Jay Ankeney /
Long GOP Editing
This summer, we're all going to start dealing with long GOP issues. Not the Grand Old Party's endless convention in New York, but the increasingly rapid adoption by production facilities of long "Group Of Pictures" (GOP) MPEG-2 transport streams, with which editors inevitably will have to wrestle.
So kick back with a cold one and journey with me under the guidance of the good folks at Pinnacle Systems into the mystical world of low bit-rate editing. After all, with everyone at NAB2004 talking about long GOP MPEG-2 editing (especially as it relates to posting the new HDV format), Pinnacle was the only major edit system manufacturer demonstrating its own practical native long GOP edit systems on the floor.
Not long ago, many wise wags would sternly proclaim that you could not edit MPEG-2 with frame accuracy. That's because MPEG-2 streams are comprised of a sequence of I ("Intra" or sometimes "Index") frames that anchor the beginning of the sequence called a "Group of Pictures," and then P ("Prediction") frames and B ("Bi-directional") frames within that sequence.
If your cut point falls on an I-frame, no problem; if it falls somewhere in between, the rules are different because some of those other frames, the P and B frames, don't exist independently outside of the syntax on the GOP. So how do you cut that string with the level of precision editors demand?
Remember celluloid film? Its sprocket holes outlined a physical frame line on which you could cut and splice invisibly. Uncompressed baseband video let you splice the signal on the vertical interval that defined the beginning of a field. But unless you transcoded MPEG-2 into an I-frame-only format, which sacrificed its compression efficiency, editing the long GOP MPEG-2 stream was like cutting up a cloud and expecting sharp edges-impossible.
Recently, however, CNN announced that it was redesigning its infrastructure for the "Feeds & Edit" project to take advantage of the low bandwidth requirements of long GOP MPEG-2 streams by installing Pinnacle's Vortex news servers. This should reduce its network bandwidth requirements down to 19 Mbps and diminish storage demands by about 30 percent, requiring only 8.7 GB/hr. versus 14.4 GB/hr. for the same video in DV25.
"Long GOP MPEG-2 is one of the foundations of the technology plan at CNN," said Gordon Castle, senior vice president of Technology at CNN in a press release. "We are leveraging the advantages of long GOP MPEG-2 in all systems."
FRAME BY FRAME
So how do we cut it? Pinnacle's solution is to add native long GOP editing to its whole Liquid line of NLE's by this fall.
"Being able to edit long GOP was one of the key criteria why CNN went with us," said Patrick McLean, product manager for editing at Pinnacle Systems. "We've earned a lot of experience handling MPEG-2, and even had a long GOP hardware card called the DC-1000 back in the late '90s that worked with someone else's editing software. But that was before Pinnacle acquired FAST Multimedia in 2001 and transformed their edit systems into Pinnacle's own Liquid line."
McLean said that Pinnacle's goal is to make long GOP editing, including HDV transferred over FireWire, as transparent as DV editing has widely become. Yet while some new HDV plug-ins for other companies' editing softwares transcode the MPEG-2 stream into their own proprietary file formats, Pinnacle is going to be cutting MPEG-2 in its native format to maintain the efficiency of its low bit-rate data stream.
First, in order to get to any given frame, the decoder may have to process several frames in front of it to generate that specific frame's image. Pinnacle has raised the performance capabilities of its new Liquid software to be able to decode up to five layers of video, or five adjacent frames on one layer, in real time. That way it can make something "real" out of the "unreal" properties of the GOP, even in high definition with HDV, without making the system feel sluggish.
Once a sequence is edited on the timeline and you want to export it or print back to tape in baseband video, Pinnacle calls upon two software decoders running in parallel. As soon as the output of the first decoder gets to a cut point, the second decoder takes over. It has looked at some of the previous frames back to the last I-frame to give it a reference and spits out video starting only with the selected frame within the GOP.
It gets a bit more complex if you want to output your project as an MPEG-2 program or transport stream. During capture, any standard MPEG-2 encoder actually records to disk just I-frames, difference frames and motion vectors. The MPEG-2 decoder only generates B and P frames upon output, and these have to be reconstructed into a sequence that follows the strict requirements of a valid MPEG file.
To accomplish that, Pinnacle uses a proprietary technology called "MPEG Splicing" to make one new flat file that follows the rules of MPEG. On either side of a given edit, the Pinnacle splicer rebuilds the necessary number of frames to generate new I, B, and P frames around the cut point, if they are necessary to produce a valid GOP.
"In effect, we are generating one MPEG-2 file for the entirety of the project," McLean said. "Only the first shot within a sequence needs to start with an I-frame. Other shots can be generated from I, P and B frames, depending upon the amount of difference or 'temporal redundancy' between them. It's a bit of black magic, but that's how MPEG-2 gains its bit stream efficiency by creating a transport stream that, unlike I-frame only editing, does not have to encode all of the temporal redundancy between frames."
MPEG-2 is the format the FCC has mandated for digital broadcasting. Now, coming into the production sphere in its native format, it's able to take advantages of the bandwidth savings from low bit-rate data streams. But there is more. There's a good chance MPEG-4 will become the standard for multimedia over the fixed and WiFi Web, and MPEG-7 may become the standard for description and search of audio and visual content. Work on the new standard MPEG-21 "Multimedia Framework" has already started and...well let's leave it there for now. Meanwhile, go pop another cold one and reflect on the time when we thought black-and-white television was complicated.