search

VForiel / Tunable-Kernel-Nulling

(PhD) Adaptive tunable kernel-nulling interferometry for the direct detection of extrasolar planets
0 stars 0 forks source link

Merge the two metrics #67

Closed VForiel closed 2 months ago

VForiel commented 5 months ago

An issue of having two different metrics for the bright and the kernel outputs is that the bright channel quickly converge to the maximum, which result in few light in the rest of the system. This can be highly limitating in real conditions because of the photon noise that will dominate the dark outputs and prevent us to determine if a shift set is better than another or not.

An idea, proposed by Romain Laugier, that can partially address this issue, is to merge the two metric by considering the ratio between the bright output and the sum of the kernels.

A complementary approach is to carfully select the order of shift mutation, starting by the ones near to the dark outputs to favor the optimization of the kernels before trying to maximize the bright output.

VForiel commented 2 months ago

I appear that merging the two metrics induce some local minimums in the optimization process that prevent the algorithme to converge to a satisfying solution. In fact, a shifter change that reduce the kernels intensity can increase the bright output intensity.