[sdiy] Digital modular backplane - update

[cheater00 .]

Hi John and Rutger,
thanks for your emails. First of all, welcome to the list - I have not
spoken with you before, and I guess this is one of the first posts for
each of you.

I am guessing you don't recollect, or haven't been involved with, the
original conversation from November 2013. I have had to ask Paul for
permission, but I have now forwarded the whole original thread to the
list. The thread is not all in the synth-diy archive, which is
troubling.


On Thu, Mar 20, 2014 at 4:00 AM, john slee <indigoid at oldcorollas.org> wrote:
> Hi,
>
> On 17 March 2014 22:24, cheater00 . <cheater00 at gmail.com> wrote:
>> Some time ago we have spoken about what it would take to make a
>> digital backplane for a modular synth.
>
> What problem is this attempting to solve? Being able to save signal
> routing and control state in a patch would be nice, I suppose, but
>
> - physical settings != patch settings
> => replace controls with "soft equivalents" => encoders/momentaries
> => current state no longer obvious => more indicator lights etc
> => may require more front panel real estate than "real" controls
> => more $$$
>
> What other benefits of going all modern do you see?
>
> John

The basic idea (which you will find more details on in the original
thread I have just forwarded) is that you use analog modules. Those
modules are connected via digital patch cables. Much like Rutger
suggests below...


On Sat, Mar 22, 2014 at 12:05 PM, Rutger Vlek <rutgervlek at gmail.com> wrote:
> I've been thinking about the concept of digital modulars too,
> but I think that until digital technology advances much further
> (lower latency, higher sampling rate, more CPU power, high
> bus speeds) the key differences between analog and digital
> domains will limit the usefulness of a 'digital modular' approach.
> Some things sound better, and are more cost effective, in the
> analog domain: such as oscillators and filters with lots of
> character, a little bit of drifting, etc. And some things sound
> better, or are more cost effective in digital: noise properties,
> patch storage, complex connections that would otherwise
> require lots of messy cables.


...the idea is to take the best of analog, and the best of digital.
Analog is obviously better at generating and processing sound. Digital
is better at moving it around, meaning recording and playback. Michael
Jackson's album Thriller was multitracked in digital.

My latest update, which I again include below (as it's been 10 days
and many readers have likely deleted the original post), talks about a
possibility of realizing a modular bus for hauling a lot of audio
around in digital with as little latency as possible. Ideally the
latency should be as much as it takes to digitize a single bit of
signal, send that single bit, and then reproduce it with a DAC. This
should make even feedback topologies useful.

Naturally, for situations where feedback isn't good in digital,
front-panel patch points should still be available.

Cheers,
Damian

On Mon, Mar 17, 2014 at 12:24 PM, cheater00 . <cheater00 at gmail.com> wrote:
> Hi guys,
> Some time ago we have spoken about what it would take to make a
> digital backplane for a modular synth. One of the issues was that most
> common digital interconnects will introduce large delays in
> transporting the audio which are not acceptable in modular synthesis.
> A requirement was mentioned of ideally having the delay as low as the
> transmission time of 1 bit. This has to take into account the
> situation that an ADC or DAC will operate at low clock speeds, while
> the backplane would operate at very high clock rates, in order to
> accomodate many ADC-DAC links in the switched, TDM fabric.
>
> I have come across the idea of using a SerDes:
>
> http://en.wikipedia.org/wiki/SerDes
>
> They are basically shift registers with additional ISO OSI Layer 1
> processing. in specific, it seems a popular type uses 8b/10b encoding
> which limits the RFI impact, and makes the layout much easier as the
> lower bandwidth limit of the physical links goes further up, while the
> upper bandwidth limit doesn't move.
>
> In addition this device family addresses the issue of having slow
> links on fast backplanes:
>
> "Bit interleaved SerDes multiplexes several slower serial data streams
> into faster serial streams, and the receiver demultiplexes the faster
> bit streams back to slower streams."
>
> It seems in this case there is no bunching or buffering so the latency
> can be kept to a minimum.
>
> Silicon for SerDes applications exists and is popularly used in loads
> of consumer technologies:
>
> "Among the areas in which 8b/10b encoding finds application are the following:
>
> PCI Express at speeds below 8.0 GT/s
> IEEE 1394b
> Serial ATA
> SAS
> Fibre Channel
> SSA
> Gigabit Ethernet (except for the twisted pair based 1000Base-T)
> InfiniBand
> XAUI
> Serial RapidIO
> DVB Asynchronous Serial Interface (ASI)
> DisplayPort Main Link
> DVI and HDMI Video Island (transition-minimized differential signaling)
> HyperTransport
> Common Public Radio Interface (CPRI)
> OBSAI RP3 interface
> USB 3.0
> CoaXPress
> MIPI M-PHY[6]
> ServerNet (From ServerNet2 onward)"
>
> It might be possible to find switched fabric chips that can route the
> links on an X/Y grid for best bandwidth utilization, and which can do
> so without buffering. At least that is what I would expect of what
> Infiniband, Gb Ethernet, and Fibre Channel are doing.
>
> Cheers,
> D.