Originally Posted by dmunsil
1) There are no actual harmonics above the Nyquist limit. Frequencies above Nyquist are aliased frequencies, which are inherently spurious. Sure, they come out of the math, but part of signal reconstruction is filtering out the aliased frequencies, because they are not actually part of the original signal.
2) Real video does have some energy near Nyquist, and rolling it off can affect the image in a visible way. There is actually more high frequency energy in modern video than there was in the past, now that we have lots of content generated and mastered entirely digitally. We believe it is useful to preserve as much of those high frequencies as one can reasonably preserve, given the limitations of real-world re-sampling filters. Thus we believe the test is useful. As to how important it is relative to other tests, that's a judgment call. For the most part, the chroma bursts test is more intended as a relative test to see if different color modes are getting you a cleaner signal with less processing, or for use to make sure nothing is being lost inadvertently because some extra processing is turned on that shouldn't be.
3) Most of the time most people won't notice moderate (or heck, severe) rolloff of the Chroma channel. But that's kind of a low bar to get over. Most people don't notice the difference between SD and HD. Home video enthusiasts are generally trying to push the envelope and get the absolute best picture possible.
4) I don't even know if you're disagreeing with me on this. My point is that the clamping is applied to all of the signal, not just the high frequencies. RGB clipping is not a reason why you would see differential amplitude of the high frequencies.
1) You mention that the points are a "Sine wave" and I agree. The issue I am pointing out is that a Nyquist filter will treat them as a Sine Wave, when actually they are representing a Square Wave. That is: Max one end of the range, then max the other, without intervening values. If a TV/Projector were able to reproduce these perfectly they are showing two, and only two colors, which is a square wave.
Since the values are treated as a Sine Wave by the Nyquist filter this can create the issue I have been mentioning. The simplest example I can think of is a burst with two equal values at max value and two at minimum value. Then if a Nyquist filter is applied too these to create samples between each pair of samples, the created values will try to represent a Sine Wave, but this is actually the problem. The resulting sample between the two max samples will be significantly higher than the maximum value and significantly overflow, and be clamped. Then the output of the clamped Nyquist filter has three max value samples, one mid-range value, and then three minimum values. This ends up as a trapezoidal Wave but the Nyquist filter did try to make it a Sine Wave. For example if the values had not been at max/min the filter would have converted what should look like a Square Wave into an approximation of a Sine Wave, since the mid-point between the two upper values would remain clamped at max while the originals values are decreased (to a certain point). Because of this the resulting waveform shape changes. That said, I am still agreeing the patterns have value. My concern is the interpretation of the results.
2) I agree there can be some energy near Nyquist, but, especially with compression, I believe for actual "real world" video it is not significant. The Chroma Burst pattern has 100% energy at Nyquist, which is not possible with a actual source (especially with a filter at the front of the video pipeline so the video can be efficiently and effectively compressed). This can lead to misinterpretation of the results of test patterns like the Chroma Burst, which is really the only thing I am trying to point out. However, once again, agree the Chroma Burst pattern is useful.
3) I actually think we are agreeing on the essential points. Specifically we agree that we want to provide absolutely best possible performance. We do need to temper actual video verses test patterns in our trade-offs as we architect the Radiance. We certainly do not want to make test patterns look better to the determent of actual video. Fortunately this is generally not an issue. The Chroma Burst is a good example. We used it to evaluate our pipeline and were able to improve the pattern, and improve actual video when this highlighted a couple places were we could increase precision in the Verilog code.
4) I agree that clamping will affect all frequencies. My point was this can "look like" roll-off on a high frequency pattern.
So I am agreeing the patterns have value and am only trying to help people interpret the results. Also, I agree that you, Stacey, and we are doing our best to provide and support the very best in video. Finally, I believe these discussion help enthusiasts see "under the hood" about the level of detail needed for quality video, which by itself is a great goal.