[sdiy] Designing Analog ICs

[Scott Nordlund]

> This is essentially what I was trying to do with my PIC projects some years ago. Take a digital chip with analog I/O and hide all the digital stuff inside the chip, so that from a users point of view, it acts just like an analog chip. I was reasonably successful for low-speed modulation stuff, even using only a $1.50 PIC.
>
> What technology would I be looking at to do the same for audio? At 96KHz? At 192KHz/24-bit? FPGA?
The Spin Semiconductor FV-1 has an onboard stereo codec. I think it's 48 kHz, 128 instructions per sample. The Alesis Ion/Micron use a Wavefront AL3101 (DSP-1K) per voice, and another (9 total) for effects. It's 48 kHz, 1024 instructions per sample, but no codec.  They all have onboard RAM/ROM and instruction sets geared for synthesis/effects. For comparison,  I think the DSP chip in the Yamaha SPX90 (YM3804) is 31.25 kHz, 64 instructions per sample, with no onboard codec or RAM, just a small control store that's modified by an external chip for modulation.
> It seems to me that digital stuff has got good enough now that you *can* build something which is functionally identical. People always go "Ah, but what about the interesting non-linearities that analog introduces? Digital is always too clean..." but that argument doesn't really hold water any more, since you can accurately model whatever non-linearities you want too. It's a lot more work than building a clean design, but it's possible - the history of the digital Moog filter emulation is essentially that process.
I think mostly what's held back digital synths is the necessity for compromise. Engineers got a surprising amount of performance out of modest hardware, but it came at the cost of sub-optimal sound quality- low sample rates, slow modulation, cheap interpolation, compressed sample data, etc. We're in a much better position now to implement algorithms that offer really good quality without cutting corners. BUT I think software still offers a lot more flexibility here, since you can scale and reconfigure things arbitrarily, and render offline at obscene sample rates.  And some of the advantages of analog (feedback especially) aren't really compatible with a DSP/analog hybrid format, due to latency. For me, I think Pure Data (or Max/MSP or Reaktor or whatever) still comes out on top as a geek-enthusiast DSP platform.