Found VME crate yesterday reporting AC power failure. Workstation unresponsive.

Power cycled everything and restarted -> all OK.

Long-term stability of the setup appears to be still poor.

According to Carl U. the DPP-PSA software is not meant to be used with negative pulses, and some of the
functions that are currently not working may be recovered fixing this issue.

Changed Pulse Polarity in V1730PSD control panel Positive -> Negative 
No changes. Note trigger is currently supplied by external CFD
QLong histograms are still not counting

Set DC offset to 64000. The negative part of the pulse is cut off. Wave histogram shows nothing below zero.
Inversion doe snot appear to be working.


Changed BNC-4 Pulser polarity Negative -> Positive and changed polarity of time signal going to CFD. CFD
triggers OK.
No changes. QLong histograms not counting.

Enabled self-triggering on Ch0 and removed CFD input. DAQ no longer triggers. Put CFD back in.

Changed DC offset of Ch0 to 64010 to remove DC offset. Pulse now looks as attach 1.
Disconnected CFD output -> DAQ does not trigger. Reconnected CFD.
QLong histograms still not counting.

Attempted to change options in V1730PSD control panel one by one to see difference
Enabled Charge Pedestal -> No difference. Disabled
Changed Charge Sensitivity 5 fC -> 5.12 pC. No difference. Back to 5 fC
Changed Discrimination Mode CFD -> LED & removed CFD input. No triggers. Reverted

Changed Baseline Mean 16 -> 1024. Wave changes as per attach 1. QLong histogram starts counting at low channel.
See attach 2.
Changed DC offset 64010 -> 60000. Peak in QLong histogram moves to ~1050 channels. See attach 2.
Changed DC offset back to 64010.
Changed Baseline Mean 1024 -> 64. No changes in either wave shape or QLong peak position.
Changed Baseline Mean 64 -> 256. No changes as above.
Changed Baseline Mean 256->16. Wave is back as per attach 1, QLong no longer counting.

Fit: 3 lines
Free parameters: 2x boundary points between lines
Algorithm: Nelder-Mead, 10 iterations. (Centroid defined incorrectly from entire simplex instead of best two
vertices)

Attach 1 - height (difference between intercepts)
Attach 2 - rise times (x-diff between boundary points). Note peak at tr=100 is spurious.
Attach 3 - alpha signal. Fit from Excel Solver (GRG nonlinear)
Attach 4 - Signal incorrectly identified as H20 tr100 signal. 
Fit displayed is correct, from Excel Solver (GRN nonlinear) tr seems 200 samples instead of 100 reported.
Attach 5 - Slow signal, poorly fitted by fitting functions
Attach 6 - Comparison between signals at H=110
Attach 7 - Heat map

Fit: line + power law + line
Free parameters: 2x boundary points between lines
Algorithm: Nelder-Mead, 10 iterations. (centroid correct)

Attach 1 - height (difference between intercepts)

Attach 2-8 - Signals with H~110, different t values (see filenames). 
Note t=370 and t=870 show discontinuities in the middle of the rising front. Pick-up?

Attach 9 - Comparison of the above. Note t=370-570 look visually similar in spite of very different fit results. 
Pulser *not* shown

Attach 10 - Heat map

Attach 11 - rise times (x-diff between boundary points), for H>20 (no noise)

Fit: line + power law + line
Free parameters: 2x boundary points between lines
Algorithm: Nelder-Mead, *20* iterations.

Attach 1: rise times for H>20. Main peak is now double.
Attach 2: heights. Second rise time peak corresponds to four peaks at alpha energies. physical origin unclear.
Overfitting?
Attach 3: comparison between waves of two rise time peaks

Fit: line + power law + line
Free parameters: 2x boundary points between lines
Algorithm: Nelder-Mead, 20 iterations.

Attach 1: Rise times. 
Peak below 200 is pulser.
200 samples should correspond to alphas, according to R3 analysis. 600+ should correspond to slow signals
(electrons?)

Attach 2: Wave amplitudes (pulser not shown).
150-200 channels should be correspond very roughly to 5-6 MeV

Attach 3: An alpha candidate signal. H 163 t 198
Attach 4: A slow physics signal (electron?) candidate. H 100 t 651

Attach 5: Noise with low Chi2. Reported as H2 t 406. Note fit is GRG nonlinear - cannot reproduce dataReader
Nelder-Mead Simplex fit in Excel (working on it).
Attach 6: Noise with high Chi2 value. Reported as H40 t150. As above.

Attach 7: 2D heatmap

Pulser frequency: 2.165 Hz (462 ms)

Attach 1: Pulser events vs. total events

Fit: line + power law + line
Free parameters: 2x boundary points between lines
Algorithm: Nelder-Mead, 20 iterations.

Attach 1: Wave amplitudes. 
Peak at 300 is pulser.
Some physics events in between obvious noise and pulser.
No obvious signal from Bismuth source

Attach 2: Alpha (?) wave sample. Rise time ~ 200 samples.
In spite of no signal from source, some alphas are visible.

Fit: line + power law + line
Free parameters: 2x boundary points between lines
Algorithm: Nelder-Mead, 20 iterations.

Attach 1: Wave amplitudes. 
Pulse not shown at 1000 channels.
Obvious signal from Bismuth source with visible peaks. Not all peak origins clear.

Attach 2: Risetimes.
Pick-up / microphonic noise quite evident (compare with background run with B-grade silicon)
Pulser relatively well-defined in risetimes
Electrons not well defined

Attach 3: Waveform for electron candidate fitted as risetime = 26 samples
Attach 4: Waveform for electron candidate fitted as risetime = 153 samples

Visual difference between the two unclear, aside from noise.
Looks like fitting function is failing. Will have to change algorithm to draw any conclusions.

Given very poor fit using line + power law + line, changes were made to the fitting algorithm.
- Starting paramters optimised for new typical wave shape
- Reverted to 3 lines, since signals are faster
- Subtracted wave minimum from all wave points to have a more realistic chi2 estimate
- Discarded all waves where 2nd half of signal > 1st half of signal. Should correspond to pick-up / noise

Fit: 3 lines
Free parameters: 2x boundary points between lines
Algorithm: Nelder-Mead, 20 iterations.

Attach 1: Wave amplitudes
Attach 2: Rise times
No major differences vs. power law method, but note candidate electron peak in rise time spectrum is now sharper.

Attach 3-9: Waveforms with H~90 and increasing rise times. 
Note fit shown are via Solver's GRG, not NM.
Fits in waves 1-4 give same result for GRG and NM.

Fit in wave 5: NM-> H72 t102 vs. GRG-> H80 t25
Fit in wave 6: NM-> H96 t128  vs. GRG-> H90 t23 
GRG risetimes are more realistic. This appears to be due to very short baseline pre-rise.
Will increase in next run.

Fit in wave 7 is rare pile-up event. Code cannot yet cope. Bad fit.