Sunday, January 18, 2009

Troubleshooting a bug with a WDK customization designed for creating S4M cuesheets

Every piece of TV content has an associated cuesheet indicating at what times different segments, such as commercials, voiceovers or show segments should be broadcast. These cuesheets would be fed into a broadcast trafficking system so the broadcast server would know what content to play at what timecodes.

In my current television broadcasting company, EMC Documentum is used to store the cuesheet information for an episode version by representing episodes as objects in the Documentum repository, and attaching the cuesheet data as attributes of the objects. I wrote a WDK (Documentum’s Web Development Kit) customization that displayed the cuesheet information as a panel available upon clicking the “i” icon of an object. Once the “i” icon was clicked, the customization would do some basic business logic checking to determine if a cuesheet for this episode should be allowed into S4M (some regions were not on Documentum yet so could not take advantage of our cuesheet customizations), and set the value of a true/false flag in the java class behind the JSP page. Then the user would be allowed to make changes to the cuesheet information. Afterwards, when the user wanted to save their changes, they could click on the OK button. This would trigger the onCommitChanges() method, which would check the value of the flag first before continuing. If the flag was set to not print the cuesheet, then the cuesheet file would not have been written to the monitored directory for S4M to import.

Over a weekend, this change was released to production, and seemed to work fine until suddenly cuesheets stopped being written to the monitored directory. After some careful examination of the code, I realized the flag was not being reset upon each view of the cuesheet WDK page. I then set the flag to reset upon every view of the cuesheet page, which resolved the issue.
The lesson learnt from this is that values set to members of the WDK code-behind class would be remembered throughout the life of the WDK session. Therefore class members should be set back to their default values upon every view of the WDK component, by adding initialization calls to the initControls() method.

Thursday, January 15, 2009

S4M bug with reading cuesheets with end credits

TV Broadcasters divide their broadcast time into segments indicated on cuesheets. These cuesheets contain information detailing what content to display at specific times. Each of these information bits contain the type of event to broadcast (show segment or commercial), as well as the timecodes when they start and end. The timecodes are a little different than regular timestamps, as in addition to hours, minutes and seconds, they contain the number of frames, as a second can be divided into 30 frames, and a frame is the smallest unit of time possible for TV broadcasting.

Each broadcast trafficking system has their own cuesheet format; S4M uses XML. However, the information contained in these cuesheets is generally the same, regardless of the type of trafficking system used. Usually this is unimportant as most television broadcasters only use one trafficking system, but what happens when a television station buys another? There is the opportunity to inherit a new trafficking system, and thus the opportunity to share information between the systems to provide visibility to the executives.

So what is the best way to store cuesheet information so it can be imported by different broadcast trafficking systems and be visible to the executives? I recommend storing the data in an ECM (Enterprise Content Management) system and providing customizations to export it in different trafficking systems’ formats.

For my particular scenario, EMC Documentum was chosen because of its rich media features and completeness of ECM features (I’m sure that my being a big fan of Documentum had nothing to do with it.) What we decided was to create a taxonomy in the repository that organized television shows by seasons and episode versions, and store the cuesheet information as metadata of the episode version objects. I’ll blog more on the taxonomy structure in the future; for now, I would like to focus on the cuesheets.

In the S4M cuesheets are 2 node types: SEC and EEC. These timecodes are computed values from the VO Squeeze TC in attribute of the episode version, and they represent the end credits of a show. On Jan 15, 2009 we discovered a bug in S4M that would schedule programs with SEC and EEC timecodes incorrectly. It was placing the end credits as separate segments on the playlist when it should in fact be one segment.

Of course an urgent fix was requested from S4M, but what happens in the meantime? Master Control cannot handle the end credits properly so we needed a workaround immediately before we could generate new playlists. What we decided was to not print out the SEC and EEC timecodes until the fix was provided.

The implementation of the fix was relatively easy. I already made a WDK (Documentum's Web Development Kit) customization in Documentum to print the cuesheets, so I just added a configuration value to the component’s XML configuration file to determine whether to print SEC and EEC timecodes. If the value was present, the timecodes would be printed. Anything else would default to hiding the timecodes.

Therefore in production we currently have the value set to not print the end credits for S4M cuesheets. Once S4M releases a fix, I will release a new configuration file with the value set to print the end credits. Configuration files are insignificant work so the turnaround time will be under 30 minutes.

Saturday, January 10, 2009

S4M - Anchor Ends

Here is an interesting facet of the S4M system in use by certain broadcasters I would like to detail below.

S4M can write playlists in a format known as H-class, which is a XML format listing events as “dlevent” XML nodes. Each event node contains a start time which is stored as the value of the “dlstart” child node. In addition, the total duration of which the event runs is stored as the “dlduration_hmsf” child node.

There are 2 main classes of events: primary and secondary. Primary usually refers to a program or commercial segment. Secondary events run in parallel to primary events, and can be voiceovers, snipes or commercial secondary events.

Anchor End
When writing H-class playlists, S4M sometimes writes secondary events that have a “dlstart” time that occurs after the primary event it is supposed to run within ends. This is a glitch with the S4M trafficking system but what this means is that the secondary event runs near the end of the primary event. The exact actual time the secondary event runs at can be determined by taking the difference of the primary event’s end time and the secondary event’s start time, and subtracting this offset from the primary event’s end time to determine the true “dlstart” time of this secondary event. To put this in the form of mathematical equations:

Primary event end time = primary event start time (dlstart) + primary event duration (dlduration_hmsf)

Now, determine if this secondary event is an anchor end or not:
If (Secondary event start time (dlstart) > Primary event end time) then
This is an anchor end.

Offset = Secondary event start time (dlstart) – Primary event end time
Secondary event offset time = Primary event end time – Offset (calculated above)