Welcome to Solid State Guitar Amp Forum | DIY Guitar Amplifiers. Please login or sign up.

March 29, 2024, 08:15:08 AM

Login with username, password and session length

Recent Posts

 

Clipping theory - a nice find this week

Started by QReuCk, May 31, 2013, 03:33:59 AM

Previous topic - Next topic

QReuCk

Hi guys!

I stumbled across this article earlier this week:
https://ccrma.stanford.edu/~dtyeh/papers/yeh07_dafx_distortion.pdf
So I though that was somewhat enlightening and I would share the link here.
It is more of a numeric modelisation project, but the approch they used to understand the behaviour of real life analog devices is interesting, me think.

For instance, look at the sine wave signal response they obtain for the Boss DS1 and for the Tube Screamer.
Typically, pieces of information you find on the interweb tell you that the diode arangement of the DS1 is hard clipping, which is often represented with a perfectly truncated sine wave where it hits the voltage limitation of the diodes. They measured the real device through oscilloscope + fourier analysis and what did they found? Both the OP amp and the diodes are distorting the signal, and both these devices do have a different real life behaviour than the typical representation. The resulting wave form is nowhere near what people usually asume for such devices, and the tone setting parts do not explain such behaviour if you model them as pure linear filters.

I don't exactly know if it really fits in there as the electrical behaviour of each component and arrangement is pretty blured in this "block box" approach. But as it deals with stompboxes, I might be in the correct forum to gather your thoughts about "real life" devices vs "how we assume it should work".

Thoughts anyone?

Roly

Thanks for that.

Yep, it's basically about how to create software that imitates or models real world circuits, and it's an interesting insight into how digital models can be created, but like an awful lot of papers written for conferences it's a bit "look at me - how smart am I?" and a bit more about ego boosting than actually illuminating the topic - in fact they often require considerable decoding to find that they aren't saying anything particularly extraordinary.

In this case the take away message for pedal builders or cloners is that the response of the Ibanez Tube Screamer (and the like) is dominated by the clamp diodes in the Negative FeedBack loop of the op-amp (because they prevent the op-amp itself from ever going into clipping).  The response of the Boss DS1 (and the like) is more complex because nothing prevents the op-amp itself from going into clipping (and thus suffering recovery delay), before the following clipping diodes across the signal path.

In short, a clone of the DS1 will be sensitive to the actual op-amp used, while a clone of the TS won't be.

Its main value I think is as an inoculation against the crud sometimes dished up by cork sniffers about the different "sounds" or "tonality" of different op-amps - in some cases this can matter, in others it can't.

More generally this paper comes from an area that normally doesn't get a lot of attention - what happens to circuits in overdrive situations.  Most audio papers concern themselves with what happens in the linear region between clipping, while guitarists generally are more concerned with what happens during clipping.
If you say theory and practice don't agree you haven't applied enough theory.

J M Fahey

Very interesting, thanks for posting.
And I bet the scientists were guitar players, in their "other" life outside the wage earning job, so they thought "today we'll write about something *we* love"
Cool !!!!  :dbtu:

Didn't read it, just superficially scanned it.
Like any Friday, I'm surrounded by anxious Musicians who want to play tonight and tomorrow, but on Sunday or Monday I'll read it many times, as it deserves.

Thanks again.

phatt

I quote from the link;

"The arithmetic introduced by the variable substitution"

"This is not a memoryless nonlinearity"

:duh xP :loco :-X  Headscratch????

I get really worried when they start talking all dat tecky stuff that actually explains sweet  nothing.

Holy crap man way to deep for me and I've only got one life time and I do want to play guitar more than learn funny words to impress people who likely never even played chop sticks on an old piano.

Don't mind Me but I'll just stick to some basic observations,, they have worked well so far.

I lack the ability to explain some observations but I get the distinct impression that a lot of modern gear both didget and analog gear may actually introduce some ugly things simply because they don't have things like transformers.

My hunch is that a lot of SS stuff passes way to much bandwidth and maybe some of the issue causing undue harshness.

I saw a screen shot somewhere of Hi freq fizz on the Plate signal but post Output transformer was gone.
I can only assume that fizz was beyond the bandwidth of the OT?
Logic suggests that high freq fizz would pass straight through with most SS designs?

My dear old Quadraverb always tormented me it was ok but it lacked something.
Even with all that cab sim and many eq options it was never liked,,,,,, Until i put some serious hi cut at 3khZ across the output.  WOW THEN it was a completely different beast.

That's about as deep as I get with teck.
Phil.

Roly

QuoteObsfication - the hiding of intended meaning in communication, making communication confusing, willfully ambiguous, and harder to interpret.

Which was what I was dancing around above.  While I read math I have to confess that I tend to glaze over when I see it used like this.

"The arithmetic introduced by the variable substitution"

The process of simulation is to mathematically manipulate the values of the samples of the sound to produce the desired effect.  In some cases this results in math that gets a bit complex to do fast with a limited processor, e.g. floating point "123.456" against integer arithmetic "123456".

In digital signal processing some operations such as "multiply A * B", or "multiply A * B and add to C" are fairly common.  There is an old saying that goes something like "A ha'peth of hardware is worth a shitload of code" so you can get Digital Signal Processing chips that contain specific instructions such as these that use internal hardware to pull the rabbit, C, out of a hat rather than grinding around "turnips times bananas equals pineapples" many times to get the same result.

But if you're cheap you take a step back and try and construct the problem so you avoid the need for such calculations in the first place, the need for "The arithmetic introduced by the variable substitution", and you can get away with using a cheaper/slower processor.

"This is not a memoryless nonlinearity"

This hinges on "memoryless".  In DSPing a "memoryless" operation is one which makes no reference to a previous state, for example simply scaling samples as a volume control or straightforward diode clipping.

An operation with memory uses a value stored from a previous operation as part of this operation, and in turn stores a value for use in the next iteration.  Sundry effects that rely on delay such as reverb, echo, chorus, &c are in this class, as are tone controls and filters.

Here they are saying that this distortion has memory of a previous state, and the most obvious thing that comes to mind in this context is op-amp latch-up or saturation recovery time, where what is happening now depends on what just happened previously.

There was a rather good article in old Wireless World a few (many?) years back that (actually) explained "z-plane" operations/transforms in exactly the way this paper doesn't. 

Putting it as simply as I can, this is looking at functions such as poles and zeros (e.g. CR networks, and more complex filters) in terms of finite time steps or samples in the z-plane (z = this sample, z-1 = last sample) rather than the more usual analogue s-plane transforms.

e.g.

new-z-value = this-z-value somefunction previous-z-value

This idea of system memory can also be understood in analogue systems such as CR networks as the voltage stored on the C at any instant as a memory of a previous instant.

Some people seem to get a bit carried away with the fact that you can make a mathematical model of reality, particularly when you can make it run on a modern processor fast enough to simulate real time, and potentially replace the analogue system it is modelling.  While it's certainly a neat trick (and there are some things you can do easily in a processor you can't in analogue) I think it tends to confuse the map with the territory.

If you say theory and practice don't agree you haven't applied enough theory.

QReuCk

You know what's funny about it? I too tend to scan forward when I see equations. I noticed the diagrams and thought these where interesting by their own merit because the curves were not what I expected them to be (more precisely, they are not what the typical guitar forum -not this one but you know what I mean - consensus assume them to be).

Then reading this discussion highlighting the non memoryless non-linearity, I finally grasped what you are saying (or what you didn't intend to say but my limitted english made me believe you did :duh ) with your Z-plane and that perfectly makes sense. We always (knowingly so or not) use some sort of fourrier analysis in our understanding of filters, but capacitors do pass intensity based on voltage drop, so they can act a bit different treating complex guitar signals (especially distorted ones containing lots of harmonics) than as pure gain reducer of each individual band.
I yet again cannot make use of that knowledge in practical circuits, but I find it interesting anyway.