone therefore assumes that the signal will end up sounding like it looks on the screen.

However these sample values are NOT signal until decoded - the way it looks is a totally invalid representation of what it will sound like.

There is no such thing as 'resolution' in the sound - the term has no theoretical validity. 'High Resolution Audio' is simply a marketing term - that would only make any sense if their systems were busted anyway:-(

I hope this helps :-)

BTW this does not deal with timing periods and sampling rates - thats another post..

I'm going to be confused...
Therefore higher bit rate is better for SNR only?
Why today we use 24 bit when recording?
From your words my conclusion is that 24 bit is audible only in some soft passages of some kinds of classical music?

[Paul Frindle]

Originally Posted by innesireinar

I'm going to be confused...
Therefore higher bit rate is better for SNR only?
Why today we use 24 bit when recording?
From your words my conclusion is that 24 bit is audible only in some soft passages of some kinds of classical music?

Higher bit width is better for SNR (not bit rate).

We are using 24 bits to record so that when many tracks are playing at the same time the noise does not get too bad. Remember that each track playing simultaneously will add to the total noise of your output programme. Theoretically doubling the number of tracks increases noise by 3dB.

So;

noise increase = 20*log(number of tracks^0.5)

For instance adding 16 tracks together will increase noise by 12.04dB

The other thing we should be doing is to use the extra SNR to give us headroom by recording/mixing at lower levels. There is a lot of spare SNR beyond what we normally need. My suggestion to reduce the inputs to the mixer by 6dB or is to create headroom..

It's entirely the fault of the user if he uses an operating level that corresponds to clipping. If you used an analogue console at say +24dBu (20dB above standard operating level - pushing up against the power supply limits) you wouldn't expect the best results either :-(

But as far as output media is concerned - much of todays modern music that lacks any dynamic range at all can easily be coded at 8 bits only and you cannot hear the difference. I tried this for fun once on a Green Day disc and had to reduce it below 6 bit coding before I could hear any difference, the noise at -33dBs was never heard amongst the awful racket of the over compressed programme!!

[innesireinar]

Therefore Paul my previous Q do work
"From your words my conclusion is that 24 bit is audible only in some soft passages of some kinds of classical music?"
Can you explain "the headroom" that I asked 3 posts before?
I'm a bit confused.

[Paul Frindle]

Originally Posted by innesireinar

Therefore Paul my previous Q do work
"From your words my conclusion is that 24 bit is audible only in some soft passages of some kinds of classical music?"
Can you explain "the headroom" that I asked 3 posts before?
I'm a bit confused.

The concept of headroom is nothing more than aiming at a level lower than maximum possible modulation. To do this all you need to do is change what the meters read - or in the absence of being able to do that, aim for some lower peak levels..

For instance aiming for peak levels of -6dBfs means you have created 6dB of headroom. Aiming for -10dBfs means you have created 10dB of headroom - it's as simple as that..

In analogue systems is was usual to aim at levels that were -20dB or so lower than maximum possible modulation. This was done mainly for 2 reasons;

1) to make the analogue mixer more flexible (i.e. you could boost things and bring down faders to compensate without clipping). In some live broadcast situations this was so important that 30dBs of headroom was sometimes used.

2) the distortion performance of most analogue kit degrades towards max modulation.

To get this to happen meters were calibrated to correspond to a deliberately (and sensibly) chosen 'operating level' (at some level below max, based on science and experience) and levels bigger than this were referred to as 'operating in the headroom'.
So for instance a typical analogue console would have it's meters calibrated to read 0Vu at say 0dBu (0.775V), and everything above that was in the 'red area' of the meter. The maximum possible output of the console was in fact something in the order of +24dBu - so the console had 24dB of headroom.. Signals in the headroom were not lost, clipped or unrecoverable.

Later on the 0dBu reference was increased to +4dBu in order to reduce noise and take advantage of higher peak outputs (around +28dBu) available on later designs. When this happened the meters were simply recalibrated to read 0Vu when the level was +4dBu, so the level people were aiming at naturally was higher etc.. It was as simple as that :-)

Similarly the actual level on analogue tape was calibrated too in order to get optimum results and some headroom for peaks and overshoot. Analogue tape had a particular issue in that the distortion performance and its abilty to record HF reduced as the level was increased. Therefore the total level you could get onto tape varied depending on what was being recorded (amongst other factors).

The reference level on tape was referred to the magnetisation flux of the tape itself (in nWb/m ) and the level one would aim at depended mostly on recommendation from the tape manufacturer (and some risking from the customer) and was typically around 10dB or so below the max possible level you could record and playback at 1KHz or so.

To achieve the correct operating level (and headroom) one would simply put on a reference test tape with recorded tones (at the recommended level) and;

- adjust the playback meters to read 0Vu.

- You would then adjust the machine's output gain to read 0Vu on the console meters.

- then you would send some 0Vu tone from the console and adjust the tape machine's record gain (whilst watching playback levels) to read 0Vu as well.


Ok - now what you can see from all this is that the complete business of setting operating levels and headroom (even in a complex analogue environment) is ONLY a question of setting gains and calibrating meters, the purpose of which is to get people to aim naturally at the optimum levels for the kit being used :-) Meters were calibrated and visually presented in order to 'encourage' people to modulate optimally.

Ok - in the digital domain (for reasons too long to go into - and avoid a rant - think of it as yet another case of long experience thrown away by the 'inexperienced') the industry has seen fit to set meters to read !00% at digital clipping - also many of the mixing applications have left no headroom within their systems either. This means that everything you do risks being hard clipped - or may cause problems down line in play out systems etc..

All that needs to be done to fix this whole problem is to recalibrate all digital meters so they read 100% and start going into 'the red' at some level lower than clipping (or put the red line further down the scale) - and hopefully sane people would naturally aim for this level instead. Yes amazingly just changing the colour scheme of the meters alone could have saved us from this madness!! But obviously no one is going to do that now at this late stage, with so much pride and history at stake as one of the many 'blunders' of digital audio decisions made decades ago and everyone batting around in panic about the loudness wars with their heads filled full of 'resolution'.

However you do not need to suffer it :-) All you need to do is to make a mental effort to aim for some meter readings below max (imagine some portion of the top of the meter is red, or even mark them with a pen) and try not to go higher than this. You will have chosen a personal operating level that makes sense (having been denied it by the rest of the industry) and provided the system with headroom (the system designers had overlooked).

It's really as embarrassingly simple as that :-)

Try it - I promise you will not be disappointed.


P.s. Apologies - that this get's perilously close to a rant I was trying to avoid.

[innesireinar]

Paul,
I love this kind of threads.
Therefore a plug that doesn't have enough headroom means that its meters are wrongly? In other words by putting the same digital signal through one plug that has more headroom than another one, it means that the first has the meters more "safety"calibrated, but rising the input beyond a certain level both start to clip?

[Paul Frindle]

Originally Posted by innesireinar

Paul, I love you
I love this kind of threads.
Therefore a plug that doesn't have enough headroom means that its meters are wrongly? In other words by putting the same digital signal through one plug that has more headroom than another one, it means that the first has the meters more "safety"calibrated, but rising the input beyond a certain level both start to clip?

Hmm... I don't think I am getting through very well?

The headroom within the plug-in is there so that the internal processing of the plug-in does not saturate (or clip) before maximum output is reached. This is needed in all cases - but a simple case is something like an EQ where you can boost freqs in one band and cut them again in another.

This headroom is internal to the plug, is not visible on meters on the plug GUI and does not affect the metering on the DAW in any way.

However you will notice that input meters on things like the Oxford Limiter (and Inflator) do indeed go above 0dBFS, revealing that the plug-in itself has headroom that you actually use in the normal operation of the plug.. In this case it's visible and the 'normal' max input level as marked on the meter is lower than the max possible input level.

This is an example of what I mean..

[innesireinar]

OK Paul, thank you.
Another Q.
I've started a post on hi end Which monitor controller do you use?
for searching a good monitor controller because mine has some little issue.
Given that you have been the designer of SSL I'd like to turn into analog world by asking you if VCAs can cause degrade of the signal. In fact the main problem of my controller is the discrepancy between the LR channels at low level and maybe this problem could be solved only with digital control and VCAs. But I've heard that VCAs tend to deteriorate the signal. Could be there other factors in a unit like a control monitor which can cause a non clear signal?
If you think that it's better to copy this Q in the dedicated post, do it.
Thank you

[Paul Frindle]

Originally Posted by innesireinar

OK Paul, thank you.
Another Q.
I've started a post on hi end Which monitor controller do you use?
for searching a good monitor controller because mine has some little issue.
Given that you have been the designer of SSL I'd like to turn into analog world by asking you if VCAs can cause degrade of the signal. In fact the main problem of my controller is the discrepancy between the LR channels at low level and maybe this problem could be solved only with digital control and VCAs. But I've heard that VCAs tend to deteriorate the signal. Could be there other factors in a unit like a control monitor which can cause a non clear signal?
If you think that it's better to copy this Q in the dedicated post, do it.
Thank you

My experience as a designer is that VCAs can indeed cause significant changes in sound quality (although I cannot be certain that this is your problem)..

Analogue VCAs are very difficult devices to make and have inherant issues. The kind of issues depends on the type of VCA. VCAs generally fall into 2 categories: mostly class A and mostly class B.

The class A types have best signal integrity performance, but not the best noise performance (which stays constant regardless of gain setting). So these tend to sound better but are more noisy - which is not good for manus chasing specifications.

The class B types have better noise performance (and helpfully the noise goes down with lower gain settings), but generally have a worse distortion performance which is less stable with freq (which btw may not be reflected in the specs quoted at 1KHz). In particular they can exhibit a phenomena akin to timing jitter, where the total delay through the device varies minutely depending on the instantaneous signal level and gain setting. This shows itself as an inability to get the best distortion null simultaneously over both LF and HF regions - a compromise has to be made instead. Things may have improved in recent times ( I was designing at SSL when I first came across this), but the effect I have had from this in the past is to muddy the HF, render things somewhat indistinct and damage stereo imaging - especially as the number of active channels increase within the mix. This effect seems to increase with number of active channels in a mix such that solo'ing any particular track seems ok, but the total of the whole mix doesn't.

This seems to be the reason that in the older SSLs a mix done on the small faders apparently sounded better than one done on the large VCA faders. When this was first suggested to me I didn't believe it. But I had it demonstrated to me when I worked at Virgin by a producer (I think it was Steve Lillywhite) who painstakingly set up an exact mix on both small and large faders so I could compare the two (and basically take back my disbelief). I was stunned! But even then, comparing any solo'ed tracks did not show the problem up - the problem only emerged as more and more simultaneous tracks were active..

IMVHO VCA's should be avoided if you are really interested in the max quality of a system.. You are definitely better off using faders in the digital domain :-)

[Paul Frindle]

Originally Posted by innesireinar

OK Paul, thank you.
Another Q.
I've started a post on hi end Which monitor controller do you use?
for searching a good monitor controller because mine has some little issue.
Given that you have been the designer of SSL I'd like to turn into analog world by asking you if VCAs can cause degrade of the signal. In fact the main problem of my controller is the discrepancy between the LR channels at low level and maybe this problem could be solved only with digital control and VCAs. But I've heard that VCAs tend to deteriorate the signal. Could be there other factors in a unit like a control monitor which can cause a non clear signal?
If you think that it's better to copy this Q in the dedicated post, do it.
Thank you

To address your original question; I understand that you are talking about a unit that has it's own DAC in it - used for only monitoring purposes?

I can't think why such a device would have a VCA in it at all - or am I being naive? As far as I can tell all that's needed in the audio path within the box is a DAC (with some gain control) a monitoring volume control and some output stages?

[innesireinar]

The problem origins when reading the Bob's book there is a section entirely dedicated to monitoring, how to calibrated them and some trick like level marks for a quickly return to a provious level.
SInce working in DAWs is very common today it's necessary to have a unit that allows to do all these things. When I switched to PT and sold my Yamaha 02R (I) the SPL was a unit that allowed to do some of these things, but it lacks of linearity potentiometer, good speaker's connectors (with daily false contacts), and trims for calibrating the monitor outs. I went for it because it was affordable.
Nicest units like Grace Design, Martinsound, Crane Song, Audient have all the features listed above plus individual monitor mute, ind. monitro solo, dim, mono, control level with a display for a precise level position, many input sources and for those that support 5.1 and 7.1 more other features. Other nice plus about these units is remote control. But because of this remote I thought that the level of these highest units is controlled by VCAs.
Some of these units have DAC on board but I'd like to have one without converters because I've already have a pretty good one and don't want to buy another one.
These are some of the units
http://www.cranesong.com/AVOCET.html
http://www.gracedesign.com/products/m906/m906.htm
http://www.martinsound.com/pd_mmu.htm
http://www.audient.co.uk/Audient_Pro...D=20&ItemID=38

Is there other ways for controlling more channel level at once other than VCAs by a remote?
The advice you suggested me to control the level via digital fader is a good way but don't it limit the resolution of the signal by doing this in a non-analog domain?
When mixing sometimes I need a very low level (even 50 dB SPL) and how good is to lower down the level via digital?

[bob katz]

Originally Posted by innesireinar

It started last week when I and a friend OM had one of the boring discussion analog/digital. After hours of discussion where everyone has remain in his position I promised to link him the Digido site. After two days he called me saying "your guru (you) is stating that analog is better" - And I "maybe this article is old" - "no, is updated 2005".
I went to read the article and I've only stated a discrepancy about your words between the PSW thread and the article, because on PSW you said that analog is far from perfect, works that you have received in your mastering studio done in digital, with the add of good analog outboard sounded better than those done mixed on the desk.
Therefore I'd like to have a your definitively point about that.

Oh. Yes, time has marched on, in my opinion.

In another thread here on Gearslutz, introduced by Charles Dye, called "Favorite Analog Summing", I think I've covered my opinion. Here's another take on it, I guess it's a rehearsal for a rewrite of my article!

In the past (say, through 1990), in my writing, it was a fight against cumulative quantization distortion and grunge, and the use of low-quality plugins and digitgal processors----and that made full analog mixing and processing much more attractive than digital mixing and processing. Now, today, it's the opposite, digital processing has come a long way, if you don't abuse it.

So today, if you wish to mix outside the box, you have to balance the loss of transparency that comes from passing the signal through low-resolution D/A/D converters (unless you spend the money on the best converters) against the supposed advantages of totally-analog-domain processing and mixing. And these advantages, in my opinion, can now only be justified when using a superb analog console whose coloration adds a desirable color (e.g., space, depth, definition) that cannot be obtained any other way. But even that color that, say, an API console can give you, can be obtained without the full console. And the tradeoff is probably less than going through the entire console to mix. For example, mix digitally in the box, use lots of good analog outboard for your prime signals, and possibly send the entire mix through a single pair of superb D/A converters and a pair of API modules and into a single pair of superb A/D converters to capture the mix, or a 1/2" tape machine. The "magic sprinkle" that the API pair add to that mix can produce a final mix with a unique combination of transparency and color that can sound superior to the use of 24 or 48 or however many "cheap, low-class" converters feeding a full API console.

I've objectively tested the premise that there is no problem with the digital summing mechanism (e.g. Pro Tools "infamous" summing bus) by simply taking a pair of good analog modules and putting them on a digital summing bus. If the sound gets WIDER and CLEARER with simply a pair of analog modules added to a digital sum, that makes it clear that most (if not all) of the "improvement" people attribute to analog summing is NOT due to the summing but rather to the desirable character of the analog gear they are using.

In other experiments, conducted by Linn Fuston, he demonstrated equal performance with some analog summers, and worse with many. I can confirm that the transparent analog summer which does not objectively degrade the sound, is very rare. A client sent me a matched gain and pan mix done with the Dangerous 2-bus versus digital mix in the box, and objectively and subjectively, there was nothing special about the Dangerous Mix. If anything, it sounded a little vaguer and less clear. In my opinion, it did not add any desirable distortion. I performed the listening tests blind on the client's files.

In another test, a client sent me a mix done with the Sumo with its converters versus in the box. The Sumo was EXTREMELY transparent. The two mixes were virtually impossible to tell apart, blind or sighted. But there was absolutely no advantage to the SUMO. In both cases, no analog outboard was used to "complicate" the test.

In my opinion, the bar on the digital side has been raised so far. There is still plenty of analog "processing" that sounds superior to digital processing, but summing is NOT one of those processes. So unless you have a virtually-totally-transparent analog summer or one whose losses are made up by its character (e.g. API), then I would currently recomend ITB digital mixing combined with lots of good character-providing analog outboard.

Does this help make clear my current thinking?

[Paul Frindle]

Originally Posted by innesireinar

These are some of the units
http://www.cranesong.com/AVOCET.html
http://www.gracedesign.com/products/m906/m906.htm
http://www.martinsound.com/pd_mmu.htm
http://www.audient.co.uk/Audient_Pro...D=20&ItemID=38

Is there other ways for controlling more channel level at once other than VCAs by a remote?
The advice you suggested me to control the level via digital fader is a good way but don't it limit the resolution of the signal by doing this in a non-analog domain?
When mixing sometimes I need a very low level (even 50 dB SPL) and how good is to lower down the level via digital?

You are NOT losing resolution - as I have said before you must lose that concept in order to go further in understanding how digital audio works. There is no such thing as resolution - ALL levels use ALL bits :-) The only time it LOOKS as though they don't is on your editing screen.

All other ways of reducing gain in the analogue domain digitally (short of a motorised potentiomenter or banks of physical relays) will cause more degradation than simply reducing the level in the digital domain.

Reducing the level in the digital domain to even -50dBfs with modern DACs is better than using an analogue VCA.

I don't know what else I can do to explain this - I am running out of ideas to explain this stuff :-(

[innesireinar]

Originally Posted by bob katz

Oh. Yes, time has marched on, in my opinion.

In another thread here on Gearslutz, introduced by Charles Dye, called "Favorite Analog Summing", I think I've covered my opinion. Here's another take on it, I guess it's a rehearsal for a rewrite of my article!

In the past (say, through 1990), in my writing, it was a fight against cumulative quantization distortion and grunge, and the use of low-quality plugins and digitgal processors----and that made full analog mixing and processing much more attractive than digital mixing and processing. Now, today, it's the opposite, digital processing has come a long way, if you don't abuse it.

So today, if you wish to mix outside the box, you have to balance the loss of transparency that comes from passing the signal through low-resolution D/A/D converters (unless you spend the money on the best converters) against the supposed advantages of totally-analog-domain processing and mixing. And these advantages, in my opinion, can now only be justified when using a superb analog console whose coloration adds a desirable color (e.g., space, depth, definition) that cannot be obtained any other way. But even that color that, say, an API console can give you, can be obtained without the full console. And the tradeoff is probably less than going through the entire console to mix. For example, mix digitally in the box, use lots of good analog outboard for your prime signals, and possibly send the entire mix through a single pair of superb D/A converters and a pair of API modules and into a single pair of superb A/D converters to capture the mix, or a 1/2" tape machine. The "magic sprinkle" that the API pair add to that mix can produce a final mix with a unique combination of transparency and color that can sound superior to the use of 24 or 48 or however many "cheap, low-class" converters feeding a full API console.

I've objectively tested the premise that there is no problem with the digital summing mechanism (e.g. Pro Tools "infamous" summing bus) by simply taking a pair of good analog modules and putting them on a digital summing bus. If the sound gets WIDER and CLEARER with simply a pair of analog modules added to a digital sum, that makes it clear that most (if not all) of the "improvement" people attribute to analog summing is NOT due to the summing but rather to the desirable character of the analog gear they are using.

In other experiments, conducted by Linn Fuston, he demonstrated equal performance with some analog summers, and worse with many. I can confirm that the transparent analog summer which does not objectively degrade the sound, is very rare. A client sent me a matched gain and pan mix done with the Dangerous 2-bus versus digital mix in the box, and objectively and subjectively, there was nothing special about the Dangerous Mix. If anything, it sounded a little vaguer and less clear. In my opinion, it did not add any desirable distortion. I performed the listening tests blind on the client's files.

In another test, a client sent me a mix done with the Sumo with its converters versus in the box. The Sumo was EXTREMELY transparent. The two mixes were virtually impossible to tell apart, blind or sighted. But there was absolutely no advantage to the SUMO. In both cases, no analog outboard was used to "complicate" the test.

In my opinion, the bar on the digital side has been raised so far. There is still plenty of analog "processing" that sounds superior to digital processing, but summing is NOT one of those processes. So unless you have a virtually-totally-transparent analog summer or one whose losses are made up by its character (e.g. API), then I would currently recomend ITB digital mixing combined with lots of good character-providing analog outboard.

Does this help make clear my current thinking?

Thank you, Bob
Clear.
I think summers is the actual fashion... and I've noticed money are going on them today, expecially in the project studios area.
What API modules are you referring to, for sending the entire mix into them? Eqs? Dyn?
A 10 slots API rack with some modules is in my wish list, expecially the 512c preamps.

[innesireinar]

Originally Posted by Paul Frindle

You are NOT losing resolution - as I have said before you must lose that concept in order to go further in understanding how digital audio works. There is no such thing as resolution - ALL levels use ALL bits :-) The only time it LOOKS as though they don't is on your editing screen.

All other ways of reducing gain in the analogue domain digitally (short of a motorised potentiomenter or banks of physical relays) will cause more degradation than simply reducing the level in the digital domain.

Reducing the level in the digital domain to even -50dBfs with modern DACs is better than using an analogue VCA.

I don't know what else I can do to explain this - I am running out of ideas to explain this stuff :-(

Ok, Ok, Ok, Ok...understood.
The only way for setting the digital level in my system is the master fader of PT.
Is this the reason why the master fader of PT has insert slots post fader?
Because from what I've learned once the signal has been dithered and "converted" to 16 bit it's not possible to change the digital level without backing into a high word lengh signal, am I right Paul?