[sdiy] Virtual Analog ICs was Re: Designing Analog ICs

[Scott Nordlund]

> The Spin chip is pretty much what I'm after, it's true, but I was wanting a bit more horsepower than it offers. It's one of those "just good enough" DSPs - built for budget effects, since that's the mass market. The use of floating point storage internally is clever though, and gets the best out of a small chip.>
> If there was a "FV-1000", with more RAM/ROM, and a 96KHz/192KHz sample rate, I'd be on it like shot!
Unfortunately the faster ones always seem to be made with the assumption that you're putting them in a larger system with more infrastructure. Maybe the idea of a pocket sized stompbox (or whatever) with 50 MIPS on tap is too much to imagine. As far as the manufacturers are concerned, maybe it's an answer to a question (almost) no one is asking. 
> I absolutely agree. Which is why the Spin chip isn't quite enough. If I want to do <boring example alert!> a Moog filter emulation, I need to have some space for those nice sigmoid distortion functions between each stage that we discussed a few weeks back. I won't be able to get that sort of thing in the Spin chip. And the reason for the high sample rates is not because I'm some audiophile nutter who thinks he can hear up to 100KHz but rather because many filter algorithms start to show problems when you get up close to nyquist, and running at double-rate or better is a simple solution to those problems. I suppose I could always interpolate on the way in, process at the higher rate, and then filter and downsample on the way out - but's that'd be even more work, and even more processing.
Of course. And I think some recent designs are working along those lines, like the Solaris. But I'd also guess a lot of manufacturers (Access?) are probably coasting on decade-old algorithms. We never get any transparency with this stuff, just some new buzzwords. 
> But hang on! - This *IS* software, isn't it? At least, I can put whatever code I like in this processor and make it do whatever I like. Ok, it's realtime, and that means there's a time limit per sample, but real time is the name of the game if I'm trying to copy analog ICs. As the sample rates get higher, the latency can go down - to one sample in theory, and that's how I'd want it in pratice. I agree that PD/Max/Etc are great for testing weird geek-out DSP ideas, but for me it's still not a synth until it's dedicated hardware - although these days, that mostly means the interface rather than the guts (Nord stuff being one example).

My take on it is that embedded/self contained stuff is better for low latency hands-on response, and interfacing with stuff that you either already have or can't easily model- treat it like a rack mount unit and patch it between your fuzz pedal and tape delay. Desktop or DAW type stuff has the advantage of arbitrary copy and paste reconfigurability, and batch/offline processing at obsessive sample rates. In Pd I find myself redesigning algorithms almost every time I use them- not totally from scratch, but making improvements here and there, or re-arranging things for a specific recording.  I can get inside it whenever I want; it doesn't have to be a black box when I'm not actively developing it. It evolves according to how it's actually used, and that's a huge advantage.
Maybe it's the Berlin School/Kosmische model versus the IRCAM/Acousmatic model. Of course in actual use, it's generally preferable to have something more hands-on and intuitive. But if we're already crazy enough to break out the soldering iron, oscilloscope, textbook and/or compiler to build and modify our tools, I don't think minor (or even major) user interface struggles are a deal-breaker here. I think we've all sort of collectively agreed by our participation here that it's worth some effort and inconvenience to get something not available in a mainstream commercial product, otherwise we'd be using ROMpler presets like everyone else.