The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
My Theory About Prosumer Audio Audio Interfaces
Old 27th June 2006
  #181
Quote:
Originally Posted by not_so_new
Exactly. What he said.....

I am sure you know this but you might not have thought of it this way before.

Yes a combination of sine waves are more complex but think of how the converter sees it. It does not "see" the whole wave like we do in a DAW, it is really just seeing the snapshot of the wave compared to the wave snapshot that just preceded it.

If you take it down to that level it really does not matter how complicated the wave form is it just needs to have a voltage shift that is 1/2 of the sample rate or lower.
I guess now we're on sample rate, not bit rate.

The only quesion I have is what about the time intbetween samples? I know that 44.1k is very fast and even more so the higher sampling rates. with digital, we're still trying to re-create fluid events with snapshots. The more snapshots, the more fluid (smooth) the re-creation and the bigger the picture, with more information, the more realistic the re-creation.

I can't see how more bits and higher samplling rate won't always be "better" to the point of infinity.

If you want to argue that we've passed the threshold of perception, which may really be what one side is discussing, that may be true, but it's different than what's the best way to digitally capture audio, or whatever you want to call the original point.

And, if we're going to discuss the threshold of perception, we've got to make a distinction between conscious and unconscious perception. For me the latter term is an oxymoron, but i'm not sure what else to call it. But ask Rupert Neve or Dr Oohashi about hypersonic effects - whether you agree with their conclusions or not, Oohashi has clearly shown that there are two levels of perception.

This is a weak argument to make, but when you look at this from that perspective, the arguement becomes, record at a higher rate even if you don't hear a difference, becuase unconsciously you may. (I'm imagining the hostile posts to follow this - let me save you the trouble, I see the problems with the arguement).

I think separate from the discussion, but slightly connected to it is the question of "Why do we listen to music in the first place and how do recording/reproducing techniques affect the effectiveness of the music on satisfying that drive to listen to music?"

I hope I've articulated the question the way I intended, but that's the the most important thing for us to address. It may or may not affect our bit/sample rate choices and opinions.
Old 27th June 2006
  #182
Sorry if you know all of this already… I started typing and I hate to waste a post….

heh

Quote:
Originally Posted by Mike Caffrey
That makes sense. it's just looking a moments in time, it has no idea what's before or after.
Yes exactly... it is a "snapshot" of a given moment in time so nothing really matters but "is it above the noise floor" and "is it below the ceiling."

Because sound is a linear event that takes place in time one snapshot means nothing by it's self. It takes 2 snapshots to make anything at all happen (in reality it takes more than that to even hear anything useful) so added to the rules above you need snapshots to take place in a very precise way. This, by the way, is one of the big keys to converter design. Making the snapshots happen at the proper time in the linear event. When the wave form is captured, as long as that wave form is "seen" by at least 2 snap shots it will be converted to a proper digital signal.

So signal complexity really has nothing to do with the converter math at all. As long as the three rules above are met, is it above the noise floor, below the headroom and slow enough to be captured by at the very least 2 snapshots you should be good to go. There is an argument that signal complexity might have an effect on the power supply of the converter hardware but that is also where better converter designs come into play.
Old 27th June 2006
  #183
Quote:
Originally Posted by Mike Caffrey
I guess now we're on sample rate, not bit rate.

The only quesion I have is what about the time intbetween samples? I know that 44.1k is very fast and even more so the higher sampling rates. with digital, we're still trying to re-create fluid events with snapshots. The more snapshots, the more fluid (smooth) the re-creation and the bigger the picture, with more information, the more realistic the re-creation.

I can't see how more bits and higher samplling rate won't always be "better" to the point of infinity.

If you want to argue that we've passed the threshold of perception, which may really be what one side is discussing, that may be true, but it's different than what's the best way to digitally capture audio, or whatever you want to call the original point.

And, if we're going to discuss the threshold of perception, we've got to make a distinction between conscious and unconscious perception. For me the latter term is an oxymoron, but i'm not sure what else to call it. But ask Rupert Neve or Dr Oohashi about hypersonic effects - whether you agree with their conclusions or not, Oohashi has clearly shown that there are two levels of perception.

This is a weak argument to make, but when you look at this from that perspective, the arguement becomes, record at a higher rate even if you don't hear a difference, becuase unconsciously you may. (I'm imagining the hostile posts to follow this - let me save you the trouble, I see the problems with the arguement).

I think separate from the discussion, but slightly connected to it is the question of "Why do we listen to music in the first place and how do recording/reproducing techniques affect the effectiveness of the music on satisfying that drive to listen to music?"

I hope I've articulated the question the way I intended, but that's the the most important thing for us to address. It may or may not affect our bit/sample rate choices and opinions.
We posted at the same time....

Yes you are correct, we are talking about something else when we talk about sample rate. It sounds like you know most of this stuff so forgive me if I am talking about things you already know... maybe it will help someone else.



Sample rate is how many "snapshots" we take in a given time frame. Think of a baseball leaving the pitchers hand traveling to the catcher's mitt. If we have 2 photos, one of the pitcher pitching and one of the catcher catching we know that the ball left the pitcher's hand and the catchers caught it… but we don't know much of what happened in between do we.

So take another photo at the half way point of the ball's movement. Now we know that the ball was not a 20 foot lob to the catcher it was traveling at about 5 feet off the ground on the way to home plate. The more photos that we add to the mix the more detail we get out of the whole motion. Once we get up to about 24 photos (or frames) a second now we can actually see the ball move in the air and make judgments like "that was a fast ball" or "that was a slider."

Now here is the hitch in all this. There is a theoretical and practical limit where the human eye can't keep up. Our eye has a refresh rate as well so at some point you can add one more frame per second and the eye will not register the difference. I forget what the limit is right now and I am too lazy to go Google it but let's say that is at 500 frames per second (I know it is much less than this). When you go to 501 frames per second you have increased the bandwidth needed to reproduce that signal, you have made the sampling machine (TV camera, film camera, digital converter whatever) work harder which will increase the possibility of error over the whole sample range and you have not gained anything that the eye can perceive as better.

The eye and the ear are not so different, they are limited evolutionally structures. The refresh rate of a dog's eye is MUCH faster and better than ours. My dogs can see things move in our back yard that I never see so I know there is information there that I am missing. Do I subconsciously "know" about the info? I don't think so..... many people mistake "subconscious" things as mystical elements. In actual fact they noticed that their dogs noticed the squirrel in the tree for instance.

Do I believe that there is subconscious acoustic info in the hyper sonic rang of our hearing? I am not convinced but I have heard very good arguments on both sides. I will say this, I have never seen a true scientific test that has been verified by other research that shows this to be true… that does not mean it does not exist it just means I have not seen anything.

Again this is all different than bit rate but the same arguments can be made there as well. There comes a point where bit rate is not able to be discerned by the human ear. Also there comes a point where the noise floor and the ceiling of the hardware just can't keep up. After 24 bits there is a pretty steep drop in the pay off of the audio quality… 32 bits for some headroom on DAW busses and plug-ins but 24 bits is pretty darn good for our analog needs.
Old 27th June 2006
  #184
Lives for gear
 

Quote:
Originally posted by Mike Caffrey:
That makes sense. it's just looking a moments in time, it has no idea what's before or after.
The problem is that's the opposite of the way it works...for digital audio to work it has to know what's before and after.

Quote:
Originally posted by Mike Caffrey:
The only quesion I have is what about the time intbetween samples? I know that 44.1k is very fast and even more so the higher sampling rates. with digital, we're still trying to re-create fluid events with snapshots. The more snapshots, the more fluid (smooth) the re-creation and the bigger the picture, with more information, the more realistic the re-creation.
Up to a point, that's true...but regardless of how high the sampling rate is, as long as the highest frequency of the signal you're trying to sample is less than half the sampling rate of the system it will be captured accurately. Obviously, the filters that are used to get rid of the higher frequencies will affect the signal to some degree, but those filters have gotten better and better over the years which is why today's converters sound so much better than older ones, even at lower sampling rates.

And it's not correct to think of more rates as making the resultant output signal more "fluid"...D/A converters fit a curve to a series of points, they don't connect the dots.

If you think about it, if you have a compass and know where the center is, how many more points do you need to draw that circle? Just one, right? Having two, or ten, or a hundred more won't make that drawing any more accurate. Digital recording systems kind of work the same way...they're more complicated, sure, but it's essentially

Quote:
Originally posted by Mike Caffrey:
I can't see how more bits and higher samplling rate won't always be "better" to the point of infinity.
They will always be better to the point of infinity, but we're not talking about infinity here...we're talking about signals within defined ranges, in which case we won't be gaining accuracy by raising sampling rates as far as, say, frequencies below 20 kHz when we go from 44.1 kHz to 96 kHz. We will be able to record higher frequencies, which does mean that we will be capturing our signal more accurately, and as we move up to infinity our sampling rates would have to go correspondingly higher as well to capture that. But that won't make the <20 kHz components of that signal more accurate.

As far as the benefits of capturing those higher frequencies, though...that's a whole different discussion. But then, so is this one I guess.

Quote:
Originally posted by Mike Caffrey:
But ask Rupert Neve or Dr Oohashi about hypersonic effects - whether you agree with their conclusions or not, Oohashi has clearly shown that there are two levels of perception.
On that side of things...not to sidetrack the discussion even further...I'm not convinced that either has shown the need for higher sampling rates. The Rupert Neve/Emerick story just shows the effect that things that happen at higher frequencies can have on audible frequencies, which can be captured at lower sampling rates (just tweak the highest inaudible frequencies on a Massenburg or Nightpro EQ and record at 44.1 kHz...you'll hear it), and as I understand it the Oohashi study showed that our brains can register those higher frequencies, but wasn't it also the case that nobody could consciously tell the difference?

In any case, my philosophy has always been to use the higher rates if your converters sound better at those rates, but I'd certainly rather see the lower rates continue to get better. Why not?

Quote:
Originally posted by Not So New:
The eye and the ear are not so different, they are limited evolutionally structures.
True, but there is a big difference in the way digital audio and film work. With film, what we're looking at is a series of stills that go by so fast that we perceive them as mostion. With digital audio, what we're hearing is an analog signal...it's not our ears that put everything together, it's the converters.

Quote:
Originally posted by Not So New:
Again this is all different than bit rate but the same arguments can be made there as well. There comes a point where bit rate is not able to be discerned by the human ear.
It's not a question of what our ears can perceived...it's a question of our analog electronics not being able to catch up. Right now even the best 24-bit converters only give us 20 or 21 bits' worth of performance. But it is also true that our ears can't perceive that wide of a dynamic range, so even if it does get to the point where a system can be designed that does more it will be of little practical usefulness.

-Duardo
Old 28th June 2006
  #185
Gear Nut
 

Quote:
Originally Posted by Dave Derr
Naren said: "Digital audio must have some limits by its nature. It was never meant to record conventional sources at "full scale."

It wasn't? Was there a law passed that I don't know about? ::wink::

And you'd better tell every mastering engineer in the world who puts the PEAKS on CDs right up against Full scale on every single record... at least within -.5dBFS.

Personally, I wouldn't even think of putting out a DAW that couldn't handle full scale signals... on every single friggin channel! By not handling full scale gracefully, DAW manufacturers are saying "Well you've got these great converters, but we won't let you get the best performance from them... and still work cleanly in our DAW system".

No, there are no laws but there are standards. Maybe they should have taken -15dBFS and redesignated it as 0dBFS giving headroom above zero. I really feel this is somewhat a semantical argument over what "0" means. 0dBFS is NOT 0dBvu, as you are well aware.

When I said "digital was never meant to record conventional sources at full scale," I meant, obviously, sources that would number in the dozens and require multiple plug ins and subsequent summing to stereo. This would be analogous to running the channels of an analog console to the extreme using ALL available headroom in every channel, running your inserts at maximum level, and then summing all of it to your stereo buss. 0dBFS is the END of the headroom. The headroom in digital is there if one is conservative with levels. Gain structuring applies to digital as well as analog.

Of course I know that CDs are mastered to almost full scale. Is there anybody who doesn't know this? :-) It doesn't matter at that point though because all the digital processing is done. But that is not the context to which I was referring.

Apparently, one big problem with DAWs and plugs is the inadequate metering which does not register the levels of intersample peaks. You may think you're running DAW signals at full scale but in fact may be abusing them by a large margin without your meters even telling you. Of course this is another huge reason to back off on digital levels.

One method suggested is to record fairly hot levels but bring down the channel gains (pre-everything) in the DAW. Makes a lot of sense to me, and it gets closer to your ideal of running convertors at their highest quality. This is actually the method I use (in combination with not absolutely slamming the convertor).

However, I question that convertors are not at their best quality when not used to full scale - at least for all practical purposes. Quality (being subjective) is subject to the limitations of human hearing, and within those limitations I would defy anyone to successfully differentiate between 24 bit signals that were recorded at -0.5 dBFS and -20 dBFS (or even lower).

-Naren
Old 28th June 2006
  #186
The Distressor's "daddy"
 
Dave Derr's Avatar
 

Quote:
Originally Posted by Naren
One method suggested is to record fairly hot levels but bring down the channel gains (pre-everything) in the DAW. Makes a lot of sense to me, and it gets closer to your ideal of running convertors at their highest quality. This is actually the method I use (in combination with not absolutely slamming the convertor).
Yes I totally agree! Good way to work. Protools HD however, has the plugs before the fader, so if the plug-in has gain and isn't written to provide headroom in its 24 bit system, the plug-in will clip with a nicely recorded full scale track. <The Bastards!> There are places where the summing busses will clip also in Protools HD. Theres a thread about this a few pages back where we question why HD is 24 bit non-floating point, while many cheaper DAWs are 32 floating point, giving you more flexibility and headroom, as well as ease of DSP.

As we stated tho, fortunately modern converters can be incredibly linear and quiet with some headroom left on them. Just seems stupid that any DAW doesnt allow for worry free full-scale tracks. Oh well. Someday we will probably all be working with 64 bit busses and headroom enough for Shaquil O'Neil.
Old 28th June 2006
  #187
Lives for gear
 
animix's Avatar
Quote:
Originally Posted by Sizzleboy
Couple of things:

As far as the DAW failsafe thing, the Ensoniq Paris system, for all it's quirks, had something going on where when you slammed it, it didn't clip hard (aside from the input A/D).

If you use Cubase in 32bit float mode, you can slam anything so long as you don't slam the master 2-bus.

That said, I agree with the 'f-em attitude. Let them figure it out themselves. These prosumer companies can sell all the equipment they want to the would-be producers, but knowledge is something that you have to work for.
I lightpipe a 48 track digital matrix from Cubase SX to Paris and pan and sum in Paris for just this reason. You can take an extrtemely phat mix that is being processed 32 bit float in SX and as long as you're not clipping the plugins or using the Cubase mix bus, you can get a a hot digital mix happening without clipping. then you can lightpipe the individual tracks to Paris and due to the Paris DSP and 52bit fixed point summing bus you've got an additional 24dB of headroom to work with. It's amazing.......absolutely amazing.
Old 28th June 2006
  #188
Lives for gear
 

Intriguing - the freeware DAW Reaper uses 64 bits internal maths - maybe it's worth mixing in Reaper for even more headroom ...
Old 28th June 2006
  #189
Gear Nut
 

Quote:
Originally Posted by Dave Derr
Yes I totally agree! Good way to work. Protools HD however, has the plugs before the fader, so if the plug-in has gain and isn't written to provide headroom in its 24 bit system, the plug-in will clip with a nicely recorded full scale track. <The Bastards!> There are places where the summing busses will clip also in Protools HD. Theres a thread about this a few pages back where we question why HD is 24 bit non-floating point, while many cheaper DAWs are 32 floating point, giving you more flexibility and headroom, as well as ease of DSP.

As we stated tho, fortunately modern converters can be incredibly linear and quiet with some headroom left on them. Just seems stupid that any DAW doesnt allow for worry free full-scale tracks. Oh well. Someday we will probably all be working with 64 bit busses and headroom enough for Shaquil O'Neil.
It's been a while since I worked on Pro Tools, but I know the problems you refer to. The solution is to insert a gain plug as the first plug on every channel. There is such a plug that is a gain-only device but I can't remember what it is.

Recently, I started using Sonar 5 for a home demo writing setup. It has a gain built at the beginning to every channel. I'm surprised this isn't standard practice as it mimics the gain pots that have been there forever at the top of analog console channels. Sonar also has a 64 bit float mix engine. Really a remarkable piece of software, and unless you're doing post work is a totally pro solution. But most can't use it obviously due to compatibility issues with other studios. Thus Pro Tools continues its market domination.

-Naren
Old 28th June 2006
  #190
The Distressor's "daddy"
 
Dave Derr's Avatar
 

Quote:
Originally Posted by Naren
It's been a while since I worked on Pro Tools, but I know the problems you refer to. The solution is to insert a gain plug as the first plug on every channel. There is such a plug that is a gain-only device but I can't remember what it is.
YUP!, And fortunately, most plugs have an input attenuator available also, so you can kind of trim out full scale signals (and lose resolution! lol)
Quote:
Originally Posted by Naren
Recently, I started using Sonar 5 for a home demo writing setup. It has a gain built at the beginning to every channel. I'm surprised this isn't standard practice as it mimics the gain pots that have been there forever at the top of analog console channels.
YES YES YES! Exactly! Let us trim the gain at the top of all virtual mixers so we can work like we did on analog consoles. Of course this takes a wee bit of processing power, but at least we could then do our favorite tricks and even do the nice visual "fader line" where every fader is about even with its neighbors as a nominal position. I still want 32 bit floating point tho. Wanna bet that Protools HD will someday switch to it (or 64 bit floating point), and all our plugs won't work properly when they do? GOOD BYE OLD MIXES!

What do ya think... 3 years?

(PS: I'm not slammin Protools HD. They aren't the only ones whose software upgrades stop older software (and mixes) from working properly.)
Old 28th June 2006
  #191
Lives for gear
 

Quote:
Originally Posted by Duardo
The problem is that's the opposite of the way it works...for digital audio to work it has to know what's before and after.


Up to a point, that's true...but regardless of how high the sampling rate is, as long as the highest frequency of the signal you're trying to sample is less than half the sampling rate of the system it will be captured accurately. Obviously, the filters that are used to get rid of the higher frequencies will affect the signal to some degree, but those filters have gotten better and better over the years which is why today's converters sound so much better than older ones, even at lower sampling rates.

And it's not correct to think of more rates as making the resultant output signal more "fluid"...D/A converters fit a curve to a series of points, they don't connect the dots.

If you think about it, if you have a compass and know where the center is, how many more points do you need to draw that circle? Just one, right? Having two, or ten, or a hundred more won't make that drawing any more accurate. Digital recording systems kind of work the same way...they're more complicated, sure, but it's essentially


They will always be better to the point of infinity, but we're not talking about infinity here...we're talking about signals within defined ranges, in which case we won't be gaining accuracy by raising sampling rates as far as, say, frequencies below 20 kHz when we go from 44.1 kHz to 96 kHz. We will be able to record higher frequencies, which does mean that we will be capturing our signal more accurately, and as we move up to infinity our sampling rates would have to go correspondingly higher as well to capture that. But that won't make the <20 kHz components of that signal more accurate.

As far as the benefits of capturing those higher frequencies, though...that's a whole different discussion. But then, so is this one I guess.


On that side of things...not to sidetrack the discussion even further...I'm not convinced that either has shown the need for higher sampling rates. The Rupert Neve/Emerick story just shows the effect that things that happen at higher frequencies can have on audible frequencies, which can be captured at lower sampling rates (just tweak the highest inaudible frequencies on a Massenburg or Nightpro EQ and record at 44.1 kHz...you'll hear it), and as I understand it the Oohashi study showed that our brains can register those higher frequencies, but wasn't it also the case that nobody could consciously tell the difference?

In any case, my philosophy has always been to use the higher rates if your converters sound better at those rates, but I'd certainly rather see the lower rates continue to get better. Why not?


True, but there is a big difference in the way digital audio and film work. With film, what we're looking at is a series of stills that go by so fast that we perceive them as mostion. With digital audio, what we're hearing is an analog signal...it's not our ears that put everything together, it's the converters.


It's not a question of what our ears can perceived...it's a question of our analog electronics not being able to catch up. Right now even the best 24-bit converters only give us 20 or 21 bits' worth of performance. But it is also true that our ears can't perceive that wide of a dynamic range, so even if it does get to the point where a system can be designed that does more it will be of little practical usefulness.

-Duardo
I've read that there has been tests done about mixes recorded and played on monitors that have been able to respond up to 50kHz. Even though the ears cannot hear it the body can feel these extra frequencies. They did double blind testing and drew the conclusion that audio frequencies are not adapted only by the ears, but by the whole body.

There is a problem with high sample rates in practise, sampling accuracy goes down when the speed goes up. So even when you could digitally describe the audio signal more accurately it will be compensated negatively by worse sampling quality.
Old 28th June 2006
  #192
The Distressor's "daddy"
 
Dave Derr's Avatar
 

Transients and our Ears

It may not be the "super-sonic" frequencies we hear so much as the transients. There is some evidence that there are outer ear hair nerves that can detect sharp rising edges, as opposed to frequencies. These hairs help determine when something is close and "sharp" in shape, as opposed to distant and diffused in shape. At least that is the proposal, still yet to be proved.

However, If as an audio engineer, you get everything right below say 10Khz: balance, eq, compression, and the music itself, I guarantee the listeners will be much happier than if you have a mediocre song or flawed production that goes out to 200KHz!

Sure, a cymbal may sound more present etc etc, but it's music and the pleasing big things that are going to get you noticed as a GREAT ENGINEER. For now, I always say to focus on the things you can control and that REALLY REALLY make a difference to everyone, and quit all the fretting about converters, clock jitter, phase alignment, sample rates, bus width, etc etc blah blah blah. You all have this amazing technology and quality at your fingertips, on your desktop, that no one had 20 years ago. And as you know, that didnt stop some of the greatest records from being made: People focused on the SONGS, ARRANGEMENT, PERFORMANCE, EQ, COMPRESSION, BALANCES ETC, and NOT the technology used to make it. As George Martin said "ALL IT TAKES IS EARS".
Old 28th June 2006
  #193
Gear Head
 

umm, sorry to butt in to this monument of digital debate, but with all the talk of 32 bit busses, i have a question:

i usually record at 24/48 or 24/44.1, with levels fairly close to dBfs. should i move to a 32 bit float after all tracking is done and i'm ready to mix?
Old 28th June 2006
  #194
Lives for gear
 

Quote:
Originally Posted by Naren
Recently, I started using Sonar 5 for a home demo writing setup. It has a gain built at the beginning to every channel. I'm surprised this isn't standard practice as it mimics the gain pots that have been there forever at the top of analog console channels.
-Naren
I'm pretty sure most other daws have that feature. PT is ... different ... in that regard, there's not really much there as standard although you can plugin whatever you want.

Although if you can just select all the parts on a track and non-destructively pull their levels down by 3db or whatever... it's the same thing. In Nuendo segment / part changes (like the volume handles) are pre-dsp / pre-fader so using that or a trim pot on the channel should give an identical result.

I assume that would be the same for PT or any other daw. So although useful at times when recording, I suppose gain pots in daw ch's after that are kinda redundant given volume handles and things like that... assuming those volume changes are pre-everything.

Sonar looks pretty darn cool though.

Lawrence
Old 29th June 2006
  #195
The Distressor's "daddy"
 
Dave Derr's Avatar
 

Swan

Quote:
Originally Posted by swan
umm, sorry to butt in to this monument of digital debate, but with all the talk of 32 bit busses, i have a question:

i usually record at 24/48 or 24/44.1, with levels fairly close to dBfs. should i move to a 32 bit float after all tracking is done and i'm ready to mix?
Thats a hard question because it depends on so many things, like what Daw you have, and how aware you wanna be of possible clipping. If you have Protools HD, I'm beginning to think you have to be careful, and possibly attenuate at or before any plug in. Some plugs have headroom built in. Also, I have personally had effects busses and other busses clip, but it wasnt so hard to work around.

Most 32 bit systems will intelligently scale the numbers and leave headroom.

Most professional engineers also record close to dBfs so you are in good company, but as we have discussed in this thread, even with a nice clean converted signal, things happen after the conversion that can leave us wondering what happened to our wonderful sounds and fidelity.

If you're mixing in analog, all you have to do is get the digitized tracks out cleanly to your mixing desk, and NOT being 32 bit shouldnt be an issue.

However, some people use a hybrid approach and do some mixing and processing ITB, and some outside the box, so once again they are prone to clipping more in the 24 bit DAW, than in a 32 bit DAW.
Old 29th June 2006
  #196
Lives for gear
 
ryst's Avatar
 

Quote:
Originally Posted by Dave Derr
YUP!, And fortunately, most plugs have an input attenuator available also, so you can kind of trim out full scale signals (and lose resolution! lol)


YES YES YES! Exactly! Let us trim the gain at the top of all virtual mixers so we can work like we did on analog consoles. Of course this takes a wee bit of processing power, but at least we could then do our favorite tricks and even do the nice visual "fader line" where every fader is about even with its neighbors as a nominal position. I still want 32 bit floating point tho. Wanna bet that Protools HD will someday switch to it (or 64 bit floating point), and all our plugs won't work properly when they do? GOOD BYE OLD MIXES!

What do ya think... 3 years?

(PS: I'm not slammin Protools HD. They aren't the only ones whose software upgrades stop older software (and mixes) from working properly.)
Dave,

Let me get this correct. What you are saying is to track as close to 0 as possible and then when mixing either trim the input on all the plugs you use or either print a "trim/gain plug" on all tracks, correct? And how many db's should you be trimming if you print a trim/gain plug on each track? Do you suggest keeping the master fader at zero or pulling the master fader down if you see the red bars light up?
Old 29th June 2006
  #197
Lives for gear
 
John Suitcase's Avatar
 

I did a session yesterday, and purposely used very low input levels. Basically, I'm recording in Cubase SX 3 with a MOTU 828 and an Behrry ADA8000 (low end, for certain.) I use a Soundcraft Spirit F1 mixer for most of the Pre's. I had the input levels on the MOTU at o, all the way to the left, and brought the level up using the Soundcraft, until they were peaking at around -6db, or so. Lower than that seemed just a little too thin and brittle to my ears, not too mention that I was having to crank up my outputs to hear back what I'd recorded. I heard some pretty nasty distortion, but determined it was mostly because I was having to crank the Soundcraft up past 12o'clock to get a decent level. I brought the MOTU input levels up a bit, and backed off the Soundcraft, and it started to sound better. I didn't use any compression or other processing on the way in.

Overall, I thought the results were good, the biggest things I noticed were the transients on the drums. They had a lot more crack than I usually get, so much so that the snare was almost painful until I eq'd and added some compression to it.

I think the key to this whole discussion is to use your ears, and be aware that you can overload circuits all the way from the mic to the convertor. Each piece should be run at it's optimum level, or as close to that as possible. Leaving a good amount of headroom allows the convertors to capture transient peaks more accurately, leaving headroom throughout the rest of the chain eliminates distortion, but you have to have enough level to overcome the inherent noise in the analogue part of the system.

My Soundcraft Pre's sound really nice when they are about halfway up, more than that and you start to get some distortion, less and they're a shade noisy. The MOTU has to be set as close to 0 as possible, if you want to avoid the coloration that it can give. But, too low and things just get difficult. Sounds that are VERY dynamic are more accurate, truer to the source, etc., but they can be much hard to mix, much harder to blend.

Anyway, I was happy with the experiment, and in the end it confirmed what I think we all already know, you have to use your ears.
Old 30th June 2006
  #198
Quote:
Originally Posted by John Suitcase
Overall, I thought the results were good, the biggest things I noticed were the transients on the drums. They had a lot more crack than I usually get, so much so that the snare was almost painful until I eq'd and added some compression to it.


My Soundcraft Pre's sound really nice when they are about halfway up, more than that and you start to get some distortion, less and they're a shade noisy.


Anyway, I was happy with the experiment, and in the end it confirmed what I think we all already know, you have to use your ears.
I want to play devil's advocate here. I think you have to use your brain too.

Are you saying that you were less happy with the snare sound than usual? If the snare was "almost painful" and your previous method wasn't this sounds like a step in the wron direction. I'm not sure if I'm reading into this, but it sounds like you're saying that you didn't like the sound until you added EQ and compression, where that might not have been the case in the past.

My real question is, how did you get the lower levels? I assume that you turned your pres down, right?

Distortion is clipping, which is a form of peak limiting and you say that your pres distort fairly easily. You may be hearing the tonal change of your pres and a reduction in peak limiting caused by clipping in you pres as they start to break up.

Also, did you listen only one track or all of them. Did you listen out of a DAW or soloed off a console?

You may be hearing differences in the sound of your summing buss.

All of these things could still support the argument that it's better to print at a lower volume, but if your intent was to evaluate the sonic differences of audio files recorded at different levels, you may not really have done that with this particular tests. There was more than one thing changing than the level of the signal as it hit the converters.
Old 30th June 2006
  #199
Lives for gear
 

This entire thread has been really informative. I've disagreed with some here but that just happens during the course of converstations... I've learned a few things.

Aside from the core of the earlier debate I'm still interested to know why converters are designed (or calibrated) so far away from 0dbVU. I tried unsucessfully to research this and could not get a definitive answer.

Just to be clear why I'm confused about that...

A lot of the analog gear we use sounds great in the operating ranges that get up to +3/+4 dbVU on analog tape which is, by most accounts, 14 / 12dbFS or so in the digital world.

Why doesn't the modern digital scale (or modern converter calibrations) match up closer with that like 0dbVU = 6dbFS or so, still leaving a little headroom over 0dbVU? Wouldn't that allow more converter "resolution" while keeping the range of I/O levels more conducive for interfacing with analog gear and not having to "overdrive" some gear that may not behave well at +15dbVU?

I'm sure there's a vaild reason for it, I'd sure like to know what that is.

Anybody?

Lawrence
Old 30th June 2006
  #200
Lives for gear
 

I don't understand what you don't understand. Converters are primarily an analog device, and they are definately designed with a particular 0dBVU level in mind. This is usually around -18dbFS. (Don't confuse dbFS with dbVU).

So that means they expect you to calibrate your system so that at 0dBVU you have around 18dB headroom.

It's a little irrelevant if you don't actually have VU meters on the analog gear feeding your converter.

EDIT - when you said "why converters are designed (or calibrated) so far away from 0dbVU." did you actually mean why aren't they closer to 0dBFS?

My question would be, why are they designed so close? Good analog gear has about 30dB headroom.

It's a compromise between the requirements of analog and the requirements of digital. They are two totally different things, so it's a trade off between headroom and resolution. I don't think it's a real problem - it's just that people like to abuse their converters by tracking too hot.
Old 30th June 2006
  #201
Lives for gear
 

Just a few thoughts about converters in general...

Quote:
In practice, the resolution of the converter is limited by the signal-to-noise ratio of the signal in question. If there is too much noise present in the analog input, it will be impossible to accurately resolve beyond a certain number of bits of resolution, the "effective number of bits" (ENOB). While the ADC will produce a result, the result is not accurate, since its lower bits are simply measuring noise. The S/N ratio should be around 6 dB per bit of resolution required.
This goes back to a previous poster noting that we will never, in practice, ever record a signal anywhere near close to the signal to noise ratio of a full scale 24-bit converter. This also answers why dynamic range plays into the conversation.

You can crank your (60db S/N) input signal up as much as you want, it's still going to be a signal with a 60bd S/N ratio. That means that you only need (roughly) 65-70db of dyanamic range to capture a pristine signal without quantanization errors or additional noise. That's way below full scale. Recording that 60db signal up to full scale only increases the converters S/N ratio. As you'll see that is practically irrelevant in real world use in a typical working range of input levels.

Only two things make up sounds... amplitude and frequency...

Note: Dave corrected me, there's many more things than those two. I just focused on those two for the sake of this discussion to keep things simple. Volume and frequency. Thanks Dave.

Amplitude = bitdepth. (available amplitude level from noise floor to peak / dynamic range) The greater the bit depth the greater the dynamic range (or theoretical maximum S/N of the converter)

Frequency = sample rate. (highest frequency you can record)

As long as your sample rate can cover the frequencies you're good.

As long as your dynamic range can cover the amplitude (working dynamic range > signal's S/N) you're good. You have all the resolution you need.

This is where this topic got a little off track.

Even though the converter may be capable of a dynamic range of 110db at full scale, if the actual S/N of the signal is 65db you can record well below full scale and not "lose any resolution" in practice. In effect, your -15 level has "more than enough bits" to capture your signal faithfully without the noise floor of the signal or the converter becoming a factor.

The following chart (16 bits) demonstates why dynamic range and bit depth are practically one and the same and why at 24-bits, the dynamic range of the converter at full scale is so far beyond anything you'll ever record (S/N 65db?) that lowering the signal (to a certain degree) does not cause "actual" physical degradation. In reality, 20 bits get the job done. 24-bits allows much more headroom. Use it... the headroom not the 24-bits.

Sure, the noise of your converter will be closer to the peak of your signal than it would be at full scale... BUT it doesn't matter since the noise of your signal is SO FAR ABOVE THAT. That's your true noise floor. Which is why it's practically irrelevant in a good working input range inside the dynamic range your using during recording. That could be -20 and it would not matter and it would not cause the signal to be "worse" or to be somehow captured incorrectly.



In the picture above, with the waveform in the same spot as it is there, if this was 24-bit conversion, there would be a lot more steps between each of those points so the rounding off would be more accurate. Those steps don't decrease as you go down or increase as you go up. They're evenly spread from noise floor to peak. A signal at -10 will get "rounded" just like a signal at -2. The 'resolution' is still apllicable to moving the amplitude from a sample point to the nearest bit which will represent that sample points amplitude. The closer they are together the fewer quantanization errors. The 24-bit "resolution" (spaces between the bits for representing amplitudes of sample points) operates effectively across the entire range of amplitudes not just up near the top of the scale.

You don't "gain" "resolution" by moving up. They bits are evenly spaced, not closer together near the top.

It's how close they are to each other, not how close the signal is to the top of the scale that makes them effective in drawing an accurate "picture" of the analog wave. If "resolution" was only valid at the top of the scale our reverb tails would sound like **** as they fade out. They don't. At least mine don't.

Bit depth, in and of itself, is only useful for reducing quantanization error and increasing dynamic range. Again at 24-bits the dynamic range is so far beyond our working S/N ratio it doesn't matter much beyond 70db or so. Hence, no need for 32-bit converters. We can't even hope to approach the maximum dynamic range of 24-bit converters. 16-bits? Yes. Which is why professional converters are 24-bits now. More than you'll ever need.

If "bit depth" universally equaled (without limits) resolution and better sound (i.e. more accurate representation of the signal) then we'd have 32 & 48-bit converters on the market. There is a practical real world 0-gain limit to bit-depth / resolution when capturing audio signals and it's not AT 24-bits. It's above 16 and below 24-bits. It's defined by what we record. 24-bits is "headroom". You don't need all of them, never have. We needed more than 16, we went to 24... more than enough. If that wasn't true then PTHD would have 32-bit converters with it's 192k sample rate.

If you were recording signals with S/N of 100db then yes, approaching full scale should "sound better" than recording into a practical 80db dynamic range since your signal's noise floor would be lower than the converter's noise floor and the practical "steps" wouldn't cover the range they need to cover. We don't. Ever. Never. Nada.

When we tracked at 16 bits that was not always true, which is why 24-bit conversion sounds so much better. Less quantanization error.. more steps. You don't need all 24 though... that's the common misconception.



At lower bit depths (16, 8) sampling quality is lost due to quantanization error, compared to 24-bits. Even at -15/-20 in 24-bits there are more than enough "steps" to represent the signals true shape from it's noise floor to it's peak. It's dynamic range... noise floor to signal peak.

Can't record below the noise floor or above the peak. That's the entire signal. Making it louder into the converter input doesn't change the S/N of the signal itself, only the S/N of the converter. Again, it's so far below the signal's noise floor... it doesn't matter.

Note: If this was not good enough you would see 32-bit converters. This is why there are no 32-bit converters. We don't need that much dynamic range. Even at well below full scale there is enough bit depth (20 bits whatever) to faithfully capture the entire "dynamic range" (noise floor to peak) of any musical signal. Anything beyond (below) the noise floor of the signal(s) is irrelevant. You'll hear the noise of the signals long before the noise of the converter with a 70db S/N signal and a 80db (well below full scale) dynamic range / recorded level.

16 million steps of resolution (24-bit) is enough to mark sample point amplitudes accurately enough so that the waveforms can be accurately represented. Actually FAR less than 16 million, maybe 10, 9, 5 million or less steprs which is FAR more than the 65 thousand steps of a 16-bit depth.

And you won't have any more quantanization errors than recording at full scale. You have more than enough 'resolution' (fine steps between amplitude points to mark) even at those lower levels.

Again, moving the captured signal up closer to full scale accomplishes only one thing, keeping the noise floor of the system/converter down lower. Since your recorded source's noise floor is FAR above that... it doesn't matter. Well actually, 2 things... it makes it play back louder.

So yes, (as many people stated correctly) recording at full scale is the converter performing at it's theoretical best,

... but not at it's "best practical sound due to more bits". It's performing at it's best "internal peak signal to noise" ratio of the converter. Which is far below any practical signal S/N ratio of any signal you'll ever feed into it. So again... it simply doesn't matter. It's irrelevant as long as your working dynamic range is greater than your signal's S/N ratio.

If that's 110db and your signals are typically 70db then ...

It doesn't matter. It's not going to record the audio "better" with less quantanization error , within the range of the S/N of the signal, the recorded signal itself is not going to be any more accurate ... only louder. And the converter noise floor is down farther from the peak but again... it doesn't matter... the REAL noise floor is far above that.

I for one am glad to have that additional "working space". A converter desiged with a 60db dynamic range would not be good. No headroom.

Now maybe we can understand why dynamic range is physically connected with bit resolution. It's not about "using all the bits". It's about using more bits than you need to represent the signal.

That's way below full scale.

I may have gotten some terms crossed up in here but I think you get the point. 24-bits (for sampling) are intentionally more bits than are necessary for accurate sampling.

Applying DSP and other things after tracking are another story.

Lawrence
Old 30th June 2006
  #202
Gear Head
 

Quote:
Originally posted by Dave Derr:

Thats a hard question because it depends on so many things, like what Daw you have, and how aware you wanna be of possible clipping. If you have Protools HD, I'm beginning to think you have to be careful, and possibly attenuate at or before any plug in. Some plugs have headroom built in. Also, I have personally had effects busses and other busses clip, but it wasnt so hard to work around.
Dave, I'm on Cubase SX, and other than a few effects, i mix entirely ITB. My question was whether or not switching to 32-bit floating point internal resolution in cubase in a session comprised of 24-bit files would have a negative effect on my mix (for example if i lowered the resolution to 16 bits from 24 bits without dithering).

P.S. Lawrence, well put.
Old 30th June 2006
  #203
The Distressor's "daddy"
 
Dave Derr's Avatar
 

Cubase SX

Sheesh I'd be lying if I could say for sure if you would have less problems switching to 32 bit. As long as the DSP is done in 32 bit floating point, I'm not sure a wider bus path would make much real world difference or not. It depends on how well the software is written and how it truncates. Perhaps you should ask other Cubase users. Sorry.
Old 30th June 2006
  #204
I think the term quantization error is what i needed to learn.

The reason I think the connection between dynamic range and bit depth is irrelevant is because, as you've pointed out, 16bits is adequate to cover our amplitude/dynamic needs.

Nature is fluid or continious. The ponential acutal amplitude of a signal in infinitely variable between zero dB and the maximum that the source is able to generate.

Clearly, the greater the bit depth the more total limtied, fixed number of little boxes digital can use to describe the amplitude of the signal.

Someone made a point earlier, that I'll try to paraphrase - no matter where you are in the scale of amplitude, regardless of how many total bits you're using, you rate of quatization errors is predetermined by the bit depth and will be the same regardless of the ampllitude. More volume may be more total bits, but it's not going to increase resolution interms of quantization errors. Using more bits does not decrease quantization errors, only a higher bit rate would do that. Is this correct?

If that's the case, then I'd have to change my position and say that as far as I know, within the digital realm, there's no advantage to recording with you're peaks as close to 0dBFS as you can get.

However, if you are using digital as a record and playback system, like a tape machine and otherwise working in the analog world (which is my perspective one this) then the only advantage is better signal to noise ratios, which based on tracks people have brought me, can still be an issue. Plus, I'm always tracking to tape first, so I've still got analog S/N to think about at the same time.

(before anyone goes to the trouble of pointing out that once the noise floor is captured it's fixed, I'll point out that there are times when you have to boos the signal when it comes back to to console and of course those circuits have noise that you'd be increasing.)

We've double our initial digital sampling rate twice at this era of digital recording (192k). I can't believe that increasing our bit rate and decreasing quatiation errors woudn't still provide audible improvements. Since there are no 48bit or 96bit converters, I can only speculate. Also, assuming the processing power exsisted, you'd be able to model and do mathematical processes with these more accurate numbers and when rinning through multiple processes/plugins/fader move have a truer signal.
Old 30th June 2006
  #205
Lives for gear
 

Dave, just as an FYI, your VU might peak at -2dbvu from "-20 dbFs" (calibrated to 18dBFs as many are) you'd definitely see the needle "flicker".

Quote:
So Lawrence, If I have a compressed vocal track that only has a dynamic range of 36dB, an 8 bit converter would be overkill?
You'll never approach anything close to 8 or even 16 bits tracking at the levels we've been talking about into a 24-bit system. You should know that.

Again... let's put this in the proper context. Noboby uses 8 bit converters...

We're talking about working within a 24-bit system. Recording ANYTHING into a 24-bit converter using only an effective 8 bits of resolution would be pointless, you'd probably have a hard time even hearing it it would be so far down.

24-bits = 16 million + steps 16 MILLION STEPS
These are your amplitude quantanization points at 24 bit. This is your "resolution".

8 bits = 256 steps Hmmm.... you didn't think 8 bits was a third of 24 bits did you?

Now tell me, at what level of the dBFS scale would you have to peak (from a signal's noise floor to it's peak on your daw input meter) to only have the bottom 256 steps of 16 million available to represent the waveform? What dynamic range (in the 24-bit system) would that be?

THE BOTTOM 256 steps of the 16 Million available in our 24-bit system. That's where you 8 bits are. You can't get them in the middle. You can only get them from the bottom up, at least from a metering standpoint.

Remember, this is a 24-bit converter so all 16 million steps are still there, you only want to use 256 so you can argue an effective 8-bit's of resolution in a 24-bit daw theory.

The moment the peak goes over the 256th step you're beyond 8 bits. You may not even see the meter flash.

Do you realize how low that signal would have to be to only have 256 of 16 million steps available from noise to peak? To record (in a 24-bit digital daw) a signal equivalent to an 8 bit signal? Are you freaking kidding me?

In your mind you can drop to 25-30dBFs and be getting 8 bits of effective resolution? Not even close. 16 bits is 65 thousand steps or so. Let's try your "lost resolution" argument on that one.

So let's divide our 16 million steps up and only take the bottom 65 thousand (or so) steps. What peak level is that? Try 1/258th or so of the scale. How far down do you have to go to even be limited to an effective 16 bits of resolution? To only use that bottom 1/258th or so (didn't do the exact math, 1/200th is silly enough)? That's how low you'd have to peak in a 24-bit system to be working with an effective limit of 16 bits of resolution.

Yeah, 1/200th or so... that's gotta be around -20.

An effective 8 bits of a 24-bit daw scale ? 1/65536th of the 24-bit scale. Sure, that'll happen. Visually chop up your daw meter into 65 thousand equal sections and mark the bottom one. That's it. 8bits. Can't use any other section because you have to cross more tha 256 steps to get there... it can only be the bottom 256 steps or bottom 65k steps for 16-bits.

NOT 1/3 rd of 24-bits like (3 * 8bits = 24bits)...

Since audio signals always start at infinity and go up in amplitude when they "sound off" in the daw and light the meters...

The ONLY way to have a signal in a 24-bit system be represented by the amount of quantanization steps to be considered limited to 16 bits of resolution (65 thousand steps) is for it TO USE THE BOTTOM 200th or so of the scale. The bottom 65k steps of the 16 million available steps.

The ONLY way to have a signal in a 24-bit system be represented by the amount of quantanization steps to be considered limited to 8 bits of resolution (256 steps) is for it TO USE THE BOTTOM 65536th of the scale. The bottom 256 steps of the available 16 million available steps.

Take 12-bits, 18 bits and extrapolate from there. Nobody recording at -15 peak is losing any "fidelity". You'd have to go down WAY farther than that. WAY down.

If this was not the case every time we faded a digital song it would sound like crap. By the time it hits those really low quantanization levels like 8 bit (during fading) it's inaudible already. -30 or so is right in the audible fade path. It should sound horrible if your "resolution" theory was correct. I suspect even there, during a fade it's hovering above 16 bits of resolution.

Please people... don't record at -15, you might only get 21-22 bits. That would destroy pro audio quality as we know it.

Even if it is (as has been said over and over and over) way beyond the dynamic range of your audio ... and more than enough bits for that dynamic range in our 24 bit daws to capture the audio faithfully.

36db vocal at a true 8 bits of resolution in a 24-bit daw? That would sound pretty bad. If you could even hear it.

Lawrence
Old 30th June 2006
  #206
Gear Nut
 

Quote:
Originally Posted by swan
Dave, I'm on Cubase SX, and other than a few effects, i mix entirely ITB. My question was whether or not switching to 32-bit floating point internal resolution in cubase in a session comprised of 24-bit files would have a negative effect on my mix (for example if i lowered the resolution to 16 bits from 24 bits without dithering).

P.S. Lawrence, well put.
Well, I'm not Dave, but I'll answer anyway.

Yes, use 32 bit float. Continue to record 24 audio bit files. When they are processed in various ways they will take advantage of the 32 bit resolution. As Dave says, having the DSP at 32 bit could only be a good thing. It doesn't negatively affect the 24 bit files in any way if that is your concern. I'm sure this is how I've worked in Cubase before and I think DP as well. Recently I set up Sonar with the same bit profile - recording 24 bits with 32 bit resolution for DSP. Sonar has 64 bit internal res as an option. I turned it back to 32 because I could hear no difference. What internal resolution have you been using? Cubase maxes at 32 so I assume you have been using 24 bit float?

-Naren
Old 30th June 2006
  #207
Lives for gear
 

Cubase SX (and Nuendo) work internally at 32-bit float. They also give you the option of actually recording 32-bit float files. I never really got the point of that when using 24-bit converters but it's there as an option.

I suppose it would be near impossible to clip digital 0 when recording that way?

Everything that comes into the mix engine from disk (or live) is converted to 32-bit float until it hits an output bus. It doesn't really come into play until you move a fader or apply dsp though... until then those extra bits are just zeros filling up space. That includes 16-bit files, mp3's dragged into the timeline, Aiff, whatever. Once you add dsp (math) some of those zeros become active until they're truncated (or dithered) to 24-bits at the output bus so you can hear it.

Now it could be useful to render a mix to 32-bit float IF you plan on doing some additional editing in another 32-bit float application like Wavelab. By doing that you will import the mix into Wavelab exactly as was in the daw without any truncation. Wavelab will directly load 32-bit float audio files from disk.

All of your dsp calculations from the 32-bit float daw would still be 100% intact. Problem is ... it still has to be truncated (or dithered with 24-bit dither) before you hear it from the speakers so I'm not sure how helpful that actually is.

I just keep a 24-bit dither (with no noise shaping) strapped across the last slot of the mix bus during mixing and switch it to 16-bit before printing a CD compatible file.

Works for me.

Lawrence
Old 30th June 2006
  #208
Why does 16million seem like a lot of steps?

Sound is a phenomenon of nature and in terms of nature, 16 million is a very small number.
Old 30th June 2006
  #209
Quote:
Originally Posted by Mike Caffrey
Why does 16million seem like a lot of steps?

Sound is a phenomenon of nature and in terms of nature, 16 million is a very small number.
So is the number 12 but think of how many songs in western music have been created with the same 12 notes.
Old 30th June 2006
  #210
Lives for gear
 

Quote:
Originally Posted by Mike Caffrey
Why does 16million seem like a lot of steps?

Sound is a phenomenon of nature and in terms of nature, 16 million is a very small number.
Try this number .... 4.2 billion !

That's the number of possible amplitude levels that a 32-bit converter would allow... if anybody made one.

I think we can safely say we don't need that many! heh

Lawrence
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Similar Threads
Thread
Thread Starter / Forum
Replies
thejook / So much gear, so little time
15
Ziggy!! / So much gear, so little time
12
DeadPoet / So much gear, so little time
1

Forum Jump
Forum Jump