On broadcast and cable, we have two sources of high-definition material on 1080i channels. We have video-sourced content, such as sports, acquired natively in 1080i60. We also have content acquired in 1080p24 with a HD camera or telecined from film, but flagged as 1080i60 for broadcast.Video-sourced content
In the case of video-sourced material such as sports, we have 60 different 1920x540 fields, all acquired at different points in time (separated by 1/60th of a second). Because the fields were all acquired at different points in time, they don't "match up" when there is any movement on the screen. Attempting to combine every two fields to form a 1080p30 image would result in severe combing. Some older (and many cheaper) TVs simply "bob" to display 540p resolution, by taking each field and making it a frame without adding any new picture information. That was the old/cheap way of doing things.
The better displays on the market interpolate new information to create a full 1080p60 signal through a process known as motion-adaptive deinterlace. Adjacent fields are compared to determine what pixels are in motion. Areas of the picture that aren't in motion -- such as background scenery that did not move in the previous 1/60th of a second -- can be weaved together at full 1080p resolution. Areas of the picture that are in motion -- and did move in the previous 1/60th of a second-- are created by bobbing, or in some cases, averaging the information in adjacent fields, and will vary in resolution between 540p and ~1080p. Whether pixels in motion appear as 540p or closer to 1080p depends on the rate of movement, as well as the quality of the video processor in the display. Not all motion-adaptive deinterlacing is equal -- far from it.
Most newer 1080p displays do region-based
, motion-adaptive deinterlace for 1080i60 video sources. With this approach, the video processor divides the screen into lots of different regions or boxes, and then compares 1-2 adjacent fields (2-3 total) to see if anything has changed in those boxes in the last 1/60 of a second. If nothing has changed in a box (ex: part of a billboard), the processor "weaves" in the information from the previous field, providing full 1080p information for that box. If something has changed in last 1/60 of a second, the processors "bobs" to display 540p resolution for that box.Even among products with region-based, motion-adaptive deinterlace, there are probably significant differences, depending on the computional power of the processor. For example, one display might divide the screen into 64 regions and another might divide it into 256.
Contrast that to pixel-based
, motion-adaptive deinterlace solutions like Gennum VXP and Silicon Optix HQV. Rather than divide the screen up into large regions, those solutions compare every pixel across three or four adjacent fields (4-5 total) to determine what pixels are in motion and which are not. Weave and bob decisions are made individually for every pixel on the screen, rather than for larger regions of the screen at once. Since they compare more fields, these processors can also better determine the properties of the motion to apply motion compensation or a multi-directional diagonal filter, if appropriate. It should be fairly obvious why this more comprehensive -- and computionally intensive -- approach can yield superior results.Film-sourced and 1080p24 content
In the case of 1080p24
content -- such as movies and television series -- all this is interpolation is unnecessary. Few people realize that virtually all movies and series content shown on CBS, NBC, HBO, Starz, and Showtime is actually full 1080p, just like Blu-ray and HD-DVD. There is no need to interpolate anything, because the full information for all twenty-four 1080p frames is already there. With 1080p24 content delivered in a 1080i60 signal, you have the following:
Frame1, Field2Frame1, Field1
Frame3, Field2Frame3, Field1
Frame5, Field2Frame5, Field1
This is known as a 3/2 cadence. You have three fields of one frame, two fields of the next, and the cycle repeats.
The fields highlighted in bold are sent using repeat flags, a few bytes which tell the MPEG-2 decoder to repeat a previous field. Only 48 unique fields of information -- each containing half the information in a full 1080p frame -- is typically transmitted every second with 24p content. Compare that to 1080i video, such as sports, where 60 different fields of information is sent every second. For that reason, 1080p24 source content requires less bandwidth to broadcast than video, not more.
The only hardware that has access to those repeat flags is the MPEG decoder inside the STB/DVR or HD-DVD player. Once the MPEG (or VC-1) bitstream from broadcast, cable, or HD-DVD is decoded by the STB, DVR, or HD-DVD player, there are no more flags. All you have at that point is the cadence.
The display processor can't simply combine every two fields, because they don't match up. Instead, the display must reconstruct the original 24 1080p frames through a process known as inverse telecine. Inverse telecine produces an image that is identical to the original 1080p source. To do this, the display processor must determine the cadence of the input signal by comparing the fields. If every field is unique, then the source is video. If every fifth field is a duplicate, then the display processor knows that the source is 24p**
; it can eliminate the duplicate fields and reconstruct the original 24p frames. Once this is done, pull-down is applied to repeat the full 1920x1080p frames to match the refresh rate of the display (i.e. 60Hz). At that point, depending on your TV, the image is output directly to the screen, [or] digitally scaled to add overscan, or digitally scaled to fit a lower-resolution panel. On a display that correctly performs inverse telecine, there will be no difference between the 1080i and 1080p output from a Blu-ray player.
Most modern displays can detect the 3/2 cadence on 480p24 content flagged as 480i60 (i.e. DVD), but only a minority can do the same with 1080i60 signals. It is more computationally intensive to do this with high-definition, and many display makers skimp on high-def processing to cut costs. On displays that cannot detect the 3/2 cadence necessary to reconstruct the original 1080p frames, they treat the source as video. They either bob to display the signal as 540p -- as was the case on older/cheaper displays -- or they do motion-adaptive video deinterlace to interpolate the remaining information for the 1080p frame.
Progressive frames created from interlaced content through interpolation will never be as good as material originally acquired in 1080p and reconstructed with inverse telecine. The greatest differences are seen when there is a lot of movement on the screen, because all the information for that motion exists in a progressive source, but does not exist in an interlaced source.
If you've ever seen combing, blurring, moire, stairstepping, or other interlace artifacts on movies or series content shown on CBS, NBC, TNT, HBO, or SHowtime, chances are it was because your display could not correctly perform inverse-telecine. Unfortunately, most displays do not have quality deinterlace -- of the displays tested by Home Theater Magazine
(more results here
), only seven of the 61
tested would offer the same performance with a 1080i input as they do with a 1080p input. In its latest issue, Home Theater Magazine reviewed and compared
the top 60" 1080p RPTVs from JVC, Mitsubishi, Olevia, Samsung, Sony, and Toshiba. Only two of the six could correctly detect and display 1080p24 content delivered in a 1080i60 transmission, such as Heroes on NBC and CSI on CBS.
Things become a bit more complicated when you have content with "bad edits," as you often get when film and video sources are mixed together. Further, while the actual movie or series may be a 1080p24 source, it's common for commercials to be video sourced. Hence, the display's processing has to be able to switch between video and film modes on the fly, based on what it detects as the source content (60i or 24p). Not all video processors and displays are able to do this well. Some display processors can detect and switch between video and film mode relatively fast (within a few seconds), whereas others may take 30-60 seconds.** It's not actually quite this simple. Most film-sourced content on cable and broadcast is distributed with the appropriate repeat flags to minimize bandwidth use. But there are times when film-sourced content is distributed and compressed like video. Generally, broadcasters and cable companies like to avoid that, because it wastes bandwidth, but not every broadcast affiliate uses modern encoding equipment.
When content is distributed without those flags and compressed like video, the cadence is still 3/2, but due to compression, every fifth field may not be bit-for-bit identical to a previous field. Hence, to provide reliable inverse telecine, a display processor must not only detect identical repeated fields, but it must also detect when every fifth field is nearly identical. This analysis requires more computational power. Some implementations like the Silicon Optix ReonVX and Realta have the processing power necessary to do this, while other solutions, like those found in Pioneer plasmas, apparently do not.