1. AGATA Pulse Shape Analysis Implementation
Status
Performance
Opportunities
Dr Andy Boston
The AGATA PSA team

2. AGATA PSA Implementation
Architecture
Algorithms
Implementation
ADL
Experimental basis
Performance

3. PSA Architecture

4. Structure of Data Processing
FEE RO PP
VME+AGAVA
GTS
Local level processing
AGATA
Readout
Pre-processing
PSA
Ancillaries
Global Level processing
Event builder
Event merger
Tracking
Post-Processing
Front end electronics
Ch. Theisen

5. Structure of Data Processing
Ancillary
1R 1G 1B ….
Digitizer
or raw data file
VME
or raw data file
CrystalProducer
EventBuilder
AncillaryProducer
PreprocessingFilter
EventMerger
AncillaryFilter
PSAFilter
TrackingFilter
Consumer
Save PSA output
Consumer
Save tracked output

6. Preprocesssing Filter
Performs
Energy calibrations and xTalk (proportional) correction
Analysis of traces
Calculation of T0 from core (Digital CFD or linear fit of the first samples)
Time calibrations and shifts
Vertical normalization of traces
Define the net-charge segments
Reformats the data
The calibration files are produced by external programs as part of the calibration procedures

7. PSA Filter Signal decomposition & diff xtalk
Implemented algorithm is the Grid Search
As a full grid search
As a coarse/fine search (AGS)
Reduces size of data by factor 20
Provides the parameters for the correction of neutron damage (can also perform it)
Must be expanded to improve timing
Takes ~95 % of total CPU time
Is the critical point for the processing speed of online and offline analyses

8. PSA algorithms

9. Pulse Shape Analysis algorithms
Singular Value Decomposition 8
Adaptive Grid Search
Artificial Neural Networks 6
Particle Swarm Optimization
Genetic algorithm
Adaptive Grid Search (with final LS-fit refinements)
now
Position resolution (mm FWHM) 4
Wavelet method
Least square methods 2
Full Grid Search ms s hr
Computation Time/event/detector

10. AGATA PSA Codes Typical PSA scheme consists of 3 components
Figure of Merit (FOM) e.g. Σ |event1i – event2i|n
Search Routine: optimization of FOM over library
Adaptive Grid Search (A. Venturelli, INFN Padova)
Particle Swarm Optimization (M. Schlarb, TU Munich)
Decomposition strategy for multiple interactions
assuming maximum 1 hit per segment
segments influenced by multiple hits excluded

11. AGATA PSA Codes Other PSA schemes Partial PSA
Matrix method (A. Olariu, P. Desesquelles, CSNSM Orsay)
Partial PSA
Recursive Substraction algorithm (Fabio Crespi, INFN Milan )
Gets radial coordinates & # interactions (~ steepest slope)

12. Practical PSA Challenges
A basis calculated on a 1 mm grid contains ~ points, each one composed by 37 signals each one with > 50 samples (for a 10 ns time step)
Direct comparison of the experimental event to such a basis takes too much time for real time operation at kHz rate
Events with more than one hit in a segment are common, often difficult to identify and difficult to analyse

13. PSA Challenges
No good theory for mobility of holes -> must be determined experimentally
Shape of signals depends on orientation of collection path with respect to the crystal lattice
Detectors for a 4pi array have an irregular geometry, which complicates calculation of pulse shape basis
Effective segments are defined by electric field and follow geometrical segmentation only roughly
Position resolution/sensitivity is not uniform throughout the crystal

14. PSA IMplementation

15. PSA Implementation The signal decomposition algorithm (AGS)
The quality of the signal basis
Physics of the detector
Impurity profile
Application of the detector response function to the calculated signals
The preparation of the data
Energy calibration
Cross-talk correction (applied to the signals or to the basis!)
Time aligment of traces
A well working decomposition has additional benefits, e.g.
Correction of energy losses due to neutron damage

16. The Grid Search algorithm
Signal decomposition assumes one interaction per segment
The decomposition uses the transients and a differentiated version of the net charge pulse
Proportional and differential cross-talk are included using the xTalk coefficients of the preprocessing.
The minimum energy of the “hit” segments is a parameter in the PreprocessingFilter  10 keV
No limit to the number of fired segments (i.e. up to 36)

17. The Grid Search algorithm
The algorithm cycles through the segments in order of decreasing energy; the result of the decomposition is removed from the remaining signal -> subtraction method at detector level
Presently using ADL with the neutron-damage correction model
Using 2 mm grids -> ~48000 grid points in a crystal; points/segments
Speed is ~ events/s/core for the Full Grid Search ~ events/s/core for the Adaptive Grid

18. Signal basis generation
Simulation: MGS, JASS, ADL
Experimental: Coincidence, PSCS
AGATA Data Library
Geometries for a wide variety of detectors
E-field solver, SIMION potential arrays
Creates the calculated basis for each detector
Bart Bruyneel and Benedikt Birkenbach IKP (Eur. Phys. J. A (2016) 52: 70)

19. AGATA Data Library
dcwc
Cross Talk
correction

20. Optimisation: Crystal orientation
•400kBq Am source +
•Lead Collimator: ∅ 1.5mm X 1cm
•Front Scan at ∅ 4.7cm: 300 cts/s
•Fitfunction Risetime(θ) = A.[1+R4cos(θ- θ4)].[1+R2cos(θ- θ2)]
Risetime [ns]
θ [°]

21. R G B
Cryst
al orientatio n
ATC1
ATC2
ATC4

22. Optimisation: Origin of Crosstalk
Pure Xtalk signal:
400
200
-200
-400
-600
-800
-1000
-1200
-1400
Proportional Xtalk (50µs decay) → Energy
Differential Xtalk (only during risetime) → PSA
1000 t [ns]
Cac PA
V0,out
Zin
With Zin = 1/sACfb + (1/sCac) + Rcold
Z01
Xtalk ~ Zin / Z01
~ C01/ACfb + (C01/Cac) + s . Rcold C01
= Proportional + Differential Xtalk PA
V1,out

23. Proportional Xtalk measurement
For any 1406keV single event in the detector:
Segment labeling:
0.0
Additional contributions from seg. to seg. capacities!
-0.5
Core to seg capacity
Baseline Shift in A1 (keV)
-1.0
A2..A6
-1.5 B1 F1
-2.0 X
-2.5
Sectors: A...F
Rings: 1...6
-3.0 6
Hit segment number 30 36

24. Crosstalk correction: Motivation
Crosstalk is present in any segmented detector
Creates strong energy shifts proportional to fold
Tracking needs segment energies !
Sum of segment Energies vs fold
Segment sum energies projected on fold
2folds : Core and Segment sum centroids vs hitpattern
…All possible 2fold combinations
Energy [keV]
B. Bruyneel, NIMA 599 (2009) 196–208

25. Cross talk

26. Cross talk correction matrix
Core - seg
Energy split
NN -xtalk

27. Cross talk correction: Results

28. How to measure derivative Xtalk?
(1) B. Bruyneel et al. NIM A 569 (2006)

29. Radiation damage from fast neutrons6 5 4 3 2 1
A B C D E F CC
/150
White: April 2010  FWHM(core) ~ 2.3 keV FWHM(segments) ~2.0 keV
Green: July  FWHM(core) ~2.4 keV FWHM(segments) ~2.8 keV
Damage after 3 high-rate experiments (3 weeks of beam at kHz singles)
Worsening seen in most of the detectors; more severe on the forward crystals;
segments are the most affected, cores almost unchanged (as expected for n-type HPGe)

30. Crystal C002 April 2010 July 2010  corrected
CC r=15mm
SG r=15mm
April 2010
CC r=15mm
SG r=15mm
July 2010
 corrected
The 1332 keV peak as a function of crystal depth (z) for interaction at r = 15mm
The charge loss due to neutron damage is proportional to the path length to the electrodes. The position is provided by the PSA, which is barely affected by the amplitude loss.
Knowing the path, the charge trapping can be modeled and corrected away -> Lars

31. PSA performance

32. Grid search algorithm result

33. PSA performance analysis
12C(30Si,np)40K
LNL commissioning
b = 5.5cm
P.-A. Söderström et al. / NIMA 638 (2011) 96–109

34. AGATA PSA and Data Analysis Schools and WS
The PSA and Data Analysis WG organises “regular” schools and WS:
Liverpool 2011 (EGAN)
GSI 2012 (EGAN)
LNL 2013 (EGAN)
GANIL 2016
The teams within the WG aim to have (at least) quarterly team meetings.

35. Summary... Lots of opportunities
In beam use AGS algorithm (Narval implemented)
Offline have AGS and Particle Swarm (Narval emulator implemented)
Continuous improvement of signal basis
Push towards experimental basis generation
Implementation of multiple segment interaction algorithm
Challenges:
Availability of AGATA capsules for characterisation
Clustering of points distributed inside detectors
Continuity of available personnel to implement PSA algorithms
Documentation + Howto guide
This work is a big effort from a large number of people.. Thanks to all.

36. AGATA Pulse Shape Analysis Implementation
Status
Performance
Opportunities
Dr Andy Boston
The AGATA PSA team