suming signals white papers

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

pucho812

Well-known member
Joined
Oct 4, 2004
Messages
14,831
Location
third stone from the sun
anyone have any good recommended reading on summing signals this is purely to prove what my ears have told me how plug in console emulators  like waves nonlinear or that  Slat VCC  offer negligible enhancement  to the audio signal coming out of the daw.
 
I don't think it's specifically the summing of the signals that's the thing to study, but the differences in distortion.
On a computer, using floating point arithmetic, to create a non-aliasing result, in a finite amount of time, you MUST limit the complexity of your distortion.
Having written many of my own plug-ins, I'm as aware of the trade-offs as anyone that tries to write such software will be.

i.e. it might be first order (a straight line, or a gain change), second order (a parabola), third order (a cubic curve) etc. or something with band-limited harmonics like a Fourier Frequency Transform.

However, none of these digital operations can, in a reasonable amount of processing time, thus far, emulate the infinite-series complexity of true analog distortions.
So all the waves, Slate VCC, etc. are all using the same algorithms (no matter what they say on their marketing speak), just tweaked in slightly different ways (which is where the real point of difference is; who can do the better job of making a result as artistically pleasing as scientifically/mathematically pleasing), but, fundamentally they are all the same, all an approximation, and all not really doing what is going on in an analog system at all.

The closest thing I've seen is the Diagonal Volterra Kernels in the Nebula plugin by Acustica Audio, HOWEVER, these are still heavily order-limited, and hence I believe are still only able to create distortions up to about order 7.

So your ears are not deceiving you, there are no plugins that currently exist that are doing anything particularly original or accurate compared to analog summing.
As you'd know even Spice gets it pretty wrong in terms of completely accurate high-order emulation, considering it mostly deals with combinations of linear approximations at local operational areas, and oversampled discrete difference solvers. The closest thing I've ever seen is the relatively recent work of simulating a single transistor down to the molecular/sub-atomic level, but that took many many supercomputers many many hours to do a simulation of a fraction of a second; and even then it still didn't totally match observable behaviour.
 
I'm trying some audio experiments mix then the mix with vcc.  if I play them both at the same time in the same session and have one out of polarity, the end result  should be that I  only hear what vcc and the others do as the rest of the information would be cancelling out.
 
pucho812 said:
I'm trying some audio experiments mix then the mix with vcc.  if I play them both at the same time in the same session and have one out of polarity, the end result  should be that I  only hear what vcc and the others do as the rest of the information would be cancelling out.

Correct.

The math for this is rather simple.

Take a signal X
Take a signal processor that generates the result f(X)

To calculate the work that the processor does that's different from the original, we calculate:

diff = f(X) - X

Which in your case would be to take the processed signal, and add it to an inverted version of the original signal; the result of which is the difference, and hence the "stuff done by" the plugin.

However, you have to be careful. This assumes that the signals are in phase, and that the processing plugin doesn't phase-shift the result.
Consider a processor that phase shifts at 1kHz by 180degrees. If you were to feed in a 1kHz sine-wave, your cancellation test would mysteriously produce a result twice as loud as the input.

So the cancellation test is ONLY useful if you know for sure that the processor you are testing is phase-coherent, else it doesn't tell you that much.
An FFT and phase comparison would be more enlightening.

Compare the FFT and phase diagrams of the original and processed signals and compare those.
The time-domain lies all the time ;-)
 
pucho812 said:
anyone have any good recommended reading on summing signals this is purely to prove what my ears have told me how plug in console emulators  like waves nonlinear or that  Slat VCC  offer negligible enhancement  to the audio signal coming out of the daw.

Please excuse my ignorance but what is a plug-in console emulator? Is this a processor that takes a dry digital stream and make it sound like a console?

JR
 
Short answer is yes... Their claim is that it does.

test last night between processed and unprocessed reveled the most bizarre  of things. Yes there was cancellation in the LF and HF but mids and upper mids were more pronounced not cancelling at all.
 
pucho812 said:
Short answer is yes... Their claim is that it does.
As an old console designer I would share that my design target for console design was to be the proverbial straight wire with gain and perhaps EQ. While that is easier said than done, I do not expect that the known performance flaws of analog console design are somehow worth emulating.

I could imagine characterizing a marginal console design based on expected performance deviation. For example a summing amp with inadequate loop gain margin would probably suffer more HF phase shift than audible distortion.  The nature of this phase shift depends on the open loop transfer function of the given opamp used for the sum amp. Most opamps with one pole compensation will have an open loop response with 90' phase shift above that compensation pole. The closed loop response is linearized by the loop gain margin (open loop vs closed loop gain). So one could imagine that a sum bus operating at high noise gain without commensurate open loop gain could suffer more phase shift with falling open loop gain at higher freqeency. This is known to designers (or should be) and I mentioned it in my 1980 console article.

While there are probably any number of marginal circuit designs out there, in my experience the sound character attributed to consoles is more about EQ voicing and ergonomic factors like control laws, than overt deviation from linear response.
test last night between processed and unprocessed reveled the most bizarre  of things. Yes there was cancellation in the LF and HF but mids and upper mids were more pronounced not cancelling at all.

When selling something that supposedly makes the sound better, first it must make it different. I would expect even marginal console designs to be more linear for midrange frequencies and to deviate more at HF and perhaps LF if using transformers.

Good luck...

JR
 
Well here is what I am finding.  I  can not really notice a difference with or without the plug in inserted which BTW is said to emulate some of the most desired consoles on the market including the SSL 4K, Old Neve 80 series, etc.  This is what started me on the audio tests of signal with and without vcombined out of phase to cancel things. I have had many a person insist that I must be joking that I couldn't hear a difference or "insert any other negative comment here". So in the interest of trying to really see what it does or does not do I can only tell what I have heard and not heard.  More over I find that there are any number of pages explaining the various types of summing done inside a console but none that I have found that really offer audio examples.  I guess audio would be a moot point as he end result is supposed to be the straight wire with gain in a perfect world.
 
pucho812 said:
I guess audio would be a moot point as he end result is supposed to be the straight wire with gain in a perfect world.

Ding ding ding... we have a winner... Most console designers worth a sh__ did not design in distortion or coloration on purpose despite the nonsense on some recording forums.  Any deviations from linear were inadvertent limits of the technology they had available.

I have been following the OTB mixing affectation since in my judgement digital summing can actually be more accurate (in theory). I really wish there was some actual benefit to analog mixing since I have ideas that I never fully developed when I stuck a fork in big analog console design a few decades ago.  So far I have not found any there there... at least regarding audio path performance.

JR

PS I am not opposed to using coloration as an effect in the studio, just opposed to pretending that is something it isn't.
 
etheory said:
However, none of these digital operations can, in a reasonable amount of processing time, thus far, emulate the infinite-series complexity of true analog distortions.
So all the waves, Slate VCC, etc. are all using the same algorithms (no matter what they say on their marketing speak), just tweaked in slightly different ways (which is where the real point of difference is; who can do the better job of making a result as artistically pleasing as scientifically/mathematically pleasing), but, fundamentally they are all the same, all an approximation, and all not really doing what is going on in an analog system at all.

The closest thing I've seen is the Diagonal Volterra Kernels in the Nebula plugin by Acustica Audio, HOWEVER, these are still heavily order-limited, and hence I believe are still only able to create distortions up to about order 7.

Except .... None of the analog systems you like have infinite bandwidth. Everything is rolled off. Your amplifier stages are band limited to prevent oscillation/instability. HF you can't hear also eats up your headroom, so why not get rid of it? And even without adding roll-off in the amplifier that amplifier still has finite gain-bandwidth.

So maybe ten harmonics is sufficient to emulate some distortions, and you need that many only because you have may have fundamentals down in the subwoofer range.

Infinite series doesn't exist in real analog hardware so attempting to emulate it in software doesn't make much sense (to me).

-a
 
Andy Peters said:
Infinite series doesn't exist in real analog hardware so attempting to emulate it in software doesn't make much sense (to me).

touché, quite true.
Yes, I guess that a chosen number of harmonics will be sufficient to emulate the original device, and that number could be calculated exactly by knowing the original system bandwidth.

That being the case, I will change my answer and say that it should be possible therefore to exactly emulate everything except the contribution of noise, by calculating the number of required harmonics, and implementing a system that can reproduce a distortion of the specified order.

The issue currently is that we aren't quite up to the point where a computer can calculate this in real-time, but you've definitely got me thinking and now I need to do some calculations to figure out how hard this would actually be.

Thanks for the prod to get me to think more.

The biggest issue I can see here is that current PDE solvers that are practical go up to order 4 for Runge–Kutta. Theoretically you'd also need to solve your PDE's with an equal order solver in order to make the most of your higher order distortions. That's where the first issue I can see starts.
 
Ran more tests last night. Pink noise and the FFT.... low end boost anyone? Frequencies above around 125hZ were able to null out but below that there was a shelf happening. It was more so on a  specific setting that was supposed to be a certain console from across the pond. This is different from before as before with program examples of pre and post we had different results.  But then again those original tests are suspect to being tainted. But anyway we confirmed the second test with program material as well.  Now in order to get any sort of noticeable changes  in audio we had to have a hot input signal going into the plug in.  This allows us to achieve coloration, distortion and the eq bump we were seeing with a high level of pink noise.  at unity level's of in and out the effect was negligible.  JR as always thanks for the insight. I'm putting this under faith based audio category. If you spend the money you hear and improvement even if you can not measure it just because you spend the money
 
Pucho,

If you have Smaart ... maybe try looking at the transfer functions.

Comparing Phase and Correlation.  This should tell you what frequencies are changing and what amount of phase shift is happening all over the audible band.  I would do this but I never found any of the console emulator plugs to be worth it.  To me they seemed to be doing very little or they became a distortion box without controls.

Michael
 
I got an AP could easily do that action with one. I am of the standpoint they are useless as far as plug in's go and fall into the category of a person paid for it so it must make it all better in their eyes even though it  does do much of anything.
 

Latest posts

Back
Top