[sdiy] Clock-controlled filters?

cheater cheater cheater00 at gmail.com
Mon Oct 22 23:25:56 CEST 2012


Hi Scott, thanks for your interesting email.

On Mon, Oct 22, 2012 at 10:50 PM, Scott Nordlund <gsn10 at hotmail.com> wrote:
> Look at the service manuals for the Kurzweil 250 and Akai S900. They do exactly what you want (variable rate DAC and switched capacitor filter).

I didn't know those two did that. Good to know! I know the emulator
did it, or at least seem to recall that it did.

> Anyway, variable rate isn't really variable unless it's clocked from a VCO.

Isn't this great? That's exactly what I've been thinking of.

> If you're not doing something tricky (using analog electronics to generate a cleaner clock signal, which I think no one has done in any musical application), it's equivalent to a fixed rate system that uses the master clock frequency as the sample rate.  The "variable clock" that the DAC sees is going to have the same jitter as an ordinary variable increment phase accumulator.

Even if I wanted to go digital... even with a Z80... at 20 MHz the
jitter not audible. But no, it's not like "oversampling to 20 MHz".
This many interpolations - even linear interpolations - would require
a very, very fast cpu. Even that many *lookups* would need a huge
amount of bandwidth. And then you need to implement the digital
oversampling filter, as well. Instead, here, you could have your 44
kHz DAC with a jitter of say 0.2% period, and a nice imaging filter at
the output of the DAC to smooth it down. However, I would be concerned
about the clock error being accumulative - I'm not great with uC's,
but without an RTC that has microsecond precision and is very light to
query in terms of clock cycles I wouldn't know how to implement a
clock output that is asynchronous to the master clock. Plus, analogs
drift and do other junk that digitals don't, so that's nice to have as
well.

> You can get the same result in software by simply oversampling.

I don't think the two are equivalent. For one thing, resampling in
digital brings artifacts which, by the way, are the reason you're
doing your resample-to-fraction-of-fs thing you describe below.

> I was thinking along the same lines as you, but unless you're going to do something that's absolutely overkill (and aside from price and complexity, overkill approaches have their own problems), I think it's not worth it.

Can you describe this in more detail, Scott?

> I'm getting better results in software by constraining frequencies to subharmonics of the sample rate (i.e. periods are always integer numbers of samples). I'll have to write more about that later.

Rational resampling always has great effect. I'm guessing you use it
in an effects box or as a post-vco module.

Cheers,
D.



More information about the Synth-diy mailing list