[sdiy] Bus for digital patching of synths

[cheater00 .]

Hi all,
I've recently been talking with someone about an idea for patching
modular synths in digital rather in analog.

Let me restate the problem. We have an analog modular. It uses a lot
of patch cables, which can get difficult to manage, and becomes even
worse in consideration of polyphonic modulars. Instead, put an ADC on
each output and a DAC on each input and connect them with the right
bus. As a nearly free bonus you get patch storage. The question is
what bus is right.

My first thought was CAN. I've just had a closer look into CAN. Some
people say can is only good up to 1 Mb/s, some say it depends on the
uC. However, perhaps the format of can itself isn't optimal. Here's
what a frame looks like:

http://en.wikipedia.org/wiki/File:CAN-Bus-frame_in_base_format_without_stuffbits.png

the "data" part can be longer, but still that's a bit of overhead.

So far I've identified that ideal bus would have the following features:

1. it is shared. That is, multiple devices share the same medium.
2. it's oriented towards setting up constant streams of data between
any two endpoints and having them active for a long time
3. shares bandwidth via TDM
4. after setup, the data should be sent bare, other than infrequent
housekeeping frames. This means once your link is established you
should be able to take a byte from the DAC and output it to the bus
directly.
5. only needs to be very local, say 2-4 meters
6. it is switched. This way the collision domains are made smaller and
network bandwidth utilized much more efficiently. So you could group
your modular into domains of 10 modules and have a fabric connecting
them.
7. needs fairly high speeds (~50-100 megabits/sec). This means on a
single collision domain, there could be 10 monophonic modules which
each have 6 inputs or outputs at 100 kHz at 20 bits (this is a rough
estimate so far). With a fully switched network with a 1 Gb/s backbone
you can realize an 8-voice or even 16-voice modular, or just use
separate networks for each voice. I have a feeling using a single
high-bandwidth network will however mean much simpler synchronization,
whereas in disparate networks the frames, link times, etc might not
line up and so some voices will receive sound later, making for
unpredictability in edge cases - voices would sound different at
extreme settings.
8. it does not do any CRC or delivery assurance. To minimize delay
there is no time for this overhead.
9. needs to have hardware support in a variety of microcontrollers

So far I feel like I'm mostly describing copper Ethernet, unless
someone can point me to something different/better.

Here are the features of Ethernet which aren't exactly necessary:
1. You don't need this to work over 100 meters or run over UTP.
However, being able to patch together fairly distant modulars (like in
another studio) could be a door opener for many things. Optical links
could take care of the ground potential issue over long distances.
2. You don't need galvanic isolation. However, magnetics are cheap,
and you might find them good to have: digital mess is decoupled from
your module, it lessens the chance of incompatibility between devices,
no need to meticulously carry differential lines over connectors and
larger distances, etc. Think of MIDI optocouplers. With very
inexpensive switches being built by the billion, you can surely find
something down at double cents. Just don't go to Element 14 ("Formerly
Farnell")
3. You don't need the ethernet frame and everything it brings with
itself. For one thing it's a huge amount of overhead unless you buffer
the data, which then means delay.

The last point is a fairly big issue. Maybe there are some tricks with
Ethernet to make it transport data in a TDM fashion. Maybe that makes
the standard unsuited.

This question is relevant for the more complicated polysynths as well.
If you were making a new CS80, and wanted it to be highly
configurable, you might want to consider digital patching. For lower
requirements, like simple monosynths, a lower-end, more localized bus
such as CAN might be advantageous, that is if digital patching is
required at all.

D.