Noisify can drive signal values negative

[cmarooney-stfc]

Noisify (herbert_core\utilities\math) modifies the signal values of its input by adding fMR where f is an input scaling factor defaulting to 0.1, M is the maximum value in the signal vector, and R is a random number.

Currently R is chosen from a normal distribution with std.dev=1 and presumably mean of zero (the documentation is coy, just says standard normal). This means that in principle the signal can go negative; this will be more likely as f is made larger. From discussion at a standup meeting, this may not generally be wrong, but really should only be expected as a result of background subtraction, where it is a permissible artifact.

Possible solutions include truncating the distribution generally or for particular pixel values, or replacing it with an appropriate uniform distribution which does not have this problem. This issue is opened for discussion of the best alternative and whether to change to it.