Cytomic "The Drop" Resonant Filter

[kmonkey]

Sounds great mate. I love my UAD-Moog filter which while relatively basic (no real envelope control etc), it has the all important sound quality but it also 'feels right' when tweaked via a real knob (automap) if that makes sense. The only other software filter I've used that did that aswell, was the Moog filter on the Creamware Minimax which strangely was one of the first I used and it was released years ago. I hoped for the filter market to be pretty much sewn up by now but I've only found the UAD filter to be comparable imo.

Wow that's just my thought too. I had same experience and i reported it here. For some reason Minimax filter just sounded plain right and impressive. And that separate minimax filter unit which is awesome. I am still shocked that it is developed almost ten years ago. Only comparable thing i had with UAD moog filter and now i am impressed with Cytomic examples.

Good things coming to native

Andy from Cytomic is one of my heroes

[AgiKvr]

When The Drop is ready I'll email everyone that owns The Glue with a discount code for it, which they can use yourself of give to someone else if you want

Super!
Mega Thanx

[Nokenoku]

Haven't read the whole thread, but will it have multipoint envelopes like Fruity Love Filter?
http://flstudio.image-line.com/help/htm ... hilter.htm

[ryandfl]

I'm really looking forward to this one.

[updog]

apparently, there's some Cytomic filters included in the new Alchemy update.

"Alchemy 1.25 is now available to download from your support account. This update features enhanced analog modelled filters by modelling experts Cytomic, an additional skin by biolabs and support for remote control with Alchemy Mobile."

[penguinfromdeep]

That's quite cool, any Alchemy owners checked them out yet?

[andy-cytomic]

apparently, there's some Cytomic filters included in the new Alchemy update.

"Alchemy 1.25 is now available to download from your support account. This update features enhanced analog modelled filters by modelling experts Cytomic, an additional skin by biolabs and support for remote control with Alchemy Mobile."

I helped Camel fix the filters in Alchemy, but the design and analog modelling was done by Antti Huovilainen and not Cytomic. There were some implementation problems which meant Camel had to clip the inputs to the filters to keep them stable. My brief was to fix the filters so they could remove the input clip, but preserve the original sound as much as possible to maintain patch compatibility, and also be as efficient on cpu as possible. If I was involved in the design process from the beginning the outcome would be different to what is there now. But, the tone has been improved, and the cpu use reduced, and they are stable, so it's a big win for all Alchemy users, which includes me. Part of the reason I wanted to help out was because I wanted the filters improved for my own use

If you are interested in the best analog modeled filters I have designed thus far that are currently for sale in a product they are in Synth Squad, which I did for FXpansion. The filters in The Drop are a completely new algorithm that offers much better results, but at the cost of more cpu. These new algorithms are much closer to analog filters since they have instantaneous feedback, and handle fast modulation very smoothly.

[meloco_go]

Andy, is your instant feedback technology applicable to compressors and EQs?

[todd_r]

Me wanty drop drop

[andy-cytomic]

Andy, is your instant feedback technology applicable to compressors and EQs?

Very much so, but the instant feedback part of the algorithm isn't really my technology, the algorithms have been around for ages to do this sorta stuff in circuit simulation packages. I modify the algorithms in certain ways, but mainly this is to get them to run with as low cpu use as possible.

[andy-cytomic]

Numerical integration of non-linear differential equations existed before circuits even existed to model with these methods. Most of the algorithms were invented in the 1600's. The main people involved with coming up with these algorithms are Gottfried Leibniz (1646-1716), and Sir Issac Newton (1642-1727):
http://en.wikipedia.org/wiki/Newton%E2% ... s_formulas
http://en.wikipedia.org/wiki/Newton_Raphson

[valhallasound]

Numerical integration of non-linear differential equations existed before circuits even existed to model with these methods. Most of the algorithms were invented in the 1600's. The main people involved with coming up with these algorithms are Gottfried Leibniz (1646-1716), and Sir Issac Newton (1642-1727):
http://en.wikipedia.org/wiki/Newton%E2% ... s_formulas
http://en.wikipedia.org/wiki/Newton_Raphson

The Drop is a great name for this plugin, as Andy's dropping some serious DSP SCIENCE here!

The math that Andy is discussing for the upcoming filters is different than the standard techniques of transforming analog filters into digital algorithms. It looks like he is attempting to address, not only the issues with stability at all cutoff frequencies (that are a big problem with, say, state variable filters), but the performance of the filter when time-varying. I am really looking forward to hearing the results.

Sean Costello

[mystran]

The math that Andy is discussing for the upcoming filters is different than the standard techniques of transforming analog filters into digital algorithms. It looks like he is attempting to address, not only the issues with stability at all cutoff frequencies (that are a big problem with, say, state variable filters), but the performance of the filter when time-varying. I am really looking forward to hearing the results.

I would like to claim that the issue with stability when it comes to the usual SVF is largely just a result of over-simplification in coefficient calculation; the structure itself is perfectly stable for any stable transfer function, but the way the coefficients are normally calculated doesn't lead to stable transfer functions for the full range of the design parameters. Like I've said before, you can pick any stable set of biquad coefficients and map them to a stable Chamberlin with the desired transfer function (though last time I suggested this, someone argued that the conversion is unacceptable since you need a square root). In fact I use a (very slightly modified) Chamberlin for all my biquads (most of which are designed with the usual cookbook formulas or something similar) now because of the nice numerical properties.

The question of time-varying behavior is kinda real though.

[meloco_go]

Numerical integration of non-linear differential equations existed before circuits even existed to model with these methods. Most of the algorithms were invented in the 1600's. The main people involved with coming up with these algorithms are Gottfried Leibniz (1646-1716), and Sir Issac Newton (1642-1727):
http://en.wikipedia.org/wiki/Newton%E2% ... s_formulas
http://en.wikipedia.org/wiki/Newton_Raphson

Ah, those are familiar! I used software for data approximation that used these approaches, although I'm guilty of not learning the inner-workings of algos too deep=) Would be quite exciting to see what you can do with it!
And I'm hopeful of many more high-quality plugins from you -- I think it is time to push the performance to a new level. I mean, I much happier use my i7 CPU to run 10 high-quality effects than 100 instances of plugins where compromises to cut down CPU usage were made.

As for Drop, I said that before in this thread -- I almost never use filtering of the type that Drop is, do you think this can be used to some tone-enhancement without using modulation?
I'm thinking of maybe using it in parallel, to get some "points" on certain frequencies by using resonances?

[andy-cytomic]

The math that Andy is discussing for the upcoming filters is different than the standard techniques of transforming analog filters into digital algorithms. It looks like he is attempting to address, not only the issues with stability at all cutoff frequencies (that are a big problem with, say, state variable filters), but the performance of the filter when time-varying. I am really looking forward to hearing the results.

I would like to claim that the issue with stability when it comes to the usual SVF is largely just a result of over-simplification in coefficient calculation; the structure itself is perfectly stable for any stable transfer function, but the way the coefficients are normally calculated doesn't lead to stable transfer functions for the full range of the design parameters. Like I've said before, you can pick any stable set of biquad coefficients and map them to a stable Chamberlin with the desired transfer function (though last time I suggested this, someone argued that the conversion is unacceptable since you need a square root). In fact I use a (very slightly modified) Chamberlin for all my biquads (most of which are designed with the usual cookbook formulas or something similar) now because of the nice numerical properties.

The question of time-varying behavior is kinda real though.

Unacceptable because of one sqrt? This is the kind of mentality that is keeping audio dsp in the dark ages. Are they trying to run this stuff on Commodore 64?

I did investigate a slightly modified Chamberlin algorithm, and with the help of Magnus Lidström came up with a normalising factor of 1 / sqrt (function of g and k) - I don't really count one division and a sqrt per sample much of an issue in terms of cpu since I'm interested in high quality filters. Although the filter is stable in linear form, and it does match the attenuation of an analog biqaud at cutoff, it turned out to be a big dead end, since as soon as you apply circuit non-linearities the normalisation breaks down and becomes unstable, and it didn't sound right when modulated either.

I have also derived the linear trapezoidal integrated svf, and unwrapped the terms so they are solved for directly to make the explicit, which is possible because the equations are linear. This form has excellent noise properties, and has the most superior coefficient rounding error, and quantization error of any 2 pole explicit filter structure I know of, and here are the plots to show what I mean (these plots are for dsp geeks so be warned!): http://www.cytomic.com/files/dsp/SVF-vs-DF1.pdf

The only way I have found to get the non-linearities and modulation to sound right is to use implicit numerical methods via newton raphson iterations, which means possibly multiple divisions per sample as the algorithm converges to a solution. These methods solve for the non-linear terms with no delay in the feedback loops of the circuit being modelled, and they sound silky smooth all the way up high, and modulate beautifully. This is what you can hear in the audio examples of The Drop, and I really don't care if it takes twice or three times the cpu of more basic filters, I am interested in it sounding good. It actually turns out most of the time only one division is needed anyway so the method is highly efficient, and division is very fast on modern cpus.