19/03/2004 18:08:05, Paul Carpenter
<
paul@pcserv.demon.co.uk> wrote:
>From what I can see of the Sony information they must be old!
They claimed
>to have stopped shipping in 93...
Indeed - but lenses don't fester at the same rate silicon does.
They're still nice, useful lumps of glassware, with some irrelevant
silicon tacked on the side.
>The horrible Sony communications 'protocol'... It was a bit of a
nightmare
>last time I lloked at it.
Yeah, LANC and its evil offspring. However, it doesn't seem to
want to talk LANC...
>For Webcam the last thing you need is resolution, most webcams
give a quarter
>VGA resolution at best, anyway the data transmission becomes
the bottleneck.
Accepted, but for things I want to do, VGA is baseline. Frame rate
is much less important than resolution, and bandwidth is (almost)
irrelevant.
>For different and simpler electronics I would look at CMOS
sensors like
>Omnivision, all the processing on chip and some versions have
digital o/ps.
I've got some usable Conaxant imager chips, and the Foveon
devices also look very tempting for my needs. Too busy with real
work to play at the moment, though.
>Hmmmm... large chunk of memory for a stream then, as
768x576 25 times a
>second is a lot of data to hand off. Probably with flags for
frames
>available etc..
Handing a PAL stream into a cheapie PVR MPEG compression chip
is the sort of thing I had in mind as being useful, or at least fun to
play with. The PXA isn't ever going to be ideal for image
mangling...
(we've been using TI TMS320DM642 DSPs in products recently -
pretty cheap, and ferociously fast fixed-point. I['ve been
impressed)
> Grabbing direct to internal RAM
>even for a field or frame can significantly reduce the board
space
>and overhead required.
PXA DMA isn't much of a win, though. You might as well just do
the transfer with the CPU LDM / STM, since you're going to be
tying up the entire external bus, and the DMA engine still requires
you to read & write every word.
DMA of that style is only really useful for middling-rate data
transfers, when data is trickling in faster than you'd like to take
interrupts - it's not actally a speed gain (or is it? Have I missed a
trick? Possibly the core won't stall while the DMA's going on, but
I'm not sure - not dug very deeply on this)
>For single field/frame grab software PIO transfer becomes
possible.
Indeed. It's all a case of who wants what...
>I will double check the Gerbers
let me know if you can't find what you need, and I'll generate it.
The backplane connector is the one thing that's really nailed
down, though. Most other things are negotiable, but the backplane
connector stays where it is, and should only ever gain functions,
not lose them...
Steve
(text/plain)