Sampo - please read my original post - [...]
Will read.
I have friends who are looking for a something like this as a cheap
multi channel solution but to be honest I wouldn't trust it for this
gig....
Since what I'm talking about doesn't exist and your gig is drawing near,
neither would I.
But given some time and some surprisingly minimal effort, what I mean to
say is that you can approach thorough perfection along the lines I
jotted out above, using reasonably priced hardware. What I'm thinking
about isn't a second best solution you can best with dedicated hardware
sync; it's the first best solution no hardware solution can even
theoretically best, at any price point, because of the trouble with high
fanout clock distribution.
*If* you can drive the differential delay of your distributed converter
bank to any quantifiable level, you *also* can drive it *much* lower
with minimal extra cost, using something like the design I alluded to
above. Then, it's possibly to get equivalent performance from a setup
much cheaper once you trade back to cost efficiency using that better
tech.
Such a tradeoff isn't theoretical. It can be done now, if somebody just
puts hir mind to it. It can then make the difference between a
prohitively expensive, ideal by pure analogue clock distribution
standards rig, which never sees the light of day, or if it does, only
proves a one time wonder; and an easily scalable, easily perceptually
competitive rig, which when once even half perfected, could serve as a
generalized, affordable substrate for all high channel count work in the
future.
Maybe I missed something; as you say, I'll have to reread or original
post in full.
But if I *didn't* miss anything, I believe what you're after is simply a
high number of well synched channels, at a reasonable price point. You
want to get 22 of them right? Possibly more? You want them to be child
and bomb proof, too, right?
You can't get that kind of hardware in over seven channels. You can't
get it compactly in over for, or perhaps six. Even then the stuff will
take the form of developer boards or SoC based hacker boards such as the
Pi.
So, you're going to have to do some integration work in any case for
channel counts as high as you're asking for. My favourite would be
something like a Raspberry Pi for each eight channels, stacking the D/A
converters on top on 2x4 daughterboards. That'd buy you 8 channels per
board, with ample processing power and Ethernet connectivity to spare.
You could even push it as far as three daughter boards, so 12 channels
per motherboard, and you still wouldn't saturate either of the USB or
the Ethernet port. But you be pushing the processor quite a lot already,
if you did any substantial processing, such as well-resampled fractional
sample delay correction which I suggested above.
In any case, if you want to make the thing truly nuclear proof, what you
do is spray the whole circuit board with insulating spray lacquer. All
connectors attached and with all contacts wetted with mercury wapor,
alloyed by a tiny impurity of silver chlorate. Let dry. Reapply in small
patches the lacquer, then hairlike aluminium shavings to cover the
thing. If you need to form thermal conduits via what follows, solder
them in place before the first lacquer application, and file/wet as need
be to ensure thermal contact, while retaining electrical insulation by
air, around; after the first application of lacquer, it displaces the
air in the negative space around in a non-geometry-deforming matter, and
takes the place of the prerequisite insulator/dieletric in any free
circuits (obviously anything like an open, air core coil would now be
fucked up by the new dielectric; don't use them, or if you do, calibrate
for the new medium beforehand). If you really want to be thorough and
weird, you'd wet the newest surface with slightly chemically polarized
mercury vapour again, imbibed with a small amount of metallic silver.
You'd single point ground the resulting, mirror-like surface one-point,
as the Faraday cage that it is.
Make sure every lead going into and coming out of the circuit has a
small, smooth, exactly toroidal ferrite ring goung around it where it
exits the black that we're going to make out of the circuit, below. Then
cast the whole thing in solid two part resin, reaching at least 2-3
millimetres from the outest envelope of the encased circuit board and
all of its components, in any and all directions. Use porous paper to
maker sure the ferrite ring derived exists for any leads stay clear of
the resin, so that the know beneath takes the stress if somebody
stretches anything, so that such tensile stress doesn't break one of the
rings and so its slight (but rising) HF suppression characteristic, and
so that the ring can at the same time serve as a permanent, rounded
outlet to the lead going through it, so any stresses from any direction
lead to approximate minimum shearing in all directions at the outlet,
over the inner curvature of the ring.
No voids, eversosmall, would otherwise be allowed. The whole end product
must be slowly poured from its corner, so as to eventually look like
something preserved in amber.
Once you learn to do that sort of treatment in bulk, pipelined, you can
easily turn out the circuits at a rate of some 30-50 per hour. If you
really need to. They will be repeated-nuclear-strike-EMP survivable, you
can feed them to your pet crocodile and have them work all the way
through without loss of function, given a sturdy enough cable
interfacing with the thing you can take hold of it and beat a grown man
dead with your D/A/computerpack, you can dip the thing in most acids and
and bases with no discernible damage, and not to mention, choosing your
resin right, you can have your smallest of babies lick, chew and fondle
at the thing, with no adverse health effects.
But it'd still go with my original, rather less expensive and involved
design, using software to compensate for the timing inaccuracies of
unmodified commodity hardware. ;)
--
Sampo Syreeni, aka decoy - ***@iki.fi, http://decoy.iki.fi/front
+358-40-3255353, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2
