By now, most readers are aware of the various DTV formats being broadcast let alone the number of possible formats in the famous ATSC Table 3. When the networks made their announcements regarding the DTV formats they would be adopting, the post-production industry was stunned. It would be unrealistic for a post-production services company to build facilities for each possible format. In addition to this new dilemma, there has been an ongoing effort since the mid-80s to electronically generate high-quality foreign distribution masters. The basic economic model of episodic television tells us that if a particular series expects to come out ahead financially, it must not only have successful foreign sales, but be marketable in domestic syndication as well. For the owners of a particular a show or series, consideration must be given to the format requirements for the first run network delivery, a 25 frame per second version for international distribution and an unknown format for future syndication.
LaserPacific Media Corp., a post-production facility in Hollywood, CA, quickly concluded that because nearly 99 percent of episodic television shows are shot on film, and film runs at 24 frames per second (24fps), post production should be done at 24fps. The post process should also use the highest spatial resolution from Table 3 to assure that only cross or downconversions are needed to create the final products.
In 1998, LaserPacific embarked on a mission to put together a 1920x1080, 24P post-production system. It partnered with Sony, Philips, NVision, Evertz, Digital Vision, daVinci and others to fast track a system that would be ready for the 1999-2000 season.
Review of the post process In episodic production, as well as long-form "made for television" movies, picture and sound are captured by a dual system. The film camera is capturing picture at 24fps while an audio recording device captures the sound on typically a tape format along with a sync reference. Today the audio recorder is either a Nagra 1/4" with center track timecode or a digital audio tape device (DAT, DA-88 etc.). The timecode recorded on the audio device normally runs at 30fps.
In post, the film is put on a telecine and run at 23.976fps. By adding a 2:3 pulldown sequence (see side bar), the temporal rate is converted to 59.94 fields per second allowing recording on a 29.97fps recorder. The production sound element is "resolved" to 29.97fps so that its speed is reduced by the same .1 percent as the film. Picture and track are synchronized with a device similar to an edit controller and recorded to a telecine master tape.
The offline editing system (nonlinear and mostly Avid-type systems) are very flexible devices that allow editors to work with a database of clips instead of the timecode numbers associated with the telecine masters. The database relates picture and sound to other information such as scene and take numbers, film keycode numbers, camera roll numbers, production sound roll numbers, shoot dates, etc. The database must track the telecine-master timecode as well as the original production sound timecode. The burden for generating these databases has been put on the post production facility. Over the years, systems have been developed by LaserPacific and others to automate some of these logging tasks. Data is collected as a snapshot of a single point in time for each take transferred. Real-time data, such as the film code and timecodes are logged automatically with the use of electronic readers that send data to a central computer. Non-real-time data is entered by the telecine operator. The offline systems also have the capability of capturing or digitizing only the original film frames running at 23.976fps embedded in the 29.97fps video. This requires the database to have knowledge of the timecode-to-pulldown relationship of the video. Normally, the database snapshots or records are taken at a telecine-master timecode number associated to an "A" frame. Once completed the database provides a map to the picture, sound and timecode of the telecine master element. Since some of the database systems are proprietary, a common interchange ASCII file format called Flex was developed.
Once the editor completed the final cut, two edit decision lists (EDLs) are exported from the offline system's database. One represents the video and audio from the telecine master source tapes and the other is audio or track only and points back to the original production tapes. The first EDL is used for the online assembly process where the high quality masters are used to re-conform the show. The second EDL goes to the post sound dialog editor who will assemble the track from the production tapes, then clean up and smooth out the track removing background noises, replacing dialog, etc. Music and effects are added to create a final track that will be married back to the final video element later.
Once the online edit is complete the "Edited Master" is given a final timing (film term for color correction) to visually smooth the transitions of the edit points and create the final "look" of the show.
Once the "Color Timed Master" is created the final audio track from post sound can be "laid back" replacing the scratch track of the master. The show can now be titled and final network "air" tapes can be made.
1920x1080, 24p basics Film is transferred to tape at a one-to-one, frame for frame relationship along with production audio. No longer is the complexity of 2:3 a part of the telecine process. All editing, effects and titling are done in the 24p domain. The result is a product that is essentially the same as if the post process were done entirely on film. At this point 2:3 pulldown can be added to create a 60i or 60p version, at either the current or reduced spatial resolution. The product can also be run a 4 percent higher temporal rate to create a 25-frame (or 50i) version.
The dirty details From the previous paragraph, it sounds like 24p post production is a "no brainer." Unfortunately, we must coexist with a 30-frame world and this coexistence can add tremendous complexity to the process. The majority of the offline edit systems in use today are 30-frame based. In most cases, either a 3/4-inch or Beta-cam SP tape are used to "input" or "digitize" the material into the system. Most post-production sound facilities also use systems that are based on 30-frame timecode and thus the production sound is recorded with 30fps code.
Much of the complexity happens during the telecine process. Remember, when the film is transferred it is synchronized with production sound and recorded onto the telecine master while simultaneously generating a database for the offline system. In this case we have production audio timecode running at 29.97fps while the telecine master is recording at 23.976fps, which is the same speed the film is running at. In order to guarantee frame accurate timecode capture by the logging system, the sync word of both timecodes must be in phase at the capture point. To achieve this two pieces of equipment need to be modified from traditional methods of operation. First the facility's master sync system needed the ability to lock the house 525/59.94Hz reference with a 1080/23.976Hz (actually 47.952Hz is used). These two rates have a coincidence every 10 fields or six times per second. LaserPacific went to NVision to modify their existing sync generator to genlock to a 525/59.94Hz master reference and output 525/59.94Hz, 1080/47.452Hz, 625/47.952Hz and a 6Hz pulse that is the vertical coincidence point of the three references. daVinci modified their TLC telecine controller system to use 6Hz coincidence point to start two time lines, one for the 29.97fps machines and one for the 23.976fps ones. This made it possible to guarantee proper timecode sync word phase at the capture point. LaserPacific helped develop and uses an Evertz 9025 encoder and "Tracker" logging system. This system also uses the 6Hz relationship to qualify the data captures.
Another hurdle is that offline edit systems are still based on 30fps (29.97fps) standard definition. This requires the telecine master to be converted to normal 525/29.97fps video by adding 2:3 pulldown and downconverting. Further, a convention for mapping between 24-frame and 30-frame timecode is needed. A Flex file is exported from the telecine logging system to be sent along with the 3/4-inch tape to the offline editor. In order to generate this file, the system must convert the 24-frame timecode on the telecine master to 30-frame and know what the pulldown will be when the downconversion is made. To achieve this it was decided early on that pulldown would be mapped to timecode and that except for the final Network delivery tape, all interim 30-frame timecode would be non-drop. This allows a simple convention of 24-frame to 30-frame timecode mapping as shown in Figure 1. The timecode numbers are identical at each whole second. Notice the modulo 4 to modulo 5 relationship between seconds that correspond to "A" frames in the pulldown sequence. The daVinci TLC and Evertz logging system edit and log only on valid, pseudo "A" frame timecode numbers (frames 00, 04, 08, 12, 16, 20). When the Flex file is exported, the mod 4 numbers are converted to their mod 5 equivalents (00, 05, 10, 15, 20, 25). Remember, because of the way the synchronizer works, these pseudo "A" frame timecode numbers are also points where the 30-frame audio numbers will be accurate.
>From the offline editor's point of view, nothing changes when >working on a 24p show. The 30-frame tape is digitized into the >system. Using the Flex database, only the original 24 frames are >captured. When editing is complete, the same two EDLs are exported >from the system. The post-sound facility continues to work as >before. The big difference comes during the online conform session. >The 30-frame list must be converted back to 24-frame, but the simple >timecode mapping convention used in telecine cannot be used. Since a >24fps frame is longer in time compared to a 1/30fps frame, simple >timecode mapping would cause run-time errors to accumulate at each >edit points. The 24-frame version would slowly become longer causing >sound sync problems. LaserPacific and others have written conversion >programs that keep track of these time deltas and trim edit points >by a frame every so often to keep picture in sync with sound. Once >the EDL is converted, the online assemble proceeds as normal.
Once the online is complete the tape to tape timing (color correction) is performed and a 24p color timed master is made ready for the track to be laid back. Once again, a daVinci TLC is used to synchronize the 24fps picture element with the 30fps sound element. Titles are added and the final master is ready for the Network delivery tape to be made.
The magic tape machine For the 24p system to be successful, a tape machine was needed that was versatile, flexible and easy to use. Engineering teams from LaserPacific and Sony's factory in Atsugi, spent eight months defining and testing the HDW-F500 HDCAM machine. To be versatile the machine needed to record and playback multiple temporal rates including 23.976p, 24p, 25p, 29.97p, 30p, 50i, 59.94i and 60i. The machine further needed to crossplay between certain rates. When cross playing between two rates, audio is resampled to maintain 48kHz on output and in some cases, timecode is remapped. The machine was also specified to add 2:3 pulldown when in either the 23.98p of 24p mode allowing not only 1080/59.94i outputs but downconverted 601 outputs as well. These outputs have embedded audio and timecode that have been delayed to compensate for the processing delays of the 2:3 pulldown adder. When making final network delivery dubs from a 24p master it is also possible to convert the code to drop frame, specifying a sync point number where the native code and output code match. This would normally be an even hour. Once specified the output code is gear-toothed and increments/decrements at the appropriate rate.
The utility of 24p 24p post production is an elegant solution that creates a universal master. The master can be played at 25fps for foreign delivery and either cross- or downconverted for first run and future syndication. It does require some discipline to achieve the highest quality product. One of the major disciplines is treating the production as if it were being done entirely on film. That is to say, all elements must have a 24fps temporal rate. At this time there in no easy way to convert material with 29.97fps temporal rates (or 59.92 field rate temporal rates) to 23.98fps without artifacts. There are companies working on products with sophisticated motion estimation and blurring processing that may provide solutions in the future.
There are still many computer graphic effects houses that do not have 24p HD tape machines requiring systems like Sierra Design Lab's HD DDR to record the 24p material andthen transfer to a standard computer environment. This allows standard graphics files to be delivered with the impediment of being much slower than real time. The problem here is that at 4.5MB to 6MB/frame (depending on whether YUV/4:2:2 or RGB files are created) a large amount of storage is needed.
One of the big debates is over aspect ratio compatibility. As long as we have to live in both the 4:3 and 16:9 worlds, decisions must be made as to how to create one master that "feels right" in both formats. The most common method is to frame to 4:3 while protecting the entire 16:9 area- no grips in the picture or edge of set. Another method is to frame for 4:3. Then when the transfer to high definition is made the image is vertically compressed so that the 4:3 image fills the 16:9 frame. During the editorial process the actors look short and fat but when the downconversion is made for standard-definition delivery, the original 4:3 is restored. The 16:9 delivery is then expanded vertically with a common top, if there are any small dogs a foot level the image can be "tilt and scanned" in a DVE device.
Facility design is also more complex as not only are there multiple and different outputs from machines but due to the various temporal rates the machines operate at, more reference signals are needed. As many as eight references may be needed to be sent to a tape machine or edit room. At LaserPacific, we elected to implement a reference router to cut down on the volumes of cables needed. There is work within SMPTE to define the next generation sync system, one that could have essence of all appropriate references on one cable.
Routing systems become more complex as now one machine can have, the primary HD SDI output, a second HD SDI output with pulldown, a third with standard-definition 601 and even a monitoring NTSC-encoded output. Control systems must be intelligent to route signals to appropriate destinations.
Caveats aside, producing a project in 1920x1080, 24p is certainly in the best interest of the content owner as it adds value in the form of flexibility in delivery. Even recording back to film will yield excellent quality. Although the technology is barely a year old, the pieces needed to build a 24p facility are available and more solutions are arriving all the time.
2:3 pulldown, or as it is more com-monly known, "3:2" pulldown, indicates a pattern by which 24 progressive (film) frames per second are mapped into 60 fields per second. There are four progressive frames to every 10 fields. This sequence is broken into four frame-to-field groupings normally labeled "A," "B," "C" and "D." At one time in the early 1980s the "A" frame was a three-field sequence starting with Field 1 followed by a two-field sequence starting with field 2. Hence the name 3:2 pulldown. Today, an "A" frame is described as a two-field sequence beginning on Field 1. Figure 2 shows the pulldown sequence for a one-second interval. Notice that for every source frame there is either a two- or three-field sequence, starting on either a field 1 or field 2. The "A," "B," "C" and "D" frames indicate where the original progressive frames fall in the 60-field sequence. There are two places in the sequence where a video frame will have information from two progressive film frames. One is the "B to C" transition and the other is the "C to D." It should also be noted that it is custom and practice to map timecode numbers to pulldown frames. An "A" frame is normally mapped to a frame 00 time-code with a modulo 5 sequence for 30 frame code and modulo 4 sequence for 24. Figure 2 shows the relationship of the 24 and 30 frame codes as they relate to the pulldown sequence.
In the case of a 60p-frame system, Figure 3 shows a similar relationship.