Login / Register  0 items | $0.00 NewWhat is KVR? Submit News Advertise

Analog modelling

mystran
KVRAF
 
4557 posts since 11 Feb, 2006, from Helsinki, Finland

Postby mystran; Thu Aug 24, 2017 1:52 pm Re: Analog modelling

aciddose wrote:There is a small effect on decay time by the accent pulse changing the output current. The circuit uses a darlington but it is far worse than a FET buffer and the effect on timing gets significant especially at the maximum decay time setting.


I'm not sure if I understand what you mean by this, since accent always forces the minimum decay time, by design.
Image <- plugins | forum
User avatar
aciddose
KVRAF
 
11316 posts since 7 Dec, 2004, from Vancouver, Canada

Postby aciddose; Thu Aug 24, 2017 2:02 pm Re: Analog modelling

Right... then the effect in the tb-303 itself is far smaller than I assumed. I'd need to put in a lot of effort to model the circuit and get the exact numbers to re-assess whether the darlington buffer + accent current has any effect that is significant.

It seems the resistance is normally approximately (guessing) 60k to 20k depending upon the settings of decay and accent. It is possible the peak level of the envelope might change but my intuition tells me it's likely 1% or less.

The "minimum" decay time does definitely change to a shorter time but again the number is likely only slightly shorter than the unaccented minimum decay. It might be as much as 5% or so but this probably isn't significant.

The discussion though should point out quite clearly how many effects are going on and how complex the process of producing an accurate model and understanding all the behaviors of all the circuits involved can be.

Most people who have claimed in the past to be working on "analog modeled ..." are full of it and not nearly investing the necessary effort (huge) to make these small improvements.
http://xhip.net/synth/ Xhip Synthesizer v8 released 21 May 2017
mystran
KVRAF
 
4557 posts since 11 Feb, 2006, from Helsinki, Finland

Postby mystran; Thu Aug 24, 2017 3:39 pm Re: Analog modelling

Urs wrote:Above that level you'll find there are numerous effects already that you do not wish to add to your model, such as humm, buzz, noise, and maybe some weird signal picked up from your smartphone.


I wanted to jump back on the topic a bit, because I wanted to comment on this.

While modelling EM pickups from cellphones might not be a good idea, some of this stuff (noise, humm) does actually affect the character of analog hardware quite significantly and I think the quest for "like analog, but without any of the noise" is somewhat misguided actually, because I firmly believe that the more you avoid the noise, the more you end up with a distinctly "sterile" digital sound.

More importantly, I don't think adding noise to the signal path or traditional modulation paths really does the trick either. As an obvious example, something like oscillator reset-trigger variation affects the resulting sounds distinctly different compared to simple frequency CV noise and we know these thing can be fairly sensitive given the "2 oscillators sharing a voltage rail can soft-sync" example given earlier.

I'm not totally convinced that one needs to simulate all of it on circuit level (let alone PCB level), but I do believe one should identify some of the secondary effects (eg. the stuff that aciddose just talked about) and deliberately introduce some of these effects at least on "macro" level in the models where it contributes to the sound.

In other words, I don't believe all noise is bad. Rather I think the desire to avoid noise is one of the things holding some of the contemporary models back the most.

YMMV.
Image <- plugins | forum
Kraku
KVRian
 
1377 posts since 13 Oct, 2003, from Prague, Czech Republic

Postby Kraku; Fri Aug 25, 2017 1:42 pm Re: Analog modelling

mystran wrote:As an obvious example, something like oscillator reset-trigger variation affects the resulting sounds distinctly different...


This is actually something I've been interested in lately. What methods are people usually using to achieve this effect? If you add a bit of white noise to each oscillator cycle length, the results aren't that good. I assume a bit of circuit modelling is required to get non-white noise that retriggers the oscillator cycles in a more pleasing way?
User avatar
aciddose
KVRAF
 
11316 posts since 7 Dec, 2004, from Vancouver, Canada

Postby aciddose; Fri Aug 25, 2017 3:59 pm Re: Analog modelling

Generally the waveform is a combination of "pink" (integrated is good enough) noise (thermal + flicker) from the PSU regulators in addition to ar(abs(mains)), where mains is approximately integrate(clamp(sin(x))).

(ar() = lossy integration with different attack/release coefficients.)

It turns out you can quite easily approximate such waveforms with trivial functions and it is important to keep in mind that the approximate spectra is the important part.

The balance between hum and noise is especially important: you need very little noise.

An example of the sort of waveform you're looking for:
Image
https://tangentsoft.net/audio/opamp-ps.html

Notice that the waveform is far more complex than a simple (abs(sin(x)) + pinknoise). This is due to the fact the waveform has been integrated and differentiated at different points and that the input was not a pure sine.

The resulting interference between oscillator frequency and mains hum is what generates the interesting "dirty" effects in the high frequency range (>1k) and all the extra side-bands you'd normally see in an analog oscillator and filter.

Integrating (lossy) these signals is essential. The usual mistake is to feed uniform white noise: this is nowhere near the "pinkish" Gaussian noise mixed with mains hum where the mains signal is impure to start with.

If you modulate the phase reset point (1.0 +/- 1e-2 or less) using the proper signal you'll find the result is ideal and very different from the effect you'd get by simply mixing noise into the phase value.

As for the strength of the modulation note that the usual reset point is 5v, with ~50mv ripple this would be 1e-2.

You should not mix noise into the phase value because this has other effects (frequency mod by noise) which do not occur in an analog oscillator. All you'll get that way is very "grainy" and "noisy" waveforms that sound like you've simply mixed noise with them directly.
http://xhip.net/synth/ Xhip Synthesizer v8 released 21 May 2017
User avatar
Urs
u-he
 
21044 posts since 7 Aug, 2002, from Berlin

Postby Urs; Tue Aug 29, 2017 4:51 am Re: Analog modelling

mystran wrote:In other words, I don't believe all noise is bad.

Neither do I, and while I don't mention it particularly in that post, when I wrote "It's up to you which of those effects you find desirable and which you don't" I think I'd also have implied that bit of noise and seemingly erratic behaviour *can* be a good thing. We're the company which refuses to "fix" our tape emulation plug-in with a completely noise free operation mode :clown:

On a related note, I'm thinking about buying a new power system for my Eurorack modulars because a recent recording had some "electronic whine" on it, from whatever was bleeding through the rails. That is the kind of crap which I meant to suggest to avoid in an analogue model :phones:
User avatar
aciddose
KVRAF
 
11316 posts since 7 Dec, 2004, from Vancouver, Canada

Postby aciddose; Tue Aug 29, 2017 3:52 pm Re: Analog modelling

A good quality article about one of the primary sources of noise:
http://www.eetimes.com/document.asp?doc_id=1278251

People seem to obsess over opamps and other active devices; justifiably so to a degree due to the impact these devices can potentially have.

Many amateurs however seem unaware that the primary sources of noise tend to be passive devices! Since white "thermal" noise depends upon total resistance it can be reduced by increasing total current. This has the opposite effect however on both active and passive devices regarding "flicker" ("pink") noise.

Due to that effect combined with the fact the main part of the low frequency spectrum (DC) is typically filtered away, every circuit is designed with a number of trade-offs balancing between low-frequency preservation, thermal noise-floor and pink noise in relation to efficiency (AKA: cost) and power supply requirements including both forms of supply noise (thermal and flicker) and ripple.

Thermal noise
Flicker noise
Shot noise

White noise
Pink (1/f) noise (-3 dB, "pink")
Red noise (-6 dB, "red")

Uniform distribution
Gaussian distribution

Uniform noise (no wiki article: a typical pseudo-RNG has a uniform distribution output.)
Gaussian noise

Fun facts:
If you integrate (lossy or not) to create pink or red noise from a uniform source, this act creates Gaussian noise. With lossy integration it's of equal peak amplitude but such a peak is impossible to reach outside an infinite period of time. The average level will be significantly reduced and will depend upon the input distribution vs. integration coefficient. The function for this is unknown to me. (I know it involves the discrete transfer function + various statistical functions.)

You can get increasingly Gaussian distributions by summing any non-Gaussian distributed values. This is why the Gaussian distribution is called "normal". Since most real world values are integrated from multiple sources they tend to have Gaussian distributions!

The explanation for that is simple. Any two values may be independent regarding probability but to generate a string of identical values (or values with the same sign relative some reference) are less likely than a single value. (Someone could link the proof for this, I'm unaware of it or forgot.) This probability becomes infinitesimal with N = infinity. So any two minimal or maximal values in a row are less likely than a single value. Any three are even less likely than two. That's where the "normal" distribution comes from and this is called the law of large numbers.

Possible proof via reducto ad absurdum: If this were untrue a coin flip wouldn't average to 1/2 but could end up arbitrarily anywhere between and including 0 or 1. What a fun universe that would be.

One possible application of this is that you can store a ring-buffer of RNG output and sum the output into a wrapping register (32-bit input, 32-bit sum.) Each time you move forward a step you subtract the value you're replacing in the ring-buffer before adding the new value.

To get the "unwrapped" output you multiply the result by the modular multiplicative inverse.

So the total cost should be only slightly more than a typical "cheapest" RNG + short ringbuffer + two adds and one mul.

Duh moment: It's also possible to just double the cost and eliminate the ring-buffer by running one of two sync'd RNGs forward by N steps in advance.


Duh moment two: this isn't actually possible except in special cases (for FIR filters?)? To store a sum of N*length bits may require a buffer of N+log2(length) bits to prevent wrapping. That said whether or not any moving average is useful depends upon the particular application and frequency range in question (can compensating filters be applied? is a comb practical? ...). As an approximation to an N times integrated (red) Gaussian noise however it may be something to look in to if only to justify the complexity of "Gaussification functions".
http://xhip.net/synth/ Xhip Synthesizer v8 released 21 May 2017
User avatar
Architeuthis
KVRAF
 
2851 posts since 27 Jan, 2006, from Phoenix, AZ

Postby Architeuthis; Wed Sep 13, 2017 1:14 pm Re: Analog modelling

We can do better than analog, modelling is not needed. Large amounts of CPU is not needed.
User avatar
Audiority
KVRian
 
575 posts since 15 Nov, 2005, from Italy

Postby Audiority; Wed Sep 13, 2017 11:13 pm Re: Analog modelling

True that, but it's also true that the same code will have a chance to sell better if you claim it's analog modelled, featuring a proper 3D GUI. #sadbuttrue

Cheers,
Luca
Progsounds | Audiority | Epic SoundLab
Need a Kontakt scripter? Contact me.
User avatar
Architeuthis
KVRAF
 
2851 posts since 27 Jan, 2006, from Phoenix, AZ

Postby Architeuthis; Thu Sep 14, 2017 3:50 am Re: Analog modelling

Audiority wrote:...it's also true that the same code will have a chance to sell better if you claim it's analog modelled...
Wait a minute, I just got done reading the "Analog-modeling or analog-emulating" thread, by the way I've been on the customer end of things for most of my life. It's only recently that I decided to create my own products, as I found things available were not to my satisfaction, and from what I gained from reading that thread is that... analog modelling can mean just about anything, as long as your intention is to describe what you've created is in fact an attempt to get "that sound", whether it be warmth, naturalness, quirky, or whatever aspects.

So, cool, yes, I can and should advertise my products as analog modelling... my products deserve that label because I've put a lot of thought into the fundamentals of analog sound, electricity, physics, feedback/chaos, and what is natural vs unnatural.

It is nice to have the option of unnatural sound, but the failure of digital synths today is it's wayyyy too easy to get unnatural sound. My goal is to design synths that sound natural first, and if you want unnatural sound, it's available, but you must intend it. E.g. an envelope generator should maintain an acoustic-like attack and decay regardless of settings, until you push the "I don't want beautiful plucky/snappy/thumpy shapes" button.

Man, why T F aren't people making better decisions with their synths? They just do whatever the next guy has done. I am first a frustrated customer creating my own tools before I call myself a developer.
User avatar
Audiority
KVRian
 
575 posts since 15 Nov, 2005, from Italy

Postby Audiority; Thu Sep 14, 2017 4:20 am Re: Analog modelling

Don't get me wrong ;) I've been on the user side for so long before start making my own plugins. I started with a DX-21, then a K2500 and then going through modular synthesizers. I love both the worlds (analog and digital) for what they offer. But as a developer I noticed the aforementioned behavior from the user base. If you place on the market two feature wise identical products, with the same code, but marketed in a different way (GUI is also a big part of the marketing imho), most of the public will go for the one that reminds the real thing... and that makes me sad, until I don't see the few exceptions around. I'm so happy when I see companies creating products that are out of this world (Zynaptic, iZotope and so on), but that market is a nice of a nice and, unless you are not licensing your own technologies to 3rd parties, there will be such a limited number of customers that eventually will jeopardize your company.

My 2 cents,
Luca
Progsounds | Audiority | Epic SoundLab
Need a Kontakt scripter? Contact me.
User avatar
aciddose
KVRAF
 
11316 posts since 7 Dec, 2004, from Vancouver, Canada

Postby aciddose; Thu Sep 14, 2017 6:12 am Re: Analog modelling

Trends are what drives success. If you were smart you would have been starting some billion dollar company to make juicers that cost $350 each where people can more quickly and efficiently juice the fruit in their hands.

The biggest mistake you could make is to try to make a "better" product; it'll be almost certainly doomed to failure.

Money was never a big motivation for me, except as a way to keep score. The real excitement is playing the game.


"The game" is manipulating people. The people who tend to be most successful in business are moderately intelligent sociopaths with excellent social skills.

Just look at Apple's success with the ipod/etc: are these "better" products? No. They were intentionally designed to offer fewer features and a simplified "trendy" and "aesthetic" interface regardless of the negative effect this has on their practical use. (Super slippery shiny phone without grips and rounded corners slips out of hands and smashes: OOPS!)

So ultimately they are "better" in that aspect: they're better at selling to customers. Currently Android phones are taking over the market although at a much lower margin: this is called "race to the bottom". Apple currently sells a crap product at an excessive mark-up because they were able to convince people it was worth more by appearing aesthetically pleasing to those people. Unfortunately for Apple, there are a lot more people in the world who care more about a product that is cheap and affordable rather than trendy and aesthetic. All it takes is to successfully market that cheaper product and grow the trend to the breaking point where Apple can't continue to compete.

Look at their most recent "solution" to this problem: add more glitter and crank up the price.

The truth is nobody wants an analog synthesizer, it's just plain trendy. Even if they had them they'd have no idea how to use them the same way they use plug-ins in their DAW.

Being able to capitalize on this is called: (ancient chinese version?) "riding on the times", think of it like surfing on a wave of retarded trends while keeping your balance.

This is why the vast majority of claimed "analog models" are in fact nowhere near what could be considered anything even remotely like any particular circuit. They are however reasonable (albeit incredibly inefficient and impractical) software synthesizers.

Just like the i-whatever, I-95?
Last edited by aciddose on Thu Sep 14, 2017 6:25 am, edited 1 time in total.
http://xhip.net/synth/ Xhip Synthesizer v8 released 21 May 2017
User avatar
aciddose
KVRAF
 
11316 posts since 7 Dec, 2004, from Vancouver, Canada

Postby aciddose; Thu Sep 14, 2017 6:15 am Re: Analog modelling

None of that however means investing honest effort to produce better software is wasted effort: for those of us who really own and regularly use analog synthesizers as well as possess an interest not in "the sound" but the technology and complete result it is very interesting just to study and understand these things we have such an interest in. Even without it having any practical application.

Do you think for example that there is a difference between Gaussian and uniform white noise? There technically is one but what are the odds that someone dumb enough to be using audio plug-ins will ever understand that? Will it actually matter?
http://xhip.net/synth/ Xhip Synthesizer v8 released 21 May 2017
User avatar
Architeuthis
KVRAF
 
2851 posts since 27 Jan, 2006, from Phoenix, AZ

Postby Architeuthis; Thu Sep 14, 2017 6:31 am Re: Analog modelling

aciddose wrote:This is why the vast majority of claimed "analog models" are in fact nowhere near what could be considered anything even remotely like any particular circuit. They are however reasonable (albeit incredibly inefficient and impractical) software synthesizers.
I have found that the simplest equations and signal routings produce the most analog/warm/natural/acoustic/interesting sounds and behaviors. That is why I say "analog modelling" could be done very cheaply. I have little knowledge on how analog synths work, but I think I know the fundamental underlying behavior, which drives all my ideas.

Audiority wrote: unless you are not licensing your own technologies to 3rd parties, there will be such a limited number of customers that eventually will jeopardize your company.
This double negative is so confusing!! lol. I think you are saying this:

"If you are licensing your own technologies, there will be such as limited number of customers"

or even better:

"There will be such as limited number of customers if you are licensing your own technologies."

:D
User avatar
Audiority
KVRian
 
575 posts since 15 Nov, 2005, from Italy

Postby Audiority; Thu Sep 14, 2017 6:51 am Re: Analog modelling

@Architeuthis: lol sorry! My english is sloppy. What I meant is (relating to the "out of this world" products): unless you don't have any other source of income, relying only on that sub-nice will jeopardize your company in the long term.
I know that's just semantics, but (to me) "analog modeling" != mere signal routing. That could be seen as "analog emulation", where you try to imitate the behavior of the machine. I used a lot of tricks in the past to get nice warm sounds, while experimenting with Reaktor over a decade ago, and it works. Warmer, richer sound out of stock oscillators and filters. Modeling, to me, is another matter and requires both the knowledge of electronics and math.

aciddose wrote:Currently Android phones are taking over the market although at a much lower margin: this is called "race to the bottom".

That's something I currently notice in our industry as well. The perceived value of digital goods is going downer and downer.
Progsounds | Audiority | Epic SoundLab
Need a Kontakt scripter? Contact me.
PreviousNext

Moderator: Moderators (Main)

Return to DSP and Plug-in Development

Who is online

Users browsing this forum: Architeuthis, CCBot (commoncrawl), Google Feedfetcher