[old_tv_nut]

Not only were there lots of compromises, but unspecified things in the NTSC standard that produced problems in the introduction of IQ demodulation in the 90s, when it was possible to produce a good IQ receiver.

One biggie was that all transmitters and receivers had an audio trap at the equivalent baseband sound frequency of 4.5 MHz. So, the upper I and Q sidebands were sharply curtailed +920 kHz. But the NTSC never specified that the Q channel should have a symmetrical trap at minus 920 kHz, either after the Q modulator or in the Q baseband. As a result, a receiver with full wideband I response would get quadrature distortion (or not) depending on the particular brand and model of color encoder. I went through a series of experiments at Zenith when RCA briefly brought out their solid state I/Q set, to see if it was worthwhile. We could not understand why RCA had the high frequency I content turned down so much until we tried it at full amplitude and saw the problem. The problem was that with a strong Q axis color, the quadrature I interference was noticeable, even if it was only 10% of the main signal in the Q.

There was another problem even with proper I/Q filtering. The idea that the eye can't see Q axis color detail is true enough for grating patterns of green/purple. But, a single narrow object is like a pulse of color and therefore contains all frequencies, high and low. This means that the typical Technicolor yellow titles would turn noticeably orange on the vertical strokes of small letters, while being yellow on the broad strokes. This effect would happen on any color that contained both I and Q. (Of course, faces weren't affected, since their color is mostly on the I axis.) When I showed this to engineering management, they said no way, it looks too startlingly different. Again, a reason not to turn up the color detail very much.

I suspect that the NTSC committees convinced themselves that things were acceptable because their test material did not contain much strong color detail that was off the I axis (remember, their repeatable material all would have been from film, which already was biased toward the I axis to prevent off-color skin tones), and because of most of the viewing being on small screens. Edit: Hypothetically, the single sideband part of the I signal should be boosted 6 dB by a "shelf filter" in the receiver to regain full detail amplitude. but I don't know if anyone actually implemented this.

Then there was the change in IF design over the years from the flat, sharp cutoff to a "haystack" response, where the NTSC assumed the former and even enshrined a phase correction filter into all color transmitters that was meant to compensate a receiver flat IF response. By the 90s, surface wave IF filters were available that could have the "haystack" amplitude response but with the exact phase response that the NTSC had assumed back in 1954, so that problem actually could have been fixed.

In the experiments, the quality went like this:

Separate baseband Y, I, Q with I and Q filtered to 1.5 MHz and 0.5 MHz: the best, but some titles and some objects that were between I and Q color phase would jump out as the wrong color on narrow parts and edges.

Composite with I and Q demods but no RF modulation or sound trap: like the above, but with more cross-color than equiband demods.

RF modulated composite with I and Q demods: quadrature distortion pops up, especially if the encoder Q filter did not have the 920 KHz trap.

In all these stages of experiment, there was improvement in the detail saturation of red objects, which was deemed worthwhile, but not with the other artifacts that crept in. Management's judgement was that it was better to lose color saturation in the details without introducing artifacts. Since the home viewer didn't have the original studio scene to compare to, this was deemed less objectionable than obvious changes in the hue of edges and narrow objects.

It must be noted that receiver color bandpass designs were simplified over the years, and simplified LC filtering means more gradual cutoff of the bandpass. So, equiband sets produced a bit more color detail than a sharp 500 kHz cutoff would. This extra detail contained quite a bit of quadrature distortion above 500 kHz, but the extended bandwidth was low in amplitude, and therefore took advantange of the eye's reduced hue discrimination that the NTSC originally touted. At normal viewing distance this gave some apparent increase in chroma resolution, and when people used composite or S-video inputs from DVDs, there was a visible increase in color resolution because there was no sound trap messing up the upper sidebands.