Improved MPV fit

[153957]

The current MPV fit is nice and fast, but the algorithm makes some assumptions which cause it to fail in some cases.
Attempts have been made to improve the algorithm (see the mpv-subtract-powerlaw branch). However, while these works better in some cases, it doesn't work in others..

In this issue a list of cases will be compiled that the algorithm should (or should not) be able to find the MPV in. In the cases where it should not find the MPV it should return the error value (-999).

[153957]

Should find MPV

I have split this in 2 and 4 detectors since they have different triggering condictions, which affect the shape of the pulse integral histogram.

4 detectors

• 501 2014/9/2 (data, csv)
• 503 2014/9/2 (data, csv) - except detector 2

2 detectors

• 104 2014/8/15 (data, csv)
• 20002 2013/10/1 (data, csv)
• 20001 2014/9/1 (data, csv)

Should not find MPV

... Please add more interesseting situations/cases to get a more complete sample.

[davidfokkema]

Good idea! The links don't work correctly? Make this a task list so that we know for which cases the current algorithm fails?

[153957]

Links are fixed.

Indeed, and perhaps also a list of the kind of special cases that might be encountered.

[tomkooij]

station 105 from sep 2015 up until mar 2016 fails the MPV fit. Probably because of strange PMT voltages (strange pulseintegrals histograms):

http://data.hisparc.nl/show/stations/105/2016/1/1/

I am investigating (https://github.com/tomkooij/lio-project/tree/master/offsets_zaanstad) but this should be "fixable". The fit fails due to the initial_guess.

[153957]

Some branches in which some work on this was performed, but were never finished/merged:

https://github.com/HiSPARC/sapphire/tree/mpv_fit
https://github.com/HiSPARC/sapphire/tree/mpv-subtract-powerlaw
https://github.com/HiSPARC/sapphire/tree/Pulseheights_integrals