AGATA Local Level Processing. - Pulse. Shape. Analysis - AGATA Week, LNL, September 15-19, 2003 Thorsten Kröll (TU München) for the AGATA PSA Team

Transcription

1 AGATA Local Level Processing - Pulse Shape Analysis - γ rel. amplitude t [ns] AGATA Week, LNL, September 15-19, 23 Thorsten Kröll (TU München) for the AGATA PSA Team

2 Outline Introduction - Task of the team - Impact of PSA on AGATA performance Software for PSA - Scheme of PSA software - Algorithms - Performance checks Hardware for PSA - PSA module - Interface Digitisation board PSA module - Architecture of PSA hardware

3 Task of the PSA team The team has to provide the optimal PSA algorithm to determine the position of gamma-rays interactions within the detectors of the AGATA array recognition of number of hits energy deposited and 3-D position of interaction points timing computing speed online-calculation Impact of the PSA results on the performance of AGATA is important!!!

4 Performance vs.. position resolution Peak Efficiency (%) Standard Ge shell with ~6 segments M = 2 M = 5 M = 1 M = 2 M = 3 5 mm standard realistic packing and smearing E γ = 1.33 MeV Packing=Smearing, Position resolution (mm) Efficiency, P/T (%) E γ = 1 MeV 5 mm packing / smearing + Pack points in segments + Center packed points P/T Efficiency Multiplicity position resolution of about 5 mm is needed for tracking with a position resolution only in the order of the segment size 25% up to 33% of the efficiency will be lost PSA is needed!!! Courtesy D. Bazzacco (INFN Padova)

5 PSA algorithms under consideration Single-step algorithms (fast, but limited capabilities) Folding Algorithm Artificial Neural Network MATLAB / FORTRAN-version Test in software Implementation on PC/(FPGA) Application of ANN chips (Milano/München) (Orsay) Iterative algorithms (better results, but slow) Genetic Algorithm Pattern Recognition C++-version MATLAB version Code improvement Extension of code Implementation on PC Portation to C Implementation on DSP (München/Liverpool/Padova) (Orsay/GSI/Jülich)

6 Scheme of PSA algorithm PSA will be done in a 2 step process: 1. Pre-processing { Zero-suppression Single-step algorithm Number of interactions First guess of positions External time reference (better resolution than individual detectors) Global time - average time of array (?) Particle detector Pulsed beam initial values 2. Full PSA algorithm Timing by algorithm Iterative algorithm Energies and 3D-positions of interaction points

7 Folding Algorithm (1) current [na] i 1 i time[ns] sharp discontinuities charge carriers collected time depends only on radial coordinate r 1 1. Differentiation 2. Folding back in time with respect to folding time 3. Superimposed wing is summed up 4. Time windows around peaks are defined r 2 r z 1z2 ϑ 1ϑ2 [u.a] [u.a] 4-4 folding time -8 time time

8 Folding Algorithm (2) 25 r 1 r 2 r i i time 5 1. Differentiation 2. Folding back in time with respect to folding time 3. Superimposed wing is subtracted 4. Polarity of signals resolves ambiguities [u.a.] Courtesy G. Ripamonti (Milano) 1 2 time

9 Folding time Increases near the corner of the section Decreases in the cap We are considering this detector shape Two different segmentations of external electrode Courtesy G. Ripamonti Induced signals from the second segmentation scheme seem to be more dependent on position

10 Artificial Neural Network (1) Planar Ge detector Wavelet transform x y Simulated signals single interactions 5 kev Gaussian noise (2 kev FWHM) NeuroOne (Windows) Wavelet coefficients Position x or y

11 Artificial Neural Network (2) Results: Courtesy C. Diarra (Orsay)

12 Genetic algorithm measured signals ABCD calculate fitness Evolution of of a population of of events calculate signals Base system of signals measured or calculated Interaction points (E; x,y,z) i signals reconstructed from base Breeding (regarding fitness): old / crossover / mutation / new current population ( E; x,y,z) + (1 E; x,y,z) + new population ( E; x,y,z) ( E; x,y,z) + (1 E; x,y,z) + (❼ E; x,y,z) + ( E; x,y,z) ( E; x,y,z) ( E; x,y,z) fittest event (❼ E; x,y,z) + ( E; x,y,z) ( E; x,y,z) Reconstructed set of interaction points (E; x,y,z) i

13 Decomposition of event by GA A4 A5 A6 B5 B4 B6 GA A4 A5 A6 B5 B4 B C4 C5 C C4 C5 C D4 D5 D6 A4 A5 A6 B4 B5 B6 C5 C4 C6 D4 D5 D t [ns] 2 interactions 1 interaction 2 interactions D4 D5 D6 A4 A5 A6 B4 B5 B6 C5 C4 C6 D4 D5 D t [ns]

14 56 Fe MeV v/c ~ 7 % Expected final results: perfect tracking 3.4 kev positional error <d> 5.4 mm 4.2 kev Correction of Doppler Broadening reconstruction of interaction points by a Genetic Algorithm 24 individual detectors with θ 9 9 single detector with θ 22 Doppler corrected using reconstructed interaction points FWHM = 4.5 kev Doppler corrected using segments FWHM = 6.3 kev non-corrected gamma spectrum FWHM = 16.5 kev 22 E γ (kev) Analysis by Th. Kröll (TUM/LNL) and M. Nespolo (INFN Padova)

15 Pattern recognition (1) Filtering of signals with the discrete wavelet transform pattern (= wavelet coefficients) 5 4 Recognition of pattern by calculating membership values with patterns from data base 6

16 Pattern recognition (2) Identified interaction Courtesy L. Mihailescu, W. Gast (FZ Jülich)

17 Performance of PSA algorithms Common data sets for performance tests 6 6 segmented detector (Surrey) / 137 Cs source (661 kev) Simulated data (Liverpool,München,Strasbourg,Surrey) GEANT3(4) Pulse shape calculations with codes available in team time jitter added to signals (up to 3 ns depending on energy) Measured data (Liverpool,Surrey) non-amplified signals from charge-sensitive preamplifier GRT4 (14 bits, 8 MSample/s) / time trigger-adc clock (jitter?) collimated source uncollimated source (imaging) } reconstructed points The quality of our results analysing measured signals can be checked ONLY by passing them to the tracking algorithm!!!!

18 Hardware for PSA Segment Preamp All segments FADC channel # energy from adaptive MWD pile-up flag time stamp from master clock (4-6 ns) traces FPGA Detector PSA module: specialised DSP board or PC optional with DSP, FPGA, ANN, cards / chips PC offers greater flexibility DSP improved efficiency in transferring data digital filters like FFT, wavelets, benefit Choice depends on speed and PRICE.

19 Hardware Digitisation board PSA Segment Preamp Segments... FADC Record data from detector De-/Compression Event building Replay recorded data at realtime speed FPGA MEMORY... enables PSA developments independent of local availabiliy of Digitiser Board and/or detector (Digitisation Team / Milano) I N T E R F A C E Detector PSA module: PC

20 PCI BOARD FOR DATA STORAGE SENSOR ANALOG SHAPER ADC PC HARD DISK PCI INTERFACE COMPRESSION ALGORITHM Courtesy G. Ripamonti State of the art - Fast and efficient real time DELTA compression - High speed (126Mbyte/sec) PCI interface - 2Mbyte/sec transfer rate - More than 1 hours storage with commercial hard disks In progress - DMA transfer mode (in background on CPU operation) - Increase speed to be compliant with AGATA spec.

21 Architecture of PSA hardware Point-to-point connection Detector Detector Detector. Flexible connection PSA module PSA module PSA module. for every detector a dedicated PSA module PSA module has only parameters of the corresponding detector # PSA modules = # detectors Detector Detector Detector. PSA module PSA module PSA module. detectors are connected to a farm of PSA modules PSA module has to have parameters of a set of detectors # PSA modules = # detectors not necessary Decision on architecture depends on number of PSA modules needed ( = speed of final PSA algorithm) has not been taken yet!!!

22 Milestones AGATA Demonstrator Scanned reference data Simulated reference data Test of PSA algorithms (offline) Definition of PSA algorithm(s) Implementation and tuning of PSA algorithm(s) (offline and on board)? DSP board prototype? AGATA PSA module prototype - development, test, and production - Memory-interface board? Interface digitisation board PSA module - development, test, and production -

23 To o do Improve computing speed (at least factor 1) Increase accuracy of results - Results of tracking algorithm will improve only slightly - Capability for Doppler correction will benefit definitely from better position resolution Transition to geometry of AGATA capsule - Cylindrical crystal of reference detector asymmetric hexaconical crystal Does this effect our results??? Hardware - PSA board based on PC or DSP board - Interface Preprocessing board PSA module Price estimate has not been done yet...

24 Team Orsay/Liverpool /Liverpool) A. Boston (Liverpool) Ch. Bourgeios (IPN Orsay) J. Cresswell (Liverpool) C. Diarra (IPN Orsay) S. Dudeck (TU München) W. Gast (FZ Jülich) A. Görgen (Saclay) X. Grave (IPN Orsay) K. Hauschild (CSNSM Orsay) P. Joshi (York) Th. Kröll (TU München) F. Le Blanc (IPN Orsay) S. Lhenoret (CSNSM Orsay) D. Linget (CSNSM Orsay) J. Ljungvall (Uppsala) P. Medina (IReS Strasbourg) G. Rainovski (Liverpool) G. Ripamonti (Politecnico Milano) N. Saito (GSI) J. Sampson (Liverpool) C. Santos (IReS Strasbourg) R. Venturelli (INFN Padova)... everybody interested is invited to participate!!!!!