[sdiy] Analog crossbars and backplanes

[Magnus Danielson]

On 11/09/2013 04:15 PM, cheater00 . wrote:
> On Sat, Nov 9, 2013 at 3:50 PM, Rainer Buchty <rainer at buchty.net> wrote:
>> I think that an programmable analog crossbar is the better
>> solution in your case. Saves you all the hassles of dealing with high-speed
>> buses and you don't even need tons of ADC/DAC. Plus, scalability is rather
>> easy to achieve.
> Dunno. Is it?
>
> Let's have a 16-voice modular, with a module that has 6 connections.
> Don't you think this is a mite huge? That's 96 connections per module.
> A reasonably sized modular will have on the order of 100-200 signal
> sources and destinations. So you'd need 16 completely separate analog
> systems that can each manage 100 inputs and outputs and route between
> them. Are there any analog crossbar solutions that do anything on this
> order of magnitude?
> Then, each connection to the main hub must have additional line
> drivers, because we're talking about distances on the order of a meter
> or two.
>
> If you find something nice, let me know, but the last time we've
> visited this topic on the list, there wasn't anything affordable or
> interesting.
First off all, you usually do not require the full 96x96 patch-freedom.
With a 16 voice modular, you however most probably want high "local"
connectivity first, and then share a few signals with others.

Another aspect, you don't build the full 96x96 matrix. The larger matrix
you build, the fewer of those matrix-points will get populated, giving a
good hint that you have grossly overdimensioned the system.

The problem has been solved in the telephony-world, and it was based on
some interesting math, but the key ideas is not that hard to understand.

One switch-type is the Clos-switch, in which you have MxNxO channels. In
the CLOS switch, you have O number of input switches, having N inputs
and M outputs. The core contains M switches of OxO size, and the output
contains O number of switches containing MxN switching. If you let N be
larger than M, then you will not have full interconnectivity, you will
only have M active signals out. For you case, you say that O = 16 and M
= 6, so you would require 16 input switches of 6x6, 6 switches of 16x16
and 16 output switches of 6x6. If we build everything in 8x8 switches,
you would require 16x1 + 6x4 + 16x1 = 32+24 = 56 8x8 switches to achieve
this. Compared to the full cross-bar it would be 12x12 = 144 8x8 switches.
As you look at different variants and sizes, you realize that you can
build it in various shapes and forms.

The Benes switch is a special case worthy of studying.

http://en.wikipedia.org/wiki/Clos_network

Clos and Benes strategies for routing signals should be well understood,
as proper use can save you a lot of money and wiring. The one thing they
are not very good at, is one-to-many.

A Benes network is easy to route through in such a way that you achieve
full interconnectivity.

Anyway, so it's not that you can't build large interconnects, it's that
you need to use the tricks that has been developed. Fortunatly, this is
a well-researched problem.

Cheers,
Magnus