[sdiy] Analog crossbars and backplanes

cheater00 . cheater00 at gmail.com
Sun Nov 10 23:30:41 CET 2013

On Sun, Nov 10, 2013 at 10:58 PM, Tom Wiltshire <tom at electricdruid.net> wrote:
> My apologies, Damian. I thought "16 x 16 crosspoint switch" meant it gave you all the points in the matrix. Apparently not, at least for AD.

No issue, asking questions about practicality and desirability is very
important. I hoped to address those in the previous two emails. I
still wish I had the time to devote to counting those patch cords. It
would be a really nice bit of data for our community to have.

> Still, I'm sure I've seen genuine analog switch arrays somewhere, where you can connect any of the X/Y points. I'll try and find it. 16x16 is only 256 switches, which is hardly beyond modern integrated circuit design.

Are you talking about a sort of "fabric network", where you hook
inputs and outputs to any point on XY and the available routes on the
fabric are allocated to directly connect them?

Magnus pointed out that I'm not great with signal theory. I have a
feeling that those have well understood limitations as well (do you
know what they are, Magnus?)

> And if you need bigger X and bigger Y, you only need gang up a few ICs.

You can gang the AD part as well. But that means even more crosstalk.

> Something more like this:
> http://www.microsemi.com/products/switches/analog-cross-point-switches/mt8808

This is a smaller version of the AD part. I'm not sure whether its
performance is better or worse. However, just the internal resistance
means that between different switches on the same chip, and including
thermal drift, one might have the system go out of tune by nearly a
semitone. Assume a 12V signal, in a 1 V/Oct system, from a source with
60 Ohm output impedance fed via the MT8808 to an input of 10 kOhm:

>>> (12 * 10000/(60 + 25 + 10000)) - (12 * 10000/(60 + 90 + 10000)) # possible drift
>>> 1/12 # one semitone in volts

I don't know if going much higher with the input impedance would
degrade the system in some other way (emi, improper loading for the
switch, ...).

> This is available as a 28-pin DIP and with four chips + a uP you could do a 16 x 16 programmable matrix. Obviously as the matrix gets larger, the number of chips goes up exponentially, so you still can't go crazy.
> This MT8808 IC doesn't offer any figures for DC accuracy, but it's a basic analog switch with Ron between 40R and 80R, so I suppose we'd have to design something for which that range was negligible and gave a satisfactory result. That means a decently high impedance on the mixers for the output buses.

I have more of a feeling this is meant for distribution amplifiers,
and the switches are meant to be buildouts / splits.

I can see how this would work OK there, but it's not suited for analog
computing (which is a more general term for analog synthesizer that's
more accepted in the electronics industry as an application domain
from what I gather).


More information about the Synth-diy mailing list