Fixing problems with TRestDetectorSignal fitting methods

[mariajmz]

DON'T MERGE: This is a work in progress

@JPorron and me detected a problem with the fitting methods: the limits for the fits are hardcoded and are only valid for a sampling time of 20 ns.

[mariajmz]

I think the problem was in this line
TH1F* h1 = new TH1F("h1", "h1", 1000, 0,10); // Histogram to store the signal. For now the number of bins is fixed.

The bin limits for the histogram were fixed. This works well with signals with 20 ns of sampling time (see figure) but not with signals of 40 ns of sampling time, for example.

Figure 1. Signal with 20 ns of sampling time. Left: histogram with 1000 bins, between 0 and 10 us. Right: TGraph of the signal.

Figure 2. Signal with 40 ns of sampling time. Left: histogram with 1000 bins, between 0 and 10 us. Right: TGraph of the signal

Figure 3. Same signal of 40 ns of sampling time with the new binning. The red point is the result of the gaussian fit (without changing the upper and lower limit)

What's new:

• The number of bins and limits of this histogram that stores the signal.
• The limits of the fit (this can be improved)

[lobis]

We can take this opportunity to rethink if we want to return energy -1 or do something else.

[lobis]

I made #120 to address this.

[juanangp]

Long time ago I did a pull request to add generic methods and this issue was addressed here https://github.com/rest-for-physics/framework/blob/48c4faed4b4942a1fa48b2b1b34acd9432ae950f/source/framework/analysis/src/TRestPulseShapeAnalysis.cxx#L396. There is no need to create a TH1 because you can do the fit directly on a TGraph that you can easily retrieve from TRestDetectorSignal

Pull request is here rest-for-physics/framework#379

[mariajmz]

I haven't changed this limits (also in the landau fit). Maybe we should adjust them too

[lobis]

Yes, I think we should. The new dynamic time window should work better in most cases than what we have now.

[mariajmz]

We are maintaining the fit with the signal stored as histogram but as Juanan commented, it is not necessary as we can get the signal as TGraph with TRestDetectorSignal::GetGraph(Int_t color). What do you think?

[lobis]

I think TGraph is a better choice but on practice I don't think it makes a difference since the histogram has one bin per signal time bin, so the information content of both representations is the same.

You can change them to use TGraph if you want but I would do it in a different PR (unless it's trivial to do so).