Smart idea? - oops!

[old_tv_nut]

For your general entertainment, here's a story of an interesting project I worked on for a year before we realized it had fatal flaws.

By the 1980s, Donald Richman was a retired engineer. He had worked at Hazeltine during the formation of NTSC color TV, and wrote an important paper on the color burst signal, showing mathematically that it was adequate for color sync under the noisiest conceivable conditions.

In the 1980s he came up with an idea (on paper) to improve the color detail in NTSC or PAL pictures. When he proposed it to Zenith, they did a most unusual thing and signed a contract to work on it and build a prototype in return for an exclusive license - and I got assigned to do the circuit design and investigation.

The idea went like this:
Consider a red object with texture and detail. In NTSC, the color difference signals are band-limited, so instead of getting the full red detail, you get luminance (Y) detail with an average smear of red color difference (R-Y) imposed on it.

Now, if that object is really all a certain red color, then the details (high frequencies) in the color difference (R-Y) should follow the details in the luma (Y), and in the same proportion that the low-frequency R-Y follows the low frequency Y. So, the idea was to build a ratio circuit to recreate the supposed R-Y details. (Of course, you would do this for all colors.)

(R-Y detail) = (Y detail) x [(R-Y low frequencies) / (Y low frequencies)]

We went for months designing the analog multiplier and divider circuits and interfacing them with a set, eager to see the result. Then, we started watching programs, getting great improvement in scenic red rocks, red roses, and so on. But we watched some sports programs that suddenly had some obvious defects!

Think about an American flag hanging vertically, with red and white stripes.
The low frequencies in both R-Y and Y are just a smear, but the high frequencies in the Y are just the opposite polarity of what you want in R-Y. The R-Y detail signal should go more positive on the darker red stripes and more negative on the bright white stripes, but they were following the Y detail per the equation. The result was a flag with alternate stripes of black and bright fluorescent pink.

Richman's process also included a mode to improve edges that depended on the first derivatives of the low-frequency signals, and it had similar but even worse problems with some images, sometimes trying to divide by zero.

After a year of effort, we supplied the modified receiver to Richman, per contract, and shelved the project. He may have proposed the project to other manufacturers then, but of course we never heard any more about it.

[bgadow]

Another fascinating tail that could have been lost to time! Thanks for sharing!