Production TV Quality Berkeley MIG Seminar

[ MIG Seminar Announcement | RTPtv Software | Open Mash Software | RTPtv Report ]

The Berkeley Multimedia, Interfaces, and Graphics (MIG) Seminar has webcast seminars on leading edge topics weekly since January 1995. The webcast started as a single audio and video stream transmitted over the Internet Mbone. During the past six years we have added a second video stream, improved the production value of the captured content, and added multiple transmissions using different technologies. The past couple of semesters we have produced two webcasts: 1) a dual-stream medium-bitrate Mbone webcast (500-1,200 Kbs) and 2) a single-stream low-bitrate Real Networks webcast (250 Kbs). A seminar "The Berkeley MIG Seminar and Internet Broadcasting" given in 1999 describes the history of the seminar and some of the technology and production changes used up to that time.

Changes to the Berkeley MIG Seminar

This week we are making two significant changes to the MIG Seminar production. First, we have installed a new computer network that supports multicast. The previous BMRC network coupled with the complex set of networking gear in Soda Hall had constant problems with multicast -- lack of connectivity, spurious packet/session drops, and the mysterious address failure errors (a working multicast address suddenly stops working for no explainable reason). We have installed a new BMRC network composed of Cisco Catalyst 350x switches with CGMP, which in early tests has worked exceptionally well.

The second major change is the introduction of a new high-speed production TV quality transmission. Last year at NAB we discovered a low-cost video capture card with a JPEG codec chip for Linux. The LML33 board is sold by Linux Media Labs for $400. The board is essentially the reference design for the Zoran JPEG and PCI bus interface chips. Over the past year we have developed new Mbone tools, collectively called RTPtv for the capture, transmission, and display of production-quality TV streams. The video tools can produce and display full-sized images (720x480, 4:2:2), called D1 video, at 60 fields/second. The audio tools can produce stereo or mono L16 audio at sampling rates up to 48 KHz. All tools use RTP encapsulation and support numerous heuristics for error concealment and recovery. Moreover, the audio and video streams are synchronized. We have tested the synchronization by playing regular television signals for over 24 hours both locally and across Internet2.

More information on the LML board and the RTPtv applications is available on the RTPtv web page.

In addition to the current transmissions, we are adding a third transmission to the seminar using RTPtv. The new transmission will have two streams:

As usual, the speaker stream will include views of the room, the speaker, and the audience asking questions. The content stream will primarily be used to show the speaker's presentation material which is typically scan-converted PowerPoint slides. On occasion the content stream is switched to show other presentation content (e.g., video tapes, overhead transparencies captured with a document camera, etc.). In those cases the content stream bitrate will be increased to capture the motion.

Some people may say "10+ Mbs is too much for this service, why not use MPEG2 or some other high-quality codec?" The answer is cost and software. We have not found a good MPEG2 encoder/decoder board for less than $2,500, with most boards or systems costing anywhere from $5,000 to $25,000. And, none of these boards come with Linux drivers. We find Windows servers to be less reliable and harder to manage remotely in our broadcast infrastructure. For now, Internet2 has plenty of bandwidth for this type of experimentation and a Linux PC with two LML boards costs less than $2K.

The RTPtv Berkeley MIG Seminar transmission will be restricted to Internet2 and will be sent using IP Multicast. It will be advertised by a SAP announcement named "UC Berkeley MIG Seminar (RTPtv)". Your SAP application will probably know how to launch vic and vat, but the RTPtv applications will need to be launched manually (see below for details).

Watching the RTPtv Berkeley MIG Seminar

You can watch this seminar using one or two LML boards or using software-only decoding. The LML board has both an overlay feature which will show the decoded images in a window on your desktop and a composite video-out feature (RCA-jack composite or svhs s-video) which will show the image on a TV. Note that up to four LML boards can be used in one Linux PC. The RTPtv software includes the following applications:

More details on installing the software and configuring Linux are available on the RTPtv web pages.

The RTPtv video streams use the 1998 RTP payload format for JPEG-compressed video and a variety of features not supported by most Mbone tools including vic (e.g., restart markers, type-specific field support for interlaced images, etc.). The most recent release of Mash has support for all of these features so you will need to download and install this software if you want to watch one or two video streams using softare-only decoding. Quicktime 5 can also play an RTPtv video stream but only if certain RTP features used by the RTPtv video server are disabled -- at the present time we do not plan to disable these features for the upcoming seminar broadcast. The Mash software is designed to line double the first field of each frame as it can only decode 25-30 fields per second on a 600 MHz Pentium 3 processor. We are interested to hear what frame rate you can achieve with the new 1.5+ GHz Pentium 4 CPUs. We have experimented with dual-stream software decoding which gives 18-20 line-doubled fields per second on the speaker stream and 2 line-doubled fields per second on the content stream. At this rate the speaker stream is still jerky on a 600 MHz Pentium 3.

Audio is more problematic. If you have a Linux box, you can use aclient to decode the audio stream. If your Linux box has the 0.5.x series of the ALSA audio driver and API library installed, then you will get synchronized audio and video when vclient is used for the speaker stream. If ALSA is not installed, then you will still receive audio but the synchronization may degrade over time (depending on what sound card you have).

Even if you do not have an LML33 card or Linux, you can still use aclient to receive the better quality audio stream if you have a UNIX system that supports the OSS audio API (such as FreeBSD). If your system can not use the aclient application, then you will have to use the standard vat stream for audio.

The following information should help configure the environment:

Two LML Boards
A vclient process can be started by typing "vclient -M 233.0.25.241 -P 51000". A Tcl/Tk GUI interface exists, but is still undergoing development -- see the RTPtv page for more details about the GUI. The vclient will tune to the first server that it receives data from on the MIG video multicast session. To change to a different video source in that session, press the 'a' key and then press enter. If you have multiple LML33 cards in a single machine, a separate vclient process will need to be started for each card (in this situation you also need to specify "-1", "-2", "-3", or "-4" on the command line to tell each instance of vclient which card it should use).
For audio, type "aclient -M 233.0.25.201 -P 21000". If you do not hear any audio, then check the audio mixer levels on your machine to verify that they aren't muted -- you will need to turn up the volume for the PCM source, as well as the speaker output.
One LML Board
The process for one LML card is basically the same as the two card case: type "vclient -M 233.0.25.241 -P 51000". Depending on your interests, you can use the switching feature (press the 'a' key then the enter key) to tune to the speaker stream (which is only necessary if your vclient selects the slide stream), but you can also use this feature to switch back-and-forth between the two video sources (i.e., you get to play "television director" for your local machine!).
If you only have one board then you probably want to use vclient to receive the speaker stream and use Mash to receive the slide stream. For information on how to use Mash, refer to the Software Decoding section below (don't forget to "mute" the speaker source in vic!).
Software Decoding
To use Mash to watch the seminar, visit the Downloads page on the Open Mash website, download Mash 5.1.4 for your platform, and follow the instructions for installing Mash. You can then watch the RTPtv seminar the same way you'd watch any of our previous seminars. Just launch nsdr, look for the announcement titled "UC Berkeley MIG Seminar (RTPtv)", and run vic and vat as usual. To view the full width of the transmission, choose one of the NTSC 601 sizes from the Size menu. We recommend using a 500 MHz Pentium 3 or better.

Berkeley MIG Seminar, Wednesday May 2, 2001

This week's seminar speaker is Paul Haskell from Harmonic Incorporated who will be speaking about "Digital Video Bitstream Manipulation." Paul is a graduate of EECS at Berkeley who has been working in the MPEG coding field for many years. We are very excited that his talk will be the first webcast with this exciting new technology. An abstract for the talk, copies of the slides, and a link to the live chat room for questions and feedback are available on the seminar announcement page.


Please send comments to: webmaster@bmrc.berkeley.edu
Last updated May 1, 2001.