Guitar Lessons by Chip McDonald - chip@chipmcdonald.com: The Problem With DAW Plugins Not Officially Discovered: Scurrilous Experiments and Non-scientific Conclusions - PART TWO

Wednesday, January 2, 2019

The Problem With DAW Plugins Not Officially Discovered: Scurrilous Experiments and Non-scientific Conclusions - PART TWO

(note to the glitterati that has contacted me, that either chooses to be argumentatively rambunctious or reflexively pedantic in a ego-needful way: I don't really care, as written in Part One this is errant, off the cuff extemporaneous "speculation".  As such I'm not willing to debate about it, nor do I care if you want to make a mental ego-measuring contest out of it: I don't need to do that, why do you...?)

.. part two, where Chip further digs an unfounded hole.....



GRIPE #2

The temporal number crunching.  This is where Ye Old Infinite Resolution steps in, but wait! I'm not talking about it in the "traditional sense", give me a moment...

 In the analog domain, your distortion pedal is instantaneously changing your guitar sound.

Every moment you play, yields

1) a unique level
2) a unique pitch
3) a unique harmonic content


 Every moment.  With zero latency, with perfect parallelism.  From a processing standpoint, in software you've got to address those 3 things based on an instantaneous sampling reduced to a single number representing level.  To get a result from your function, you have to determine a modifier for those 3 things.

 This should be perfectly digital model-friendly, it would seem.  The problem I think, is that you have to do math on the single sample one at a time serially, or you have to do it component-wise and then add it together.  You're applying basic math to the number to represent the change in level, the change in pitch, and the harmonic content.  It's really just one number across a set of numbers  - a grouping of 1,024, or some such.  A processing "clump".

That "clump" then leads to another clump, etc..  The math applied to each clump will be the same.
The buffer is NOT instantaneous, however.  So while in theory the sample rate is "fast enough" to represent any audio signal, the software is trying to modify that signal faster than reality.  It's not that the analog world has Infinite Resolution, it's that it has Infinite Parallel Processing Power.  It's not doing anything in a buffered state.  It's not doing anything serially, or in modules paralleled.   No clumping.  One continuum.  The variability changes with infinite granularity; all aspects are not fitted to a curve and composited serially. 

 Comb filtering is (effectively) errors in sound that occur at mathematically regular intervals across the spectrum.  It's my belief that as a byproduct of the math in software happening temporally, clump by buffered clump - but with metered regularity of delimited by the buffer size - that across a longer time scale (a second, 2 seconds), there is a "temporal comb filtering" happening.

 "Temporal Comb Filtering": yes, I made that up.  Normally one describes comb filtering as an instantaneous phenomenon.  "Here is the sample of this moment, and we can see peaks at 100 hz, 200 hz, 400 hz, etc.".  What I am describing is this happening at some ratio across time.

 The buffer z is processed, then z+1, then z+2, etc..  But, because the same math is being applied to every buffer, there could be artifacts/errors introduced that creates a harmonic series only seen in multiples of the buffers.  On a waterfall plot it would be buried among the resulting signal.  A number being rounded up or down, 1,024 times modified by whatever other functions,  creating an artifice that is not visible in a graph, or even a waterfall plot because - how do you know it's an artifact when it's the result of math on a test signal that's changing?

 The rest signal is *variable*.  Guitars are not perfect signal generators.  The math applied to a perfect sine wave would be confusing, because you are making a function that is intentionally truncating values to yield distortion.   You have no way of knowing if your mathematical system across time is making a harmonic series alteration that is not linear to a Real World Analog Amp.

 Even if you have a sweep, or a set complex wave, you wouldn't know because you can never measure it against an analog equivalent perfectly.  Comb filtered sound can measure frequency wise as being "close" - but again I claim the human mind can discern the difference across a large sample set.

 Your brain realizes "there is a commonly reoccurring series here" that doesn't happen in the analog world. A non-humanly testable phenomena, and a non-scientifically testable phenomena.

 The result being, for most distorted guitar sounds I hear an amount of comb filtering I don't like in the mids/highs.  When that doesn't change - it sounds "digital" to me.

 I first had an inkling of this thought when the first Line 6 gear came out.  When I first heard it I was super impressed - it does sound like, in time slices, the real thing.  But then, if you hold a chord and spin the dial while the presets go by, you'll notice a harmonic coloration to *all* of the presets.

 That is software artifacts I think, and it's evidenced by comb filtering in the same manner on everything.  All digital sims have this I realized, when I tried the Fender Cyber Twin for the first time: spin the knob, and there it is, comb filtering.  Plug into the Vox modelling amp next to it, spin the knob - all the presets have that comb filtered sound, maybe at a different frequency/spread.

 Once you hear it, it's always there.  You can fool yourself into thinking you don't notice it, but it's there.  Every electrical system is going to have comb filtering artifacts, particularly speakers, but it's not a fixed thing between devices.  And it's state-variable; more or less evident depending on the input signal.

 As an example of this, I'll point to a video by John Segeborn that is tremendously great and educational.  In this video he plays the same thing back through different models of a Celestion Greenback speaker.  You'll hear comb filtering on each as a "shhhhh" harmonic coloration, but it will be different on model.  Which is fine - that's what speakers do.  The problem is when your software is adding another coloration on top of that one, or homogenizing it:



  In each example you can hear a spike in treble.  BUT, you're not just hearing a spectral peak, it also has comb filtering: a vaguely "smeary" sound, that changes in dominance depending on the signal.  My belief is that humans are super sensitive to this, and THIS is what software is messing up in sims.  I think it is too linear in general to signal level in sims.

 So, at some volumes it might be spot on.  At other levels it's too loud, or maybe buried by the upper harmonics.  This interaction is flawed in digital recreations I think.


 I think.  I do not feel like trying to provide proof or documentation.  I've been (unfortunately) doing all sorts of tedious comparisons and tests for years at this point that has led me to these assertions.  I think there is a problem here in the comb filtering and harmonic decay linearity.  I could be wrong.  Harmonic decay errors, and comb-filtering problems.

$.10.

POST SCRIPT

 Here's a yet another free idea I wish I had the resources in which to patent, but I don't:
Without a doubt, at some time within 3 years a company will come out with a post-processing VST plugin that will use A.I./confrontational machine laerning to conform a track output to mimic anything.


















1 comment:

  1. Besides, if you are the one in search any such guitar then it is suggested to make an effective purchase from the best online guitar store. The guitar offered for sale at the online store will surely fit your taste and preference. The best thing about making online is its easiness.

    ReplyDelete