[sdiy] ADDA & MIDI over Ethernet (long)

[ASSI]

On Monday 26 May 2003 05:33, j vallier wrote:
> >I've looked at both mLAN and MaGIC and both seem to be needlessly
> >complicated if not to say bloated. I'll have to think about this
> > some more...
>
> You are right--these specs (and others) are quite complex--what you
> see as "needlessly bloated" is the fact that these full
> specifications need to be "everything to everybody" and get rather
> involved (ever try reading the 802.11 spec???).

I might agree when it comes to mLAN as it has to fit into the quite 
complex scheme of how IEEE1394 is working.  But for MaGIC this argument 
doesn't hold IMHO, as it tries to comply with parts of the protocol 
stack of IEEE802.3 that are only a hindrance for this application. 
Streaming data and packets don't fit together unless you use a really 
big hammer. This would only make sense if you were to share the 
physical medium with normal Ethernet traffic, which is something I'd 
avoid at all cost. It might make sense when you try to use existing 
backbone links, but then existing QoS profiles and a tunneling engine 
at either endpoint is better than grafting all that bloat onto each 
single port.

The feature set I want looks like this: 
- A/D and D/A should be as close as possible to the producing
  or consuming device
- audio channels and MIDI/control can be bundled into cables
- channels can optionally be divided into subchannels for increased
  capacity at lower quality
- a cable has head and tail and provides up- (tail to head)
  and downstream (head to tail) simultaneously
- each device provides two physical ports per cable to allow
  chaining of devices until the maximum capacity in terms of
  audio channels is reached
- a device is called a mixer if it accomodates connections to at
  least two cables
- the master functions, for instance clock and conflict resolution,
  are negotiated on each cable to be assumed by the most capable
  device, with the device nearer to the head of the cable taking
  precedence

I give you that this can be reaonably implemented in both mLAN and 
MaGIC, but how much will the ports on each device cost? Analog mixer 
ports can be had for about $15 a pop, but I don't see these going below 
$50 anytime soon.

For me IEEE1394 is dead because to even read the spec I need to shell 
out $1000 (I'm still unsure whether I even get this CD when I'm not a 
member) and becoming a member runs me $4000 a year plus they really 
don't think individuals are existing, so I have to be a company. On the 
other hand, getting the full spec on IEEE802 costs no more than 
acknowledging some legalese and waiting for the download to complete. 
Further you find many more FastEthernet PHY chips and/or uC with 
builtin PHY and even complete switch fabrics at quite reasonable cost.

Now looking again at the requirements and translating that to IEEE802.3:
each cable is made up of 100Base-TX segments, each port is fixed 
capability at 100MBit/full duplex and there is no need for any 
addressing scheme at the port level (but you need channel naming in the 
control data), so you could well drop both the LLC and MAC layers and 
work just down from the MII level. The master constructs frames at a 
fixed rate of 192kHz and with a fixed format, any slack is filled so 
that the medium is never idle when there is a master. Each device takes 
in the frames (synchronizing clock to the master along the way), works 
on the slots it owns and sends it out again in cut-through style. This 
gets us to 90% of the theoretical channel capacity whereas MaGIC can 
use only less than about 70% of it.


Achim.
-- +<[ Q+ & Matrix-12 & WAVE#46 & microQkb Omega sonic heaven ]>+ --

Factory and User Sound Singles for Waldorf Q, microQ and rackAttack:
http://homepages.compuserve.de/Stromeko#WaldorfSounds