Re: Intel & Video Conferencing

Dick Cogger (R.Cogger@cornell.edu)
Sun, 28 May 1995 17:26:48 -0400


At 3:11 PM 5/27/95, Don Johnson wrote:
>About 3 weeks ago I read in Computer Reseller News that Intel will produce
>video conferencing hardware technology for resell at about $300.00. They
>apparently want to establish the market very much like their CPU's did.
>Possibly even to the detriment of their own commercial product line of
>Pro-Share.
>
>I would very much like to see a discussion of software based systems (ie.
>CUSEEME) vs. hardware. Obviously, I'm not talking about external periferals
>such as cameras neccessary to any system, but of compression technology and
>video handling methods, etc. Also, I have heard that the power pc chip has
>features superior to pentium technology in this regard.
>

In setting out to build CU-SeeMe, we chose to aim at the low end,
reasoning that plenty of others were working the high end. Also, we chose
early on to focus on multiparty conferencing, with multiple windows showing
other participants. The low end implies software-only codec, pretty much.
However, even when specialized hardware is available at an affordable
price, multi-party implies that you need to be able to decode multiple
video streams at once. What that means is that decoding must be cheap.
Encoding usually is more work than decoding, but the real powerful
compression techniques need quite a bit of compute power on the receive
side.
In any case, I havn't heard of specialized chip-sets for video that
will decode multiple streams. Since it is only recently that many folks
have believed the Internet to be real, most designs (Ptel, etc.) have been
oriented toward dial-up, point-to-point sceanarios where you only get one
stream coming in (maybe getting switched between sources by a MCU), so
there has been little perceived need (I'm guessing) to design highly
asymetric codings or to implement chips that do multi-stream decodings.
With the increasing power of general-purpose desktop systems, a
nice approach might be to (1) design an asymetric coding and build special
hardware for doing the heavy work on the send side (after all, you need
digitizing hardware anyway to be a sender), and (2) do decoding in
software. I'd like to hear about any efforts going on to implemnt this
approach.

So I don't think you're going to see 320x240, full color, 20fps,
using less than 50Kbps/video stream anytime soon in a system where you can
have 7 remote windows showing. But it's all about tradeoffs-- I realize
that with a modem link, you're not going to be able to do much of a job
showing 7 windows anyway. Also, if you assume 2-party, it's a lot easier
to support various workspace sharing and remote control paradigms. We'll
probably get around to better 2-party support, but for the most part, at
Cornell, we'll keep our focus on multiparty with all that implies.

Cheers, -Dick

-Dick