DTV

3:2 Pulldown:
The Importance of System Test

By Frederick Gotwald
Sun Microsystems
Chapter 36 Secretary/Treasurer

The discussion of 3:2 pulldown at the June meeting brought back memories for me. In 1991, I was at General Instrument (GI) and was the lead engineer for the video section of DigiCipher I. The video section consisted of four printed circuit boards, the one which dealt with 3:2 pulldown was called the Digital Frame Store (DFS).

3:2 pulldown from the Quantel Digital Fact Book. Courtesy Quantel Corporation.

The original designer of the DFS left GI before the board design was complete, so Marti Johnson finished the design. The DFS consisted mainly of 2 parts: The digital video anti-alising filters and a big circular memory buffer (or frame synchronizer). As was mentioned in the meeting, 3:2 pulldown inserts duplicate fields in the original film frame sequence. If you can detect the sequence, you don't have to send the duplicate fields through the system; you just have to send a synchronization bit to the decoder telling it when to duplicate the appropriate field.

When it came time to test the detection of the 3:2 sequence, Marti built video sequences into a test pattern generator. The sequences checked that the detection circuit would sense the start and end of the 3:2 sequences. We anticipated that broadcasters would interrupt the film sequences to insert video commercials (to the chagrin of many directors). Anyway, after some small tweaks, the detection circuit worked well.

The video source for much of our integration was industrial laser disk players. We had them on repeat mode and they literally ran 24 hours a day for many years. When all the pieces of the DigiCipher system came together in the system test area, someone noticed occasional jumps in the decoder's video output. They found a repeatable scene in "Back to the Future" where the jump would occur (at the beginning of the movie when Michael J. Fox is on a skateboard and grabs the back of a pickup exiting a drive-through).

The problem was due to the DFS 3:2 detection exiting on a boundary where the decoder did not have a valid frame. OOPS. The problem was fixed by inserting more memory on the DFS to drive the 3:2 exit given the constraints of the decoder memory.

It was interesting to hear that Snell & Wilcox is still seeing encoders which have this problem. If you're going to buy a MPEG2 encoder, you'd be well advised to buy one from a company which is on their second or third generation product. Not only because of 3:2, but also because of other algorithms such as scene change/fade detection and statistical multiplexing.

Previous Story |  Back to Top |  Home Page |  Next Story
Archives |  Contacts |  Jobs |  Calendar |  Subscriptions 
Members |  Links |  Sponsors |  EAS |  Search |  Home 

Web site contents ©1999 Society of Broadcast Engineers Chapter 36 San Diego. For more information, to become a member or a sponsor, or to make suggestions or comments, e-mail sbe36@broadcast.net. Write to P.O. Box 710702, San Diego, California 92171-0702.  

Edited by Gary Stigall. Posted 29-Jun-99.