• Blogs
• Forums
• Tradeshows
• Partners
• White Papers
• Podcast Show Notes
• Events

HDV@Work

Feb 13, 1997 3:58 PM

Just What is 1080i? Part 2

In the last HDV@Work newsletter we learned that “1080i” can have horizontal pixel counts that range from 960 to 1920 — a whopping 2:1 range. However, the one constant was that there were always 1080 pixels vertically. Alas, both vertical resolution and horizontal resolution are not as high as these numbers suggest.

To convert the number of horizontal pixels to “horizontal lines of resolution” we must consider the Kell Factor. The Kell Factor was first measured by Ray Kell in 1934, who was working with an all electronic, 240-line, progressive video for RCA. About 36 percent of scanned lines were lost. Hence, the Kell Factor in 1934 was defined to be .64. (This is a number still used by many.) With more modern equipment, in 1940, Kell re-defined the value to be .85. With HD, the factor typically used today is between .90 and .95. (I employ values of .94 for horizontal Kell and .95 for vertical Kell.)

Converting horizontal pixels to effective lines of horizontal resolution is done by the equation: TVL/ph (TV lines per Picture Height) = (Horizontal Pixels x .94) ÷ 1.78. (The latter is the image aspect ratio.) Thus, when the pixel count is 1920, slightly more than 1000 TVL/ph lines will be available. With only 960 pixels — even with the use of horizontal green shift — only 550 TVL/ph lines will be available.

When a progressive-scanning camera uses three chips, resolution will be decreased by both Horizontal and Vertical Kell Factors. (The latter is .95.) Therefore, 1280x720 progressive video will typically have 680 TVL/ph and 680 TVL of horizontal and vertical resolution, respectively.

With 1080i, the effective vertical resolution — in TV lines — is the number of vertical pixels multiplied by the Interlace Coefficient. Our NTSC system, and the 1080i HD format, are based upon interlace scanning. When a CCD employs Dual-Line technology, only half the rows are read out. For every upper field, CCD driving logic selects the top-most row (Row #1). The first row scanned out is the sum of each element in CCD Row 1 added to the corresponding element in Row 2. The last row scanned out is the sum of each element in CCD Row 1079 added to the corresponding element in Row 1080. In this way, the CCDs output 540 rows (one field).

For every lower field, the CCD logic driving logic selects the second CCD row (Row #2) as the first row. The first row scanned out in the lower field is the sum of each element in CCD Row 2 added to the corresponding element in Row 3. The last row scanned-out is the sum of each element in CCD Row 1080 added to the corresponding element in Row 1081. Once again, the CCD outputs 540 rows.

A simple low-pass filter is created by adding two data samples. The process is called Row-Pair Summation. When a CCD adds row pairs, the filter helps prevent “interlace flicker.” Interlace flicker is the 30Hz on/off flashing of one-pixel thick horizontal lines.

As a hard horizontal edge moves vertically, flicker moves up or down the screen, causing “interline twitter.” Both flicker and twitter are reduced, but not eliminated when horizontal edges and lines are softened by the low-pass filter. (Line twitter and interlace flicker are the two most objectionable artifacts of interlace scanning.)

Two significant benefits arise from using Row-Pair Summation. First, CCDs are made 6dB more sensitive to available light. Second, noise is averaged and thereby canceled, increasing the signal-to-noise ratio.

Unfortunately, a price must be paid for the benefits of Row-Pair Summation. The filter decreases video’s vertical resolution. The amount of fine detail removed by the low-pass filter is called the “Interlace Coefficient.” The Coefficient is approximately .66, which means an interlace-scanned 1080-line CCD can output only about 715 lines (TVL) of effective vertical resolution. (Because of the filtered nature of interlace, the Vertical Kell Factor need not be employed.)

As we work through these laborious equations, the bottom line is that the real resolution of 1920x1080 touted by marketing is being incrementally reduced. Can resolution go even lower? Oh, yes.

When a single CCD or CMOS chip is employed there will be about a 25-percent reduction in total — equal horizontal and vertical — image resolution. This resolution reduction comes from the circuitry that must combine four pixels to obtain one luminance pixel. This is accomplished using a sliding, two-sample process that works both horizontally and vertically. (When generating interlace video, this process is accomplished by Row-Pair Summation.)

The desire for video free of “interlace artifacts” is most directly satisfied by shooting progressive video. However, by de-interlacing, a 1080i camera can eliminate interlace artifacts. The first step in this process is a brutal one — discard every other field. And that occurs after the Interlace Coefficient has already eliminated about 34 percent of the vertical information. (Thankfully, light sensitivity remains the same as for non-de-interlaced interlace video, which is an advantage of not using progressive scanning.)

Next the camera’s DSP must interpolate the discarded lines from the remaining lines. The key to image quality is the use of a sophisticated “intelligent” interpolator to construct a 1080-line frame. (We’ll discuss de-interlacing in far more detail when we look at the how most HDTVs mangle 1080i.) My calculations indicate that a re-constructed frame can have up to 25 percent more information than that contained in a single field.

The Table below summarizes what we have covered so far. These resolution estimates come from a computer model I’ve developed. This model inputs a CCD’s resolution and outputs an estimate of video resolution. These estimates are slightly lower than I’ve used in the past because they estimate dynamic rather than static resolution. Thus, they discount the theoretical potential of green-shift to increase resolution by 150 percent. (The model employs a green-shift factor of 108 percent.) Also discounted — the potential for intelligent de-interlacing to almost perfectly reconstruct a frame when the image is static.

My model estimates, with a total (horizontal and vertical) average error of less than one, the measured resolutions of eight HD cameras running in a total of eleven modes.

Can the resolution of 1080 get worse than this? Can its quality be made better? The answer to both questions is “yes.” In the next newsletter, we’ll look at a how Sony’s new XDCAM HD significantly increases 1080i and 1080 24fps quality.

Continue the discussion on “Crosstalk” the Millimeter Forum.
© 2009 Penton Media, Inc.