I've seen similar banding in some tests when I had set analog gain to 1.0 on a warm night. With analog gain set to 16.0 and Peltier cooler attached to back of camera PCB it was gone.
My guess is that IMX477 does some software de-noise that causes banding. I tested this guess by calculating power spectrum of a dark frame and there was noticeable drop in higher frequencies. Unfortunately, even disabling hot-pixel detection did not help. I did not investigate it further.
Fan is standard 40mmx40mm, there is some generic copper thermal paste between Peltier and radiator. The plywood mount is DIY made with with hole drills. It is a bit too thick so I used 2x C-CS mount adapters (like the one bundled with the camera) to get more space, otherwise I would not be able to screw in lens adapter.
There was a bit of neoprene foam between radiator and fan to reduce vibrations but it looks like I lost it somewhere.
About condensation: it was always wet or even icy in operation but it never caused any electrical issues, I guess condensed water was not very conductive? Once on a cold night (about 5C) I had water condensation on the sensor itself but it did not cause any lasting damage. I've read that putting a fresh silica gel desiccant packet into lens adapter helps. I had plans to attach temperature probe to lens mount and control cooling power form Pi itself, will get back to it in the future. The camera sensor itself has embedded thermometer but it is not exposed in camera driver.
My guess is that IMX477 does some software de-noise that causes banding. I tested this guess by calculating power spectrum of a dark frame and there was noticeable drop in higher frequencies. Unfortunately, even disabling hot-pixel detection did not help. I did not investigate it further.
reply