R/E/P Community

Please login or register.

Login with username, password and session length
Advanced search  

Pages: 1 2 3 [All]   Go Down

Author Topic: Summing  (Read 18858 times)

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Summing
« on: October 17, 2006, 10:29:04 pm »

i've been seeing some arguments that sending tracks directly to a/d outputs from within a DAW sounds/sums better than sending tracks to a bus or master bus and then to the a/d output.
isn't it all 1's and 0's? any validity to this?
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #1 on: October 18, 2006, 05:51:56 pm »

dobster wrote on Wed, 18 October 2006 03:29

i've been seeing some arguments that sending tracks directly to a/d outputs from within a DAW sounds/sums better than sending tracks to a bus or master bus and then to the a/d output.
isn't it all 1's and 0's? any validity to this?


Did you mean to say DA (instead of AD?)

Regards
Dan Lavry

Logged

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Re: Summing
« Reply #2 on: October 18, 2006, 07:31:45 pm »

yes dan, absolutely. my mistake..apologies. I meant D/A, thanks for the oversight

i've been seeing some arguments that sending tracks directly to d/a outputs from within a DAW sounds/sums better than sending tracks to a bus or master bus in the DAW and then to the d/a output.
isn't it all 1's and 0's? any validity to this?
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #3 on: October 19, 2006, 01:43:54 pm »

dobster wrote on Thu, 19 October 2006 00:31

yes dan, absolutely. my mistake..apologies. I meant D/A, thanks for the oversight

i've been seeing some arguments that sending tracks directly to d/a outputs from within a DAW sounds/sums better than sending tracks to a bus or master bus in the DAW and then to the d/a output.
isn't it all 1's and 0's? any validity to this?


The question itself is about subjective opinions and tastes - what "sounds better". I can not answer it in a subjective manner. I can see a lot of explanations as to why one may prefer the summing done in the analog world. They may like a particular sound of a DA, an analog mixer, and AD... I also heard some people preferring the smooth response and feel of the analog sliders (compared to on screen sliders and knobs).

If your question is strictly about summing, then it is difficult for me to see any validity to arguments against doing it in the digital domain. Once the data is in the digital domain, summing in digital eliminates the need for additional DA, analog mix and AD processing, which can only cause more deviation from the original waveform.

But if the question goes beyond summing, and some of the processing is done in analog (beyond just summing), then we are getting into comparing analog processing to digital processing. As a rule, linear processing (such as summing, EQ, re-verb...) is well suited for digital work. At the same time, any processing that calls for non linearity (such as compression, limiting, tube emulation...) may be better in analog, because a non linear processing in digital generates alias energy, which is very non musical. I am not condemning all non linear digital processing, but I am suggesting that it needs to be "carefully evaluated". While compression in digital may or may not be OK (implementation dependent), hard limiting in digital is probably bad news no matter what you do to try and fix it...

But even the above comments may not stand up to "what sounds good". For me, high degree of aliasing is always real bad news. The alias energy, unlike harmonics of musical instruments, does not fall on frequencies that are multiple of the fundamental pitch. So it sounds horrible to me, but in fact even that is a subjective opinion.

Regards
Dan Lavry
http://www.lavryengineering.com        
Logged

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Re: Summing
« Reply #4 on: October 19, 2006, 04:16:28 pm »

Dan thanks for the response

I know my question is easily confused with "digital summing vs. analog summing: which is better?" But, I'm actually not concerned with summing "outside the box" but whether or not summing tracks through bus's ITB is a good idea or not. And by that I mean, should bussing be kept to a minimum? Some of the advocates for NOT using bus's in the DAW claim that the audio sounds "mushier" and/or smaller than simply letting the tracks  in the DAW sum to the D/A (the track outputs directly to an output). Is there any validity to this? Does that process even exist? In other words, is the summing happening in the DAW at all times anyway when one remains digital? There is no summing at the converters? Just, conversion to analog for monitoring?

You have though, in your response, sparked another question that I might as well ask now.

You, having said this...

" At the same time, any processing that calls for non linearity (such as compression, limiting, tube emulation...) may be better in analog, because a non linear processing in digital generates alias energy, which is very non musical."

I'd like to know then, how would you feel about the idea of recording the output of a digital track in the DAW with some type of non-linear processing plugin in realtime? I guess I'm asking, do non-linear processes only present alias artifacts when processed digitally and rendered? Or do they also present those artifacts while audio is playing through the plugin listening to it in realtime? Regardless, if so or if not, is there any benefit to letting audio play through the plugin and capturing that realtime playback back into the DAW? Assuming of course and putting aside the question are the converters worth it, etc. This is directed strictly towards the process in and of itself
thanks
Logged

ruffrecords

  • Full Member
  • ***
  • Offline Offline
  • Posts: 130
Re: Summing
« Reply #5 on: October 22, 2006, 12:41:55 pm »

dobster wrote on Wed, 18 October 2006 03:29

i've been seeing some arguments that sending tracks directly to a/d outputs from within a DAW sounds/sums better than sending tracks to a bus or master bus and then to the a/d output.
isn't it all 1's and 0's? any validity to this?

Some. Going direct means essentially the bits are sent directly to the D/A. Going via the bus means the bits get multiplied by the fader value. Assuming this value is 1 then it depends on the DAW as to what happens. A smart DAW might realise x1 means do nothing and it would do nothing. Other DAWs might just do the x1 multiplication. Depending on the DAW this might or might not give exactly the same output as the direct connection. It depends on how math is handled in the DAW and whether or not it has any rounding errors. This means some samples may not come out at exactly the same value as they went in.

Ian
Logged

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Re: Summing
« Reply #6 on: October 22, 2006, 09:56:53 pm »

hey Ian...thanks

so, how does one quantify whether or not a DAW possesses there rounding errors or not? and if that bus fader is at "unity" then is there much of a difference then going directly to the D/A directly?
Logged

Jon Hodgson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1854
Re: Summing
« Reply #7 on: October 23, 2006, 05:04:24 am »

dobster wrote on Mon, 23 October 2006 02:56

hey Ian...thanks

so, how does one quantify whether or not a DAW possesses there rounding errors or not? and if that bus fader is at "unity" then is there much of a difference then going directly to the D/A directly?


Any DAW will have rounding errors

HOWEVER, in setting the level and summing (as opposed to say, EQ) the process is so simple (one multiply and one add per channel) that the error is only going to be half a bit maximum. You are not going to hear a half bit error in a 24 bit system.

Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #8 on: October 23, 2006, 04:00:15 pm »

dobster wrote on Thu, 19 October 2006 21:16

Dan thanks for the response

I'd like to know then, how would you feel about the idea of recording the output of a digital track in the DAW with some type of non-linear processing plugin in realtime?
thanks


Aliasing occurs the instance one tries to include energy (signal) containing frequencies above half the sample rate (at a specific point in the circuit).
To avoid aliasing when converting with an AD, you must make sure that there is no energy there at frequencies above Nyquist (half the sample rate of the converter input circuitry - called the modulator). You can do it by pre filtering (analog filter) the high frequencies, by operating the modulator at higher rate (oversampling) to a high enough rate.... In practice you get to do both (filter and fast operating modulator).

You can do all sorts of processing in analog, and there will be no aliasing. Non linear process will generate high frequency, which when converting to analog can be dealt with as stated above by:
1. Pre filtering
2. Faster rate of conversion (oversampled modulator).

But when you are in the digital domain, as soon as you did a non linear operation, the aliasing is already there, and whatever falls on the audible range can not be removed, it is too late, because the non linear process and the aliasing happen simultaneously. You can help it some by doing the process at high oversampling rate, but whatever does alias to the audio, stays there. You can not filter it before the fact, you can not do it after the fact.

So you can do non linear processing in analog. If you want it in digital, make sure that the level of aliasing is acceptable.

Regards
Dan Lavry
http://www.lavryengineering.com
Logged

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Re: Summing
« Reply #9 on: October 23, 2006, 08:53:10 pm »

thanks Jon and Dan...

jon, just to clarify, what if you bus @ unity then? meaning, no level adjustment at the bus?

dan, is it possible to create a plugin that upsamples momentarily locally on a track to process and then downsample again? or is that just a dumb idea?
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #10 on: October 24, 2006, 02:30:55 am »

dobster wrote on Tue, 24 October 2006 01:53

thanks Jon and Dan...

jon, just to clarify, what if you bus @ unity then? meaning, no level adjustment at the bus?

dan, is it possible to create a plugin that upsamples momentarily locally on a track to process and then downsample again? or is that just a dumb idea?


Not dumb at all. That is the best way, and maybe the only way to overcome the aliasing. The problem is that depending on the nature of non linearity, you may have to up sample to 10MHz or beyond, and that is heavy computational requirement.
Why so fast? It is true that as a rule, the overtones made by the non linearity decay as you go tom higher frequency, but when we impose a logarithmic curve on the harmonic decay, it does not decay fast enough. The ear is nearly logarithmic in terms of response to amplitude, so a decay to say 1% is really only "40dB down", and even a decay to .1% is only 60dB down....

So one needs to evaluate how much aliasing is there on a "case by case" basis. As a rule, I think that a real "hard" non linearity, such as a digital hard limiter, would be nearly hopeless. I suspect that a non linearity of say a digital compressor may be handled well, because most of the non linearity is at very slow speed (the envelope)....

Regards
Dan Lavry
http://www.lavryengineering.com    
Logged

ruffrecords

  • Full Member
  • ***
  • Offline Offline
  • Posts: 130
Re: Summing
« Reply #11 on: October 24, 2006, 05:57:27 am »

Jon Hodgson wrote on Mon, 23 October 2006 10:04


Any DAW will have rounding errors

HOWEVER, in setting the level and summing (as opposed to say, EQ) the process is so simple (one multiply and one add per channel) that the error is only going to be half a bit maximum. You are not going to hear a half bit error in a 24 bit system.

The internal word length (number of bits used) will be longer than the required final answer. As you say, a maximum half bit rounding error may occur when this is converted back to the required word length.

There is another source of error though and that is in the math itself. Some DAWs use integer math but with large word lengths (as much as 56 bits) and others use a floating point representation of the data. The results form these will themselves contain small errors but usually they are second order compared to the rounding error in converting back to the required word length. However, even some simple multiplications where the fractiional part of the answer is close to o.5 bit will occasionally have a small error which casues the subsequent rounding to go in the 'wrong' direction giving occasional one bit errors.

Ian
Logged

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Re: Summing
« Reply #12 on: October 24, 2006, 12:44:14 pm »

Dan, excellent clarification... So, what i gather is, a good solution to digital processing of non-linearity overall is to either use outboard analog gear to perform those operations assuming great converters though. Or, and forgive me if this still isnt the right idea, use a quality upsampler, export the tracks to the upsampler using great converters, once again, and then into the DAW to process? But this as you said, still requires alot of CPU right? Because then the local sample rate must change to match those newly converted tracks.

damn it, somebody make that upsampling plugin!  Rolling Eyes


So Ian, thanks for that and as I understand it then, floating point should be more superior to fixed, no? And even better is to use a high native bit depth and on top a high floating point for processing within that DAW
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #13 on: October 24, 2006, 02:00:57 pm »

dobster wrote on Tue, 24 October 2006 17:44

Dan, excellent clarification... So, what i gather is, a good solution to digital processing of non-linearity overall is to either use outboard analog gear to perform those operations assuming great converters though. Or, and forgive me if this still isnt the right idea, use a quality upsampler, export the tracks to the upsampler using great converters, once again, and then into the DAW to process? But this as you said, still requires alot of CPU right? Because then the local sample rate must change to match those newly converted tracks.

damn it, somebody make that upsampling plugin!  Rolling Eyes


So Ian, thanks for that and as I understand it then, floating point should be more superior to fixed, no? And even better is to use a high native bit depth and on top a high floating point for processing within that DAW



Hi again,

First, regarding your comment about a loss of 1/2 bit at the final truncation:
It is best to have that final truncation done with dither. At first glance, that 1/2 least significant bit may not "look as bad". But it is worse then it looks, and here is why:

If you add dither, the overall noise energy will be a little higher, but it will be "spread out" evenly across the spectrum.

If you do not add dither, the overall noise will be lower, but you may end up with "spikes on the FFT" - concentration of energy at harmonics frequencies of the tones. A tiny sine wave, when quantized by say 2 levels only, is in fact a square wave, thus it has a bunch of odd harmonics... It also sounds like a square wave... And also, without dither, there is "noise modulation" - the noise floor changes with the music (instead of being constant). As a rule you will not hear the lack of dither at a 24 bit level, so the above may not be so impotent, but it is a good form to dither the truncations.

Regarding the "upsampling plugging". Again, as a rule we are not talking about up-sampling to 192KHz or 384KHz... An up sampler for say "hard limiter" (very non linear operation) may call for up sampling at 10MHz or more. I heard of such stuff being done for some speaker project. The problem is far greater AFTER the up sampling. Doing DSP on say 100Hz sine wave sampled by say 44.1Khz is in fact dealing with a ratio of 441 to 1 (sample rate to signal). In other words, it takes 441 data points to "express" a single cycle of 100Hz. Now lets try to store 1 cycle of 100Hz with 10MHz sample rate - it would take 100000 samples to do that!

What I am getting at is a general comment: the required DSP power grows very fast as you go to higher sample rates. The growth is often far beyond linear. The very brilliant Dr. Richard Cabot did a great presentation on that subject in an AES workshop I was chairing about 2 years ago.

Regards
Dan Lavry
http://www.lavryengineering.com
Logged

ruffrecords

  • Full Member
  • ***
  • Offline Offline
  • Posts: 130
Re: Summing
« Reply #14 on: October 24, 2006, 05:19:22 pm »

danlavry wrote on Tue, 24 October 2006 19:00


It is best to have that final truncation done with dither. At first glance, that 1/2 least significant bit may not "look as bad". But it is worse then it looks, and here is why:

If you add dither, the overall noise energy will be a little higher, but it will be "spread out" evenly across the spectrum.

I agree any final truncation should be done with dither for the best sound. The interesting word though is "final". I suspect 'final' applies to any mix within the DAW for example just after a bounce. If it does then it implies noise will build up with successive digital bounces just the way it used to in analog days.

Ian
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #15 on: October 24, 2006, 05:48:56 pm »

ruffrecords wrote on Tue, 24 October 2006 22:19

danlavry wrote on Tue, 24 October 2006 19:00


It is best to have that final truncation done with dither. At first glance, that 1/2 least significant bit may not "look as bad". But it is worse then it looks, and here is why:

If you add dither, the overall noise energy will be a little higher, but it will be "spread out" evenly across the spectrum.

I agree any final truncation should be done with dither for the best sound. The interesting word though is "final". I suspect 'final' applies to any mix within the DAW for example just after a bounce. If it does then it implies noise will build up with successive digital bounces just the way it used to in analog days.

Ian


Hi Ian,

In principle, any truncation, should be dithered, because as I pointed out, not truncating may lead to energy concentration in some specific frequencies. The energy concentration I am talking about may have peaks far above the noise floor of a dithered signal. Again, the average value and the rms value of the un dithered signal may be lower, but the peaks of the un dithered signal may be higher then the dithered noise floor.

So in principle, any word length reduction (truncation)should be dithered.

There are times when one decides not to dither, such as when the noise content is already dominating the lower bits of the truncated signal. But from a pure standpoint, it goes beyond just the final truncation.

However, when it comes to the final truncation, the use of dithered noise shaping should be reserved for the final truncation.
Noise shaping dither is far better then dithers without noise shaping, because the dither noise is being pushed away from hearing sensitive ear regions (such as 1-3KHz) to places we do not hear as well (such as say 16-22KHz). But one can not noise shape again and again, you can get the desired shaping by doing it only once, in the final truncation.

Regards
Dan Lavry
http://www.lavryengineering.com
Logged

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Re: Summing
« Reply #16 on: October 24, 2006, 10:06:33 pm »

Dan from what you are saying, I am now picturing the notion of all truncation should have dither as something akin to analog. Because somebody mentioned before the dither adds up, and you mentioned that in essence that is better than the peaks and prevalence of odd harmonics, maybe that dither adding up emulates the sort of character analog gives to audio. I dont mean exactly but along the idea of sometimes imperfections of a system can sound nice to cover really ugly imperfections of audio. So is the case with dithering all truncations. Am I off here?

I should ask, when you say truncation does that include going from the inherent floating bit of a DAW down to the file bit depth, say 64 or 32 back down to 24? should a simple dither be added here?

using shaped dither only once makes perfect sense however.
Logged

Jon Hodgson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1854
Re: Summing
« Reply #17 on: October 25, 2006, 05:32:21 am »

dobster wrote on Wed, 25 October 2006 03:06

I should ask, when you say truncation does that include going from the inherent floating bit of a DAW down to the file bit depth, say 64 or 32 back down to 24? should a simple dither be added here?


In purist terms, yes. In the real world, it will probably make no difference (depending on the signal and/or what you do with it after the conversion).

If you're going to send the result to a DAC and out through your playback system, then you'll be adding more than 1 bits worth of analogue noise anyway, so that will completely swamp any artifacts from truncation.

If you were converting to 16 bits it would be a different story. It would also be different if you took that 24 bit integer stream applied gain to it (effectively the same thing).
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #18 on: October 25, 2006, 12:30:53 pm »

Jon Hodgson wrote on Wed, 25 October 2006 10:32

dobster wrote on Wed, 25 October 2006 03:06

I should ask, when you say truncation does that include going from the inherent floating bit of a DAW down to the file bit depth, say 64 or 32 back down to 24? should a simple dither be added here?


In purist terms, yes. In the real world, it will probably make no difference (depending on the signal and/or what you do with it after the conversion).

If you're going to send the result to a DAC and out through your playback system, then you'll be adding more than 1 bits worth of analogue noise anyway, so that will completely swamp any artifacts from truncation.

If you were converting to 16 bits it would be a different story. It would also be different if you took that 24 bit integer stream applied gain to it (effectively the same thing).



Jon and I are in agreement about "purist" vs, "practical", and of course it brings a question: where does one draw a "line" between the two.

Let's examine adding random noise into a signal that already contains random noise.
Clearly, if the already existing random noise is higher in amplitude then the dither noise, the dither will be "swamped", "overpowered", "buried" by the already present noise in the signal.  
And if the signal itself has very little noise, then the dither can "rule", "overpower" the noise in the signal.

That is why Jon is suggesting that truncating to 24 bits and to 16 bits are different stories. As a rule, the signal out of a converter will have some of the least significant bits contain noise. You can take such signals and run them through all sorts of processing in a DAW, with very wide mix bus (it can be 1000 bits wide), but if bit 20-24 recorded noise, the noise is still there at bit 20-24, no matter what you do. So setting the dither at say bit 24 may make little sense. But assuming that bit 16 was noise free, it makes sense to dither at bit 16 when you truncate to 16 bits.
The need for dither exist only at the point that a signal is noise free.  
 
So that brings us back to: how many bits is noise free? That depends a lot on the micpre input noise floor, on the gain setting of the micpre (see an article on my company web under tutorials), and it may depends on the AD. In practice, a recorded sound will have at lease the last 4 bits carry noise and no sound at all. There is still a long way from 20 bits "true", to 16 bits format, so dither does have it's place...

Regards
Dan Lavry
http://www.lavryengineering.com
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #19 on: October 25, 2006, 01:05:29 pm »

Here is a plot showing what I said earlier.
It is an FFT showing a 3 LSB peak to peak 1KHz sine wave tone. Both the red an blue plots share the same tone at 1KHz (the blue 1KHz tone "covers" the red 1KHz tone almost completely so it is not easy to see, so take my word for it).

But other then the energy at 1KHz, the RED (Truncated to 16 bits with NO DITHER), and the BLUE (Dithered and truncated to 16 bits) are very different plots.

The RED (no dither) has less total noise, and the overall noise floor is lower then the BLUE (dithered).

But the RED (no dither) has some peaks that go way above the Blue noise floor. In this example, there is a "big peak" around 9KHz, way above the dithered noise floor.

Note that those peaks keep changing with the music in a very non musical manner. Also, such peaks "disappear" when you raise the signal amplitude to a high enough level. So dither is not going to help loud music, it helps signals such as "the tail end" of a decaying piano note....


index.php/fa/3638/0/


The point is: While dither adds to the overall noise, the truncation without dither result with some undesirable energy that peaks way above the dithered signal noise. The dithered noise is "spread out" sort of evenly accross the audio. You hear it as noise. The "energy peaks" of the un dithered signal "stand out". They don'y belong there and they have a "non random sound" which is undesirable.  

Regards
Dan Lavry
http://www.lavryengineering.com

 
Logged

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Re: Summing
« Reply #20 on: October 25, 2006, 02:22:50 pm »

Right, non random by its nature will even stick out audibly speaking, yes? So the dither sort of act as stilts while keeping a "constant" random flow under the actual music; keeping it steady in some way. I've listen to a file with and without dither and with the dither, certain harsh points were gone? or, not as harsh? point is, it sounded more natural with the dither. Thats why it makes me think of analog and possibly the kind of "stilt" and electronic noise personality analog circtuiry possesses.
From what I learned here, i will assume any one track within a mix i process outside the box I will not look to add noise or dither to ITB because of the conversion adding its own dither.
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #21 on: October 25, 2006, 05:14:45 pm »

dobster wrote on Wed, 25 October 2006 19:22

Right, non random by its nature will even stick out audibly speaking, yes? So the dither sort of act as stilts while keeping a "constant" random flow under the actual music; keeping it steady in some way. I've listen to a file with and without dither and with the dither, certain harsh points were gone? or, not as harsh? point is, it sounded more natural with the dither. Thats why it makes me think of analog and possibly the kind of "stilt" and electronic noise personality analog circtuiry possesses.
From what I learned here, i will assume any one track within a mix i process outside the box I will not look to add noise or dither to ITB because of the conversion adding its own dither.


Be careful not to jump into the wrong conclusion.

You first need to know how much (or little) noise you already have in the signal. As  rule, I would not worry too much about dither when truncating to 24 bits. I would certainly dither when going to 16 bits. There are a lot of "cases in between".
Usually, when in doubt, add dither. It does not raise the noise level that much, and the possible benefits tend to far outweigh the little increase in noise.

Also, adding dither again and again is not as bad as it sounds initially. Say you added dither and it cost you a little by increasing the noise floor by XdB. The next time you add dither (for the same word length) the noise will not go up by another Xdb. The second time it will go up by a lot less then Xdb. It does not add linearly. Losing another XdB (for a total of 2X dB) will take adding dither 4 times. Losing 3XdB (where XdB is dithering once) will require 8 times of adding dither...

So once you added dither, much of the increase in noise floor is already done...  It is a bit complected, I wish I could simplify it, sorry....

Analog noise and what it does is a whole subject to itself, maybe another for thread another time..

regards
Dan Lavry
http://www.lavryengineering.com


Logged

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Re: Summing
« Reply #22 on: October 25, 2006, 09:18:30 pm »

yes I understand. i think i was making digital analogous to analog but without the coloration and circuitry noise and harmonics of analog. Smile if that makes sense.. i think it makes sense then, that without previous noise build up or dither 20x, some dither and some times, while remaing 24 bit or higher can be helpful.
Logged

ruffrecords

  • Full Member
  • ***
  • Offline Offline
  • Posts: 130
Re: Summing
« Reply #23 on: October 26, 2006, 05:37:16 am »

danlavry wrote on Tue, 24 October 2006 22:48


Hi Ian,

In principle, any truncation, should be dithered, because as I pointed out, not truncating may lead to energy concentration in some specific frequencies. The energy concentration I am talking about may have peaks far above the noise floor of a dithered signal. Again, the average value and the rms value of the un dithered signal may be lower, but the peaks of the un dithered signal may be higher then the dithered noise floor.

So in principle, any word length reduction (truncation)should be dithered.

Hi Dan,

I agree. I wonder what happens though in a typical DAW? Suppose you make a stereo sub mix of a drum kit, for example; presumably this sub mix has been dithered during its creation? Repeat this for other elements of the music then make a final stereo mix of the sub mixes. Presumably dither is applied again? I know the answer will be DAW specific but there must be some general principles to follow.

Ian
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #24 on: October 26, 2006, 12:55:09 pm »

ruffrecords wrote on Thu, 26 October 2006 10:37

danlavry wrote on Tue, 24 October 2006 22:48


Hi Ian,

In principle, any truncation, should be dithered, because as I pointed out, not truncating may lead to energy concentration in some specific frequencies. The energy concentration I am talking about may have peaks far above the noise floor of a dithered signal. Again, the average value and the rms value of the un dithered signal may be lower, but the peaks of the un dithered signal may be higher then the dithered noise floor.

So in principle, any word length reduction (truncation)should be dithered.

Hi Dan,

I agree. I wonder what happens though in a typical DAW? Suppose you make a stereo sub mix of a drum kit, for example; presumably this sub mix has been dithered during its creation? Repeat this for other elements of the music then make a final stereo mix of the sub mixes. Presumably dither is applied again? I know the answer will be DAW specific but there must be some general principles to follow.

Ian


I am not sure what all the DAW do. Ideally, the DAW would keep everything without truncation until the end of the process, when one MUST truncate to "fit" the data into some format (such as AES or SPDIF).

I do not know what word length is used for the "sub mix". Ideally one can keep it stored with wide words, such as offered by the mix bus. Dither is needed only when you reduce the word length (less bits).  

Say you have some sub mix of some word length, and you want to add to it another track, or another sub mix, or do some EQ... The DAW should allow you to do such operations, and when the operations call for more bits (to the left and or to the right), so at this point there is no reason to truncate, thus no dither is required.

From dither standpoint, I would treat a sub mix the same way I treat a single track. It has some given number of bits. The more significant bits (hopefully 15-21 bits) carry music, the lower bits (22 to whatever) carry noise.

I have said it before a number of times, but here again:
There is the issue of bits for digital processing and DAW.
There is the issue of bits for conversion.

The DAW needs a lot of bits, to allow adding tracks, amplifying / attenuating / EQ / reverb.... all take "work space", so you need a lot of bit to do the processing. Once done, you reduce the number of bits.

The conversion bits is a different story. Say your music is recorded with a 24 bit format where the noise floor is such that only the top 16 bits carry music. In this example, the bottom 8 bits are of no value. All that noise is due to limitations such as mic pre noise, AD noise or what not...

Just because one loads such a file into a DAW with say 48 bits bus, does not mean at all that the extra DAW bits will make the music on that track end up with better then 16 bit. It will not. Say you decide to do an EQ and boost the 10-20KHz by 6dB. The computation will immediately call for a lot more bits, one on the MSB side, and many on the LSB. However, when you boost the music signal over 10-20KHz, you also boost the noise over 10-20KHz...

So how many bits? For conversion, 20 bits result out of an AD is very rare! For DAW, 20 bits is very limiting, you need much more then that....

Regards
Dan Lavry
http://www.lavryengineering.com  
   
Logged

ruffrecords

  • Full Member
  • ***
  • Offline Offline
  • Posts: 130
Re: Summing
« Reply #25 on: October 26, 2006, 05:40:42 pm »

danlavry wrote on Thu, 26 October 2006 17:55

I am not sure what all the DAW do. Ideally, the DAW would keep everything without truncation until the end of the process, when one MUST truncate to "fit" the data into some format (such as AES or SPDIF).

snip

I have said it before a number of times, but here again:
There is the issue of bits for digital processing and DAW.
There is the issue of bits for conversion.

Of course, it makes sense for the DAW to use as many bits as necessary until it has to dither. I suppose on PC DAWs with unlimited tracks and FX this is mostly not an issue because dither probably therefore only needs to be applied to the final mix. Everything else is just processing of the original data.

I suppose my interest in intermediate dithering is born out of the way I tend to work plus the fact that I have done a lot of bouncing in 40 years of recording. I don't use a PC, I use a stand alone DAW, an AKAI DPS24. For various reasons I tend to bounce and print several submixes. These, along with the remaining tracks are mixed to the final stereo mix. Clearly it is important to the quality of my final mix that the printed bounced sub mixes are dithered, and not truncated, before they are printed.

Ian
Logged

dobster

  • Full Member
  • ***
  • Offline Offline
  • Posts: 152
Re: Summing
« Reply #26 on: October 26, 2006, 09:56:46 pm »

i know this is a dumb question, but i still wonder where floating point falls? doesnt floating point imply the virtual bit depth at which the daw processes and not the actual file bit depth? so one could have a 24 bit file bit depth but processes at 32 bit float. in this case, whats happening between the 32 bit float processing point and after the rendering down to the 24 bit depthe actual file? does a simple dither (not shaped dither which should be for the final product)apply here? is that even considered truncation?  
Logged

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #27 on: October 27, 2006, 01:05:29 pm »

ruffrecords wrote on Thu, 26 October 2006 22:40

danlavry wrote on Thu, 26 October 2006 17:55

I am not sure what all the DAW do. Ideally, the DAW would keep everything without truncation until the end of the process, when one MUST truncate to "fit" the data into some format (such as AES or SPDIF).

snip

I have said it before a number of times, but here again:
There is the issue of bits for digital processing and DAW.
There is the issue of bits for conversion.

Of course, it makes sense for the DAW to use as many bits as necessary until it has to dither. I suppose on PC DAWs with unlimited tracks and FX this is mostly not an issue because dither probably therefore only needs to be applied to the final mix. Everything else is just processing of the original data.

I suppose my interest in intermediate dithering is born out of the way I tend to work plus the fact that I have done a lot of bouncing in 40 years of recording. I don't use a PC, I use a stand alone DAW, an AKAI DPS24. For various reasons I tend to bounce and print several submixes. These, along with the remaining tracks are mixed to the final stereo mix. Clearly it is important to the quality of my final mix that the printed bounced sub mixes are dithered, and not truncated, before they are printed.

Ian


I have not examined in detail the internal working of a DAW, but at first glance, I would think that more bits calls for much more then then just more storage and a wider buss. It may call for a more powerful compute engine.

Regards
Dan Lavry
http://www.lavryengineering.com
Logged

Ronny

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 2739
Re: Summing
« Reply #28 on: November 09, 2006, 03:39:45 pm »

ruffrecords wrote on Thu, 26 October 2006 17:40

danlavry wrote on Thu, 26 October 2006 17:55

I am not sure what all the DAW do. Ideally, the DAW would keep everything without truncation until the end of the process, when one MUST truncate to "fit" the data into some format (such as AES or SPDIF).

snip

I have said it before a number of times, but here again:
There is the issue of bits for digital processing and DAW.
There is the issue of bits for conversion.

Of course, it makes sense for the DAW to use as many bits as necessary until it has to dither. I suppose on PC DAWs with unlimited tracks and FX this is mostly not an issue because dither probably therefore only needs to be applied to the final mix. Everything else is just processing of the original data.

I suppose my interest in intermediate dithering is born out of the way I tend to work plus the fact that I have done a lot of bouncing in 40 years of recording. I don't use a PC, I use a stand alone DAW, an AKAI DPS24. For various reasons I tend to bounce and print several submixes. These, along with the remaining tracks are mixed to the final stereo mix. Clearly it is important to the quality of my final mix that the printed bounced sub mixes are dithered, and not truncated, before they are printed.

Ian



First order, reducing wordlength is truncation, with or without dither added. You dither the truncation, it's not a one or the other scenario. IOW, dithering a 24 to 16 bit wordlength is still a truncation, just a dithered truncation. Second, I also use stand alone HD-R's and prefer to print submixes and plug-in processing, as you do when I'm in a DAW, it eases CPU taxation and eliminates the power needed to run processors in real time, I get better performance in my systems that way, it takes a little longer, but I feel it's worth it. However not all printed processes while remaining at the native rate in the DAW need user applied dither. For example if you are operating at 32 bit, process and print at 32 bit, you won't need to add dither until you truncate the wordlength, no word reduction, no dither needed in most cases that I'm aware of. Someone can correct me if they know different.
Logged
------Ronny Morris - Digitak Mastering------
---------http://digitakmastering.com---------
----------Powered By Experience-------------
-------------Driven To Perfection---------------

Ronny

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 2739
Re: Summing
« Reply #29 on: November 09, 2006, 04:18:13 pm »

danlavry wrote on Fri, 27 October 2006 13:05




I suppose my interest in intermediate dithering is born out of the way I tend to work plus the fact that I have done a lot of bouncing in 40 years of recording. I don't use a PC, I use a stand alone DAW, an AKAI DPS24. For various reasons I tend to bounce and print several submixes. These, along with the remaining tracks are mixed to the final stereo mix. Clearly it is important to the quality of my final mix that the printed bounced sub mixes are dithered, and not truncated, before they are printed.

Ian


Quote:

 I have not examined in detail the internal working of a DAW, but at first glance, I would think that more bits calls for much more then then just more storage and a wider buss. It may call for a more powerful compute engine.

Regards
Dan Lavry
http://www.lavryengineering.com


The beauty of printing the process rather than running it in real time, is that the power of the computer engine need not be extravagant. The less power and processing speed, the longer it typically takes to print, but that's about it from my experience of printing processes, rather than trying to run a bunch in real time. The benefit is that the computer can take the time it needs to process the signal effectively. For real time processing, yes, the more process' that you run, the more power that you'll need, for printing processes not so important as the computer doesn't have to keep up with the power needed for real time processing. You have less chance of errors to occur and you completely eliminate the latency that you get from processing in real time chaining through various processors.  
Logged
------Ronny Morris - Digitak Mastering------
---------http://digitakmastering.com---------
----------Powered By Experience-------------
-------------Driven To Perfection---------------

danlavry

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 997
Re: Summing
« Reply #30 on: November 09, 2006, 06:56:55 pm »

Ronny wrote on Thu, 09 November 2006 21:18

danlavry wrote on Fri, 27 October 2006 13:05




I suppose my interest in intermediate dithering is born out of the way I tend to work plus the fact that I have done a lot of bouncing in 40 years of recording. I don't use a PC, I use a stand alone DAW, an AKAI DPS24. For various reasons I tend to bounce and print several submixes. These, along with the remaining tracks are mixed to the final stereo mix. Clearly it is important to the quality of my final mix that the printed bounced sub mixes are dithered, and not truncated, before they are printed.

Ian


Quote:

 I have not examined in detail the internal working of a DAW, but at first glance, I would think that more bits calls for much more then then just more storage and a wider buss. It may call for a more powerful compute engine.

Regards
Dan Lavry
http://www.lavryengineering.com


The beauty of printing the process rather than running it in real time, is that the power of the computer engine need not be extravagant. The less power and processing speed, the longer it typically takes to print, but that's about it from my experience of printing processes, rather than trying to run a bunch in real time. The benefit is that the computer can take the time it needs to process the signal effectively. For real time processing, yes, the more process' that you run, the more power that you'll need, for printing processes not so important as the computer doesn't have to keep up with the power needed for real time processing. You have less chance of errors to occur and you completely eliminate the latency that you get from processing in real time chaining through various processors.  



I agree, but many of my customers want to hear what they are doing as they are doing it (real time).

Say you are a mastering engineer and you have to do a sample rate conversion from 96KHz to 44.1KHz. Say you have 2 options:

1. Use a real time SRC, a bit costly to buy, and may yield good results.

2. Use an much less expansive software SRC that will take an hour or a few hours but yield very good results.

What would you do?

There are many people in the music production business that need to monitor every step they do by ear. In many cases, they like to do multiple tasks simultaneously and listen to what they are doing. That way, if something is "off", you can "fix it" immediately, instead of waiting for a long time.

I am not advocating one way or the other. It depends on needs,  style, experience...

Regards
Dan Lavry
http://www.lavryengineering.com
Logged

ruffrecords

  • Full Member
  • ***
  • Offline Offline
  • Posts: 130
Re: Summing
« Reply #31 on: November 10, 2006, 02:19:05 pm »

Ronny wrote on Thu, 09 November 2006 20:39

 However not all printed processes while remaining at the native rate in the DAW need user applied dither. For example if you are operating at 32 bit, process and print at 32 bit, you won't need to add dither until you truncate the wordlength, no word reduction, no dither needed in most cases that I'm aware of. Someone can correct me if they know different.

I guess this is DAW specific. If I am running a 24 bit project, the internal processing of tracks may be at 32 or some other number of bits but when I print a bounce it is back at 24 bits (certainly it is in my AKAI DSP24) so I hope but don't know for sure that the print is dithered from the internal bit length to the printed 24 bits. Not the sort of thing DAW manufacturers tend to spec IME.

Ian
Logged

Ronny

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 2739
Re: Summing
« Reply #32 on: November 15, 2006, 09:15:37 pm »

danlavry wrote on Thu, 09 November 2006 18:56

Ronny wrote on Thu, 09 November 2006 21:18

danlavry wrote on Fri, 27 October 2006 13:05




I suppose my interest in intermediate dithering is born out of the way I tend to work plus the fact that I have done a lot of bouncing in 40 years of recording. I don't use a PC, I use a stand alone DAW, an AKAI DPS24. For various reasons I tend to bounce and print several submixes. These, along with the remaining tracks are mixed to the final stereo mix. Clearly it is important to the quality of my final mix that the printed bounced sub mixes are dithered, and not truncated, before they are printed.

Ian


Quote:

 I have not examined in detail the internal working of a DAW, but at first glance, I would think that more bits calls for much more then then just more storage and a wider buss. It may call for a more powerful compute engine.

Regards
Dan Lavry
http://www.lavryengineering.com


The beauty of printing the process rather than running it in real time, is that the power of the computer engine need not be extravagant. The less power and processing speed, the longer it typically takes to print, but that's about it from my experience of printing processes, rather than trying to run a bunch in real time. The benefit is that the computer can take the time it needs to process the signal effectively. For real time processing, yes, the more process' that you run, the more power that you'll need, for printing processes not so important as the computer doesn't have to keep up with the power needed for real time processing. You have less chance of errors to occur and you completely eliminate the latency that you get from processing in real time chaining through various processors.  



I agree, but many of my customers want to hear what they are doing as they are doing it (real time).

Say you are a mastering engineer and you have to do a sample rate conversion from 96KHz to 44.1KHz. Say you have 2 options:

1. Use a real time SRC, a bit costly to buy, and may yield good results.

2. Use an much less expansive software SRC that will take an hour or a few hours but yield very good results.

What would you do?

There are many people in the music production business that need to monitor every step they do by ear. In many cases, they like to do multiple tasks simultaneously and listen to what they are doing. That way, if something is "off", you can "fix it" immediately, instead of waiting for a long time.

I am not advocating one way or the other. It depends on needs,  style, experience...

Regards
Dan Lavry
http://www.lavryengineering.com



If you would have asked me that 10 years ago, I would have said real time, every time, but now a days processors are so fast that printing a SRC takes little time on 2 track mixes. Printing dynamic processes or eq tweaks, just a matter of seconds, but you have to know what you want, there isn't the experimentation that real time processing allows, so it's not for every application. Typically when I run real time I'm incorporating analog processing, but if it's all digital, printing provides stability without latency issues and has benefits regarding less CPU taxation on large multi-track projects.
Logged
------Ronny Morris - Digitak Mastering------
---------http://digitakmastering.com---------
----------Powered By Experience-------------
-------------Driven To Perfection---------------

Pages: 1 2 3 [All]   Go Up