KVR Audio

pdxindy · Post by **pdxindy** » Sun Jan 25, 2009 5:54 am

zvenx wrote:Each method has its advantages and disadvantages...... I guess to each his own...to me that stance makes you miss out on some wonderful stuff out there that require dongles...I am sure most of Angus customers however appreciate their method of protection.
rsp

Sure... each person can make the choices they want... Fortunately, there is more than enough top quality stuff that is dongle free that I can never want!

Also, it happens that some of my favorite smaller companies who I am most pleased to support express thoughts like Angus did... makes me happy!

zvenx · Post by **zvenx** » Sun Jan 25, 2009 5:54 am

Angus_FX · Post by **Angus_FX** » Sun Jan 25, 2009 9:37 am

To be clear - I don't personally mind using stuff with dongles when I absolutely have to - which I pretty much do. However, it's a) not something I ever want to inflict on our customers, and b) the thought of companies like PACE profiting from the whole sorry situation makes me quite uncomfortable.

BTW - for those who asked earlier on in the thread about Animator audio demos - I will get some put together, still catching up from NAMM right now but will post in this thread as soon as we have some, shouldn't be too long.

pdxindy · Post by **pdxindy** » Sun Jan 25, 2009 9:57 am

Angus_FX wrote: BTW - for those who asked earlier on in the thread about Animator audio demos - I will get some put together, still catching up from NAMM right now but will post in this thread as soon as we have some, shouldn't be too long.

Thanks, would love to hear Animator in action... also, got a screenshot of it that you are ready to post?

hakey · Post by **hakey** » Sun Jan 25, 2009 10:09 am

Hey Andy, a general dsp question from a layman - How do you attempt to model non-linearities in a system? As I (poorly) understand it, non-linear equations aren't solveable, they're recursive so, do you 'run' non-linear equations, but limit them before they become unmanageable, or do you use some kind of pseudo non-linear algo that gives something like non-linear behaviour.

Also, you say:

The amount of cpu required to really be near indistinguishable from analog is beyond current computers

How much quicker do they need to be and how would that extra speed be used? If, as you say, the current models aren't yet indistinguishable from real analog, in terms of their audible qualities, where do they fall short?

Kriminal · Post by **Kriminal** » Sun Jan 25, 2009 10:53 am

Angus_FX wrote:@ Krim - if you're in London, drop me a line - if I'm about in the office, you can pop by and see us.

will do, be good to see you

andy-cytomic · Post by **andy-cytomic** » Sun Jan 25, 2009 11:55 am

hakey wrote:Hey Andy, a general dsp question from a layman - How do you attempt to model non-linearities in a system? As I (poorly) understand it, non-linear equations aren't solveable, they're recursive so, do you 'run' non-linear equations, but limit them before they become unmanageable, or do you use some kind of pseudo non-linear algo that gives something like non-linear behaviour.

Also, you say:
The amount of cpu required to really be near indistinguishable from analog is beyond current computers
How much quicker do they need to be and how would that extra speed be used? If, as you say, the current models aren't yet indistinguishable from real analog, in terms of their audible qualities, where do they fall short?

I use standard circuit modeling methods that circuit simulation software like qucs and spice use. Here is a technical run down on the engine that powers these packages which I've found invaluable for my models: http://qucs.sourceforge.net/tech/technical.html

The main idea is to solve the system using dc analysis. You can turn things that change over time (eg capacitors, inductors) into things that can be numerically integrated and solved for by using dc analysis. A run down of what that involves for a simple diode circuit is here: http://qucs.sourceforge.net/tech/node16.html

So as you can see for a simple circuit with one resistor and one diode you are already having to loop around 8 times and evaluate an e^x and a division every time, and that is for 1 sample. Now oversample this 8 times to reduce aliasing and you end up with 64 evaluations of e^x and 64 division and thousands of assignments and multiplies and adds and subtracts per sample for the most trivial of non-linear circuits. Luckily there are many different ways, which I won't go into here, to optimize and simplify circuits. In this particular case, ignoring self noise and very high frequency capacitance, and if you have two op amps, one before and one after the resistor and diode pair, then you can model this with a table lookup.

Now take one of the filters with the fewest number of discrete components, the moog ladder. For one filter it has around 16 transistors, around 30 resistors, and around 12 capacitors. As you can see here: http://qucs.sourceforge.net/tech/node70 ... 0000000000 each transistor needs four evaluations of e^x so that's 16 * 4 * 8 * 8 = 16 * 4 * 8 * 8 = 4096 evaluations of e^x per sample. As soon as you have more components joined together this leads to an exponentially larger combinatorial time to solve as well. Then you need to add self noise to each component eg http://qucs.sourceforge.net/tech/node70 ... 0000000000 . This is for one filter of one voice of a synth! I haven't done the sums to work out exactly how much faster computers will need to be, but I put it conservatively at several thousand times at least. Even then what you are left with is a simplified model of the actual physics going on.

So for the dsp side of the work my job is bascally to understand all this circuit stuff, what the most important bits of the circuit are to the sound, and how to work out algorithms to run those parts of the circuit in realtime. Knowing all the maths and physics and making informed decisions on what sounds musical and what gives things their character are very closely linked, either in isolation isn't enough. This applies even more so to the brilliant people that came up with these circuits in the first place - hats off to them.

Andrew Simper

Angus_FX · Post by **Angus_FX** » Sun Jan 25, 2009 1:45 pm

@ pdxindy -

It's still work in progress to some extent, but it should give you a pretty good idea where we're going with it. The three single-function modes (Note Sequencer, Mod Sequencer and Arpeggiator) present a very much simplified user interface geared to the specific task, and you can mix and match across the four engines (i.e. two arps, a mod sequencer and a note sequencer) should you wish to do so.

@ hakey - following on from Andy's explanation, a simple example in the real world as to the limitations of digital.. any time there is nonlinearity in an audio signal path, it adds harmonics. Which, in the digital domain, causes aliasing any time the added harmonics go over Nyquist, and is why an "anti aliased oscillator" on its own is not good enough - the moment you waveshape it, ringmod it, or put it through a filter with any balls at all -- hello aliasing, goodbye analogue. Even the most trivial analogue synth circuit has multiple nonlinearities in series. So you either have to run the whole signal path highly oversampled (which the D-CAM synths in fact automatically do in offline rendering mode), or bandlimit around each individual nonlinear stage (which is insanely CPU expensive for a polyphonic synth with enough stages to sound good).

Sometimes modeling the weird things that the analogue circuit designers did to make their hardware synths cheaper, and which gives them character, actually makes things more expensive in digital. For instance, on the paraphonic string machines (on which Amber is based), they sometimes used a biased diode circuit per-key to get a crude envelope amplifier, as this was much cheaper at the time than having 61 VCA's. It's actually really quite poor at the task - as the envelope decays away, the shape of the waveform changes radically - but it's part of the character of the sound. If you listen closely to an RS-505 or an Electrapiano, you'll hear this weird PWM-y, phase-y sound during the decay stage which is actually the waveform being shaped by this circuit.

Unfortunately it's so expensive to do this properly with a digital simulation, we had to take something of a short cut there. Simulating the equivalent circuit based on good quality VCAs, on the other hand, is cheap to do in digital.. the "perfect" VCA, so expensive in analogue, amounts to little more than a multiply in digital, whereas the circuit they used in those synths, which costs pennies per key in analogue, would require more CPU per voice than we can afford in digital. By and large, the cleaner the circuit, the cheaper it is to reproduce in digital (for the most part -- the cost doesn't go to zero, even 'clean' stuff can oftentimes benefit from the extra mathematical headroom afforded by e.g. oversampling), but it's the dirt and nonlinearity which makes analogue sound interesting in the first place, especially if we are talking synths. Some of the FM patches I've cooked up on Cypher really illustrate this - although the FM is essentially just 2-op, because it's modeled VCOs rather than naive digital sinewaves a la DX7 doing the FM'ing, there is just so much more richness in the sound.

Another less concrete example from the real world can be found in the sheer musicality to be found in (for example) the SH-101, all the more so given the relatively tiny number of controls on that synth - yet just as even the very best digital instruments struggle to approach the musicality of analogue, the very best analogue electronic instruments struggle to approach that of a well-made physical instrument. An awful lot of VSTi synths, like a lot of too-cheap guitars and blinged-up workstation keyboards, seem to have little or no thought put in to musicality. A good musical instrument, whatever it might be, should be so playable that it almost has music waiting to pour out of it. The D-CAM: Synth Squad instruments are, to my mind, a significant step in the right direction - Strobe in particular delivers "more music per knob" than pretty much any software synth I've played with - but there is no doubt further room for improvement there.

djsubject · Post by **djsubject** » Sun Jan 25, 2009 3:13 pm

Angus_FX wrote:

i think i just pooped

now that looks like fun to program!!

Subz

SWAN808 · Post by **SWAN808** » Sun Jan 25, 2009 4:00 pm

Thanks Andy and Angus for the enlightening info - its great to have such expert developers communicating with the userbase - and I love hearing about how these things work...

It sounds like there has been some serious commitment to modeling and studying synthesis - and its great how you are able to discuss that openly here - so we as users can get an idea about how much you've put into it.

Very much looking forward to tasting that pudding!

hakey · Post by **hakey** » Sun Jan 25, 2009 4:00 pm

Andy and Angus thanks for those interesting explanations.

Angus_FX wrote: The D-CAM: Synth Squad instruments are, to my mind, a significant step in the right direction - Strobe in particular delivers "more music per knob" than pretty much any software synth I've played with

Well I'm looking forward to testing the truth of this myself.

pdxindy · Post by **pdxindy** » Sun Jan 25, 2009 8:41 pm

Angus_FX wrote:@ pdxindy -

It's still work in progress to some extent, but it should give you a pretty good idea where we're going with it. The three single-function modes (Note Sequencer, Mod Sequencer and Arpeggiator) present a very much simplified user interface geared to the specific task, and you can mix and match across the four engines (i.e. two arps, a mod sequencer and a note sequencer) should you wish to do so.

Thanks for the screenshot... wow, quite a set up... so each of the engines can run 1 of the 3 modes... nice, it has some version of the graphs from GURU...

Can Animator switch patterns on the fly?

:::pulls out a gun::: Okay, I want it now!! hehehe

teknobryan · Post by **teknobryan** » Sun Jan 25, 2009 10:17 pm

so what ever happened to that magic compressor that was supposed to come out with the D-squad?

djanthonyw · Post by **djanthonyw** » Sun Jan 25, 2009 10:18 pm

You mean this?

http://www.kvraudio.com/forum/viewtopic ... ic&start=0

MitchK1989 · Post by **MitchK1989** » Mon Jan 26, 2009 1:53 am

So do these synths have audio input?

Does that mean that you can use fuzor as an effect? If so, is it possible to do it without a synth in the chain, or at least with the filter bypassed on the synth so that we can just use the D-Cam and digital effects?

FXpansion D-CAM: Synth Squad for sale now!