I thought I’d ruminate an idea around image noise reduction that should help any of the cameras being used and see if it would be of interest.
I wouldn’t require much other than an intermediate PCB board between the camera module and the RPi to work with mounting holes to sit behind the camera. The PCB board would then profile galvanic Schmitt triggers isolation, isolating the control signals in both directions and allowing for a low noise power supply to be introduced for the camera (a number of options here are LDO to provide ultra low noise and if further noise reduction is needed without lag - a BJT shunt can be used to prevent opamp comparator phase delay). The result is an extremely low noise power supply for the camera module yet the RPi can operate as if there’s no change.
I’ve done something similar for low noise floor ADC conversion reaching -150dB, below this it becomes exponentially more difficult. In this case the entire system works at 24MHz so the rate for the cameras shouldn’t be a problem. The board was a 4 layer board as they had that on offer for prototyping, I’d also created a low phase noise clock source with buffering and fanning at the same time, so production costs could be reduced by panelling small boards then manually cutting. The board here is larger because it isolates both i2c and i2s separately, signal on onside and the power on this side facing:
This is the noise floor of the ADC. I’ve not used the shunt, just a lower spec LT3080 power supply due to it providing enough noise reduction for the intended purpose (the differential ADC is 124dB dynamic range, the noise shaping of the ADC can be seen as the rise right). The shunt would remove the spikes for mains power too plus drop the noise floor further:
The reason for suggesting this is that isolation board would enable a lower noise floor on the analogue side of the camera board ADC. The result would be a lower gaussian noise over the image, less noise from the RPi and any of the steppers etc from finding its way into the image prior to conversion (during the sensor chip reading process).
The change/addition would have no impact on software (other than characterisation), and would be entirely reversible (assuming the mounting doesn’t require a reprinted part!).
Just ruminating the idea and any suggestions welcome.