Working on an Avid Media Composer Adrenaline at online resolutions most of the time, I have been curious about the various resolutions and compressions available to me. I have not been able to find any satisfactory comparisons focusing on the visual quality of the compression techniques.

So, I have conducted some research of my own, I think it provides a reasonable overview of the quality of the resolutions. This was all done with PAL footage, but I assume that roughly the same results will be had with NTSC.

Technique

I captured a small clip from a Digital Betacam tape at all the resolutions I wished to test, and then exported a representative still from Avid using an uncompressed TGA export setting.

Opening each file in Photoshop, I created a single file that contained each still on a layer on it’s own, and applied the ‘Difference‘ blending mode. Each layer was differenced against the 1:1 reference capture to issolate places where the recorded image varied. I then applied a adjustment layer than cropped the levels to between 0 and 32 (or 1/8 of the image’s total range) to highlight the variances.

To get some sort of accurate and consistant measure of the level of variance, I used Photoshop’s histogram tool to give me statistics on the resulting difference image (without the levels applied) - this gives us two useful measures, Standard Deviation telling us how far the lumanance levels in the diff vary, and the Mean, which demonstrates the average lumananance in the picture. As I figure it, the lower the number, the less the picture has varied from the 1:1 reference image.

	Frame Segment
2:1	3:1	10:1
MPEG50	MPEG40	MPEG30
	DV25 4:2:0

Results

(These Meridian bitrates are estimates)

Based on these results, I’d rate these compressions in the following order:

Notes

Obviously DV25 suffers in this case due to it’s 4:2:0 sampling method, although it does general produce what I think is a better picture than 10:1 in most cases. MPEG IMX is a 4:2:2 I-frame MPEG2 codec. I am not clear what sampling method is employed by the Meridian codec.

The footage I have used in this case represents what I feel is a reasonable representation of an average frame of video. It has a wide dynamic range, a good range of colours, skin tones and moderately complex fine detail in some areas of the picture. I suspect other images will yeild different results - complex tree patterns are a prime example.

The sample footage was shot on a Sony DVW-790P, onto Sony stock, played in a DVW500AP, captured via SDI into an Avid Media Composer Adrenaline.

An increasingly frustrating problem for online editors is a fairly recent tedancy for Avid to experience a type of sync drift while digitising whole tapes at an offline stage.

This is not apparent at the offline stage, as the captured footage appears to have continuoius timecode, and in sync audio, however at online stage many clips can drift out of sync with the provided audio by as much as 18 frames (or potentially even more).

The problem is two-fold…

Firstly, it is a camera problem. A number of cameras (only Sony models it seems) appear to skip a frame or two when continuing timecode after powering off or entering standby mode. They do not miss any actual frames but the timecode of the first frame after this power cycle does not immedidately follow the timecode of the last frame recorded, instead it is one or two frames further on.

The second part of the problem is to do with the way Avid records timecode while capturing footage. Rather than recording exact timecode from the deck, it gets in sync with the deck when the capture starts and from then on simply counts frames. This isn’t normally an issue, as any discontinuities in timecode will be detected as a timecode break, and Avid will halt the capture. However, and this is a big however, Avid does not detect timecode breaks of less than six frames, so in when digitising tapes with the above problem, the timecode Avid records will get more out of sync every time it passes one of these break points.

While the audio and pictures are tied together as one stream in the offline, this is not a problem, however when it comes time to reconform the edit on another suite, problems arise. Because the batch capture process generally cues to within a few second of the required material, it syncs with the tape timecode at that point, and is again correctly in sync with the actual recorded timecode. However, this is not the timecode that has been recorded as part of the edited sequence, nor is it the timecode that the original audio was captured with.

To correct this problem, it is necessary to resync each shot in the new online sequence to the recorded sound, to counter the original inaccurate timecode recorded at offline time. There are a number of ways to achieve this in a non-linear environment, some are not even that difficult. However if the show is to be onlined on a linear suite, there is no way to correct the problem. It will introduce problems with audio sync, and often created flash-frames where shots have slipped enough to pass the end of a take.

There are a number of options available at the time of the offline, or even during the shoot, to help reduce this problem.

• Digitise tapes in smaller chunks
  If offline tapes are digitised in segments of five or ten minutes, these small breaks to not continue to accumulate to create large sync discrepencies toward the end of tape, as they will if a tape is digitised as one single clip.
• Shoot ‘Time of Day’
  If shooting ‘Time of Day’ timecode it is impossible to create breaks of less than six frames between takes. However this can introduce other problems if tapes are used over multiple days or past midnight (00 comes before 23, decks have trouble figuring out how to cue). Another option might be ‘Free Run’ - However, in this case timecode breaks WILL be created, and care must be taken to allow for pre-roll at the beginning of all takes.