Hi folks - We've been experimenting with a higher bit rate mjpeg video stream. This message describes what we are doing and asks if anyone wants to setup a similar machine so that we can do experiments across Internet2. We have a working prototype that sends mjpeg RTP streams using either unicast or multicast at bit rates between 3 - 30 Mbs. The streams can be 30 fps full-sized D1 or CIF images. We're using it to play video from a satellite dish on another computer. The setup works as follows: 1. Both the send and receive hosts are Linux PC's with a Linux Media Labs LML33 mjpeg card ($410 at http://www.linuxmedialabs.com/). 2. The composite output of the video source is connected to the card. 3. We run a program, named "rtpvb" for RTP Video Bridge, that grabs the video and transmits it. 4. The receive side rtpvb receives the packets, reassembles the jpeg data and passes it to the decoding board. Currently the video must be displayed on a separate NTSC display (TV). We're working on getting it to run in overlay mode so the video output, at least for the CIF stream, can be displayed in a window on the screen. The board uses a Zoran jpeg codec chip. Interestingly enough, the encoder does rate control by adjusting the quantization table on a frame-by-frame basis. This caused some problems with RTP since the normal way of specifying the table required that the table variation match a particular model for encoding it in one number - using the Q parameter in the RTP jpeg header. Using the rate control, the chip produced a constrained bit rate - i.e., default was 5 Mbs and it stayed below that. We added features to the device driver that lets us modify the bit rate dynamically - we've run tests up to 25 Mbs. We also tried setting the board so that it would use a fixed quantization table - the attached message reports on the bit rate variation - in short 3.6-13 Mbs which is actually a pretty large range. SO, the question is whether anyone connected to Internet2 has a Linux box and TV that they'd like to use for some experiments? You'll need to get an LML33 board and install the drivers we have. If you're interested in participating in this experiment, let me know. Larry -- Professor Lawrence A. Rowe Internet: Rowe@BMRC.Berkeley.EDU Computer Science Division - EECS Phone: 510-642-5117 University of California, Berkeley Fax: 510-642-5615 Berkeley, CA 94720-1776 URL: http://bmrc.berkeley.edu/~larry
-- BEGIN included message
- To: "Lawrence A. Rowe" <Rowe@BMRC.Berkeley.EDU>
- Subject: fixed Q table bitrates
- From: Matthew Delco <delco@CS.Berkeley.EDU>
- Date: Wed, 23 Aug 2000 15:41:54 -0700 (PDT)
- In-Reply-To: <39A44FE6.B0084001@BMRC.Berkeley.EDU>
I added the sender code to track bits/second over 5 second intervals, and so far the range of values I get is between 3,672Kbps and 13,261Kbps. This range seems much larger that what I'd expect (the max is basically 3.6x the min), although the most common values are generally around 4.7Mbps--during commercials (especially those with bright colors) the average is closer to 6.7Mbps (with occasional peaks in the 12Mbps range). MD
-- END included message