search

AstarVienna / ScopeSim

A telescope observation simulator for Python.
GNU General Public License v3.0
14 stars 10 forks source link

Calibration data should have persistent structure #30

Open hugobuddel opened 4 years ago

hugobuddel commented 4 years ago

The image artifacts that are introduced by the instrument have a relatively stable structure, this should be represented in the simulations.

E.g. the quantum efficiency varies per pixel, but is relatively stable over time. Therefore it is possible to recover the quantum efficiency by observing many flat frames and averaging them: the master flat will show this structure. Similarly for the dark and illumination correction.

However, this structure is random in the current simulations. E.g. simulating a raw dark frame can be done as described in test_simplecado.py. However, this will always result in a dark with an entirely different structure.

Two possible ways to solve this (that can be combined):

Being able to set the random seed would also generalize to other effects. ESO seems to already have some header keywords for this, @hugobuddel will look those up.

astronomyk commented 4 years ago

Just a quick comment on the random seed - this is already possible. Just use the following config keyword:

rc.__config__["!SIM.random.seed"] = 42

Also possible to directly set this in each Effect discription in the yaml files:

For example in MICADO_H4RG.yaml we could add:

-   name: shot_noise
    class: ShotNoise
    kwargs:
        - random_seed: 42

The same works for the ReadoutNoise effect

This would be a nice quick 5-liner for the documentation, as the !SIM.random.seed is called where-ever a random number or sequence is needed in scopesim

hugobuddel commented 4 years ago

For the 5LS see #34 (just linking the issues).

However, this would not really solve the issue. The pattern should be generated with a fixed seed, but on top of that there should be shot noise with a different seed for each exposure. (The noise can just be on the total signal I think, but maybe there should be specific noise for different signal sources.)

There is no pattern generation in either DarkCurrent or QuantumEfficiencyCurve. So I think we should either