Side-by-Side versus Top-and-Bottom 3D Formats
CableLabs, a standards and research body funded by the cable industry, recently released a specification for stereo-3D formatting for CATV systems. This spec is now issued, and publicly available:
Cablelabs has a couple of figures that visually explain these two alternatives:
It’s pretty obvious that these approaches are two ways to pack two left-right stereoscopic views into a single video frame. Both have the distinct advantage of utilizing the existing HD video transmission and distribution equipment and infrastructure. It’s also obvious that some of the original resolution in the Left and Right frames is sacrificed.There are some interesting implications in their specification, however:
The thinking going into this has several dimensions. These reasons have to do with the overall tradeoffs between horizontal, vertical, and in some cases, temporal resolution of the video formats being used.
A pattern emerges here — why are interlaced signal formats restricted to Side-by-Side? …and progressive formats restricted to Top-and-Bottom? This question and derivatives of it seem to come up frequently, and I’ve seen a variety of answers that seem to miss the real issues.
In the case of progressive video formats, the tradeoff is fairly simple to visualize and explain. When applying one of these 3D formats, you are choosing between sacrificing one half of the vertical resoluton (TaB) or one half of the horizontal resoluton (SbS) in order to squeeze twice the information (a left and right eye image) into a single frame.
Interlaced and Top-and-Bottom do not get along well
With interlaced formats, it’s a more complex question. First, remember that interlaced formats alternatively transmit the even lines, then the odd lines of every video frame, in an alternating sequence. But these sets of even lines, and odd lines — referred to as fields — do NOT join up to form a single frame at a single instant of time the way a progressive frame does. Each field is actually captured by a camera at twice the sampling rate — so moving objects actually register differences with every field. While this means that motion is sampled at twice the progressive rate (e.g. 60 hz instead of 30 hz), each field has half of the vertical resolution.
The loss of vertical resolution is “shared” between odd lines and even lines at alternate times — while doubling the apparent frame rate — the temporal resolution. In interlaced displays, the consequence can be smoother perceived motion, especially for sports and other fast-moving action.
With interlaced video formats, the choice of TaB formatting can have a catastrophic effect. Recall that TaB formatting will sacrifice one-half of the vertical resolution. With interlaced video, one has two choices, none particularly attractive: 1) sacrifice MORE vertical resolution 2) sacrifice temporal resolution.
In the first case, the encoding system takes each field (which is already at half vertical resolution), and throws away every other line (again!).
In the second case, the encoding system throws away every other field and packs it into one or the other of the top or bottom views.
Neither of the two possible ways to carve up 1080i into TaB are attractive — the result is that 1920x1080i at 59.94 fields per second with TaB becomes 1920×260 @ 59.94 effective resolution per eye in the first case. This is an abysmally low amount of vertical resolution. TaB becomes 1920×540 @ 29.97 effective resolution per eye in the second case, which is throwing out half of the motion information. On the other hand, with SbS, 1920x1080i at 59.94 fields per second becomes 960×540 @ 59.94 effective resolution per eye. One does not have to sacrifice temporal resolution (motion is preserved), and is a good tradeoff of spatial resolution.
Why not keep it simple and use Side-by-Side for all formats then?
Let’s consider the 720p60 format. This is 1280×720 with a 60hz frame refresh rate. With SbS, the effective resolution becomes 640×720 @ 60 per eye. This horizontal resolution is actually closer to VHS than DVD, let alone HD, and becomes quite noticeable.
With TaB, 720p60 becomes 1280×360 @ 60 hz per eye. This vertical resolution puts this square into the precise definition of VHS again, not HD… so why do this at all?
This would seem to be especially perplexing for 720P — why not just use SbS for 720P and TaB for 1080P?
Even more perplexing: in spite of CableLabs’ document, Comcast and several other MSOs broadcast the Masters golf torunament in 1080i SbS format, which was produced by ESPN — who shot the event itself in 720p60.
In my next post, I will cover the SbS format in detail — this is also not as simple as it would seem at first blush.