noise - compare

[dasbin]

I think Dasbin means the gradient lines inherent to the compression algorithm, not color banding or FPN?

FPN is always present and looks like faint dark checkerboard pattern across whole frame.

Color banding looks like purple or yellow streaks that somewhat follow the same pattern as FPN but is only present in -2.5 or less underexposed video. Sometimes it's all over, sometimes it's just in certain areas of the frame. The more contrast in the picture, the less banding problem you have generally. This is why sometimes it happens and sometimes it doesn't.

Gradient lines occur on surfaces that are smooth and uniform in color/texture and generally follow the radius of light-to-dark transitions. This seems to happen regardless of the above problems and regardless of bitrates.

Just to clarify, I am speaking purely of FPN and not at all about compression. I have never seen color banding that to me looked like it was something other than a manifestation of FPN. I'm curious what you mean by "somewhat follows the same pattern as FPN." I think that maybe the reason you can sometimes see color banding but not general FPN in video footage is because the processor is removing evidence of the rest of the FPN through noise reduction.
I mostly shoot in, and run my tests in, RAW stills, to see what the sensor is really doing and try to remove processing and compression from the equation. But I still think there is some noise reduction being done even in RAW mode. My theory is the camera has a constant "dark frame" reference (seperate from the one it takes during long exposures) which it subtracts from all images. I don't have any evidence to back this up unfortunately, but it would explain a few things, including why some people claim FPN/banding gets better over time with their camera, and why there is so much more FPN viewable during live-view in certain situations than in the actual recorded file. But that one may just be due to different sensor read mode.

 

[dasbin]

I don't know how qualified you are to make statements re the camera - but what you just said there is the best fitting explanation re banding I have seen yet (and I have read a LOT of threads on the issue.)

What you are saying fits exactly with my experiences - anyone care to disagree?

I'm not qualified in any way. This is based on my having spent far too long reading into CMOS sensor theory among other things, then running my own tests on my GH1, which have corroborated with what I've read.
Anyway, I'm happy to be proven wrong, but based on what I've read and seen, there is no reason to think that there is any situation in which the CMOS sensor would *not* be producing the same FPN *all the time*. It should always be there, it's just a matter of how much it's been drowned out by real image signal.
The 5%-25% figure is just a rough guess based on my own observations. In reality you can't quite put numbers to it like that, because the FPN is just part of the noise floor and is essentially unaffected by signal... but how much we see in the final image IS affected by signal, so talking about it in those terms is fine for practical discussions like this, I think. It will be more visible over certain color backgrounds than over others, and this is especially true when recording a white-balanced image in which certain color channels (and their noise) are greatly amplified digitally.

I guess I should say that I am a professional audio technician for a living, and as a result I know a lot about the process of capturing and converting an analog signal into a digital one. This has been surprisingly relevant to learning about digital cameras. The way image noise and signal behaves is very similar to the types of noise, distortion, and signal we get in the digital audio world. However I try not to make any assumptions that one is the same as the other - they just usually end up being that way, and I can draw on my audio experience to understand the explanations why.