Dino Bazzacco INFN Padova

Dino Bazzacco INFN Padova
AGATA Data Processing Dino Bazzacco INFN Padova

Topics General structure of data acquisition
Data processing model == data analysis model Narval (ADA/C/C++) and actors  emulators Data flow  adf Survey of actors Hands on

38 high-resolution signals / detector AGATA Triple/Double Cryostat
AGATA detectors 80 mm 90 mm 6x6 segmented cathode Cold FET for all signals Volume ~370 cc Weight ~2 kg (shapes are volume-equalized to 1%) Energy resolution Core: keV Segments: keV keV) A. Wiens et al. NIM A 618 (2010) 223 D. Lersch et al. NIM A 640(2011) 133 38 high-resolution signals / detector AGATA capsules Manufactured by Canberra France AGATA Triple/Double Cryostat Manufactured by CTT

Structure of Electronics and DAQ
Up to 180 detectors GL Trigger Detector preamp. Digitisers Ancillary Analogue Clock 100 MHz T-Stamp Preprocessing Electronics PreProcessing & PSA Core + 36 seg. Ancillary Synchronous Event Builder Tracking Control, Storage… Ancillary Buffered Other detectors interface to GTS via mezzanine merge time-stamped data into event builder (merger)

M.Bellato , Agata Week Darmstadt - 6-9 September 2011
GTS : the system coordinator All detectors operated on the same 100 MHz clock Downwards 100 MHz clock + 48 bit Timestamp (updated every 16 clock cycles) Upwards trigger requests, consisting of address (8 bit) and timestamp (16 bit) max request rate 10 MHz total, 1 MHz/detector Downwards validations/rejections, consisting of request + event number (24 bit) Up to 256 requesters. Present implementation limited to 40 M.Bellato , Agata Week Darmstadt September 2011

Architecture of the system
Local Level : where the individual detectors don’t know of each other. Electronics and computing follow a model with minimum coupling among the individual detectors, which are operated independently as long as possible Electronics is almost completely digital, operated on the same 100 MHz clock Data processing (in the electronics and in the front-end computers) is the same for all detectors and proceeds in parallel Every chunk of produced data is tagged with a 48 bit number (time stamp) giving the absolute time (with a precision of 10 ns) since the last hard-reset of the system  roll around takes place every 32.5 days. Global Level : where the detectors do know of each other By means of the real time trigger, which reduces the data rate by selecting the class of interesting events Via the event builder and merger that assemble the event fragments (including the ancillaries) into complete events that are further processed In the Tracking and in the Physical analysis stages The fact that 3 (and 2 at GSI) crystals are packed in one cluster does not produce correlations in their data; only seen in the name of the detectors: [1-5][RGB] at LNL  2G where 2 refers to the 2nd cluster produced [00-14][ABC] at GSI and GANIL  12B where 12 is the “hole” in the frame

Producer reading psa_xxx.adf
00A 00B 00C 01A 01B 01C…. PSA Hits Auxiliary Digitizer+PreProc or raw data file Producer reading psa_xxx.adf VME or raw data file Ndets Ndets CrystalProducer (Save original data) EventBuilder AncillaryProducer PreprocessingFilter AncillaryFilter EventMerger PSAFilter should be split in two actors global actions and TrackingFilter PostPSAFilter Consumer Save PSA output Consumer Save tracked output

Narval Topology at LNL 5 ATC + 1 Ancillary
Produced by the run control GUI Krakow at LNL

Data processing and replay
Offline = Online A series of programs organized in the style of Narval actors Director of the actors Online : Narval Offline : Narval or narval emulator(s) Depending on the available computing resources Single computer or Farm with distributed processing The system is complex and “difficult” to manage Very large number of channels NumberOfCrystals*38  LNL 570  GSI 836  GANIL >  No chance to take care of them individually Rely on automatic procedures Hope that the system is stable Result depends on average performance Need GUIs, Histogram Viewers and and scripts to manage the work Processing/Analysis model developed with contributions of several people Xavier Grave, Olivier Stezowski and their groups for the architecture and the basic tools Joa Ljungvall, Daniele Mengoni, Enrico Calore, Francesco Recchia The users and their feedback

Emulator(s) femul GammaWare and Watchers
Originally intended to help developing and debugging the user libraries which is very hard to do in a distributed processing environment like Narval. Gradually developed as a full emulation of the Narval framework with the limitation of being a single process running a specific machine (threads used to distribute the work on the available cores). Configuration (detectors, actors, ..) specified by a “topology” file Generation of configuration parameters vie gen_conf.py GammaWare and Watchers Don’t know well its present status but this package is possibly more suitable for specific tasks and test, instead of inserting everything in an actor that is going to be used in the DAQ

Use of actors, managed by the framework (Narval or femul)
Creation process_config and object creation … process_initialise Read parameters from confPath/Actor.conf (e.g. Conf/3R/PSAFilter.conf ) Generate configuration-dependent internal objects, open data files process_start (Important for Producers/Consumers) process_block (or adf::ProcessBlock)  the event loop where the actual work is done process_stop (Important for Producers/Consumers) Destruction Save spectra and close files A fair amount of repeated code (boilerplate) saved by inheriting from proper base classes (NarvalInterface, NarvalProducer, NarvalFilter, NarvalConsumer provided by the adf library. The C Narval API (and its C++ binding in the adf library) contain other less-used/less-useful functions

Olivier Stezowski

Olivier Stezowski Olivier Stezowski

Structure of analysis directories
The directory where you produce your data ( e.g. /agatadisks/data/2011_week41/zCurrentNarvalDataDir) contain some standard sub-directories Conf  configuration of actors, calibrations, ... for each detector LNL) 1R 1G 1B 2R ... 5B Ancillary Global with minimal differences between online and offline Data  data and spectra produced during the experiment 1R 1G 1B 2R ... 5B Ancillary Global Online writes data here Offline replay takes data from here Out  data and spectra produced during data replay 1R 1G 1B 2R ... 5B Ancillary Global Offline writes data here Conf, Data and Out are often symbolic links to actual directories

Typical configuration directories
Conf/12A Conf/Global CrystalProducer.conf CrystalProducerATCA.conf PreprocessingFilter.conf PreprocessingFilterPSA.conf PSAFilter.conf PostPSAFilter.conf xdir_ cal xinv_ cal BasicAFC.conf BasicAFP.conf EventBuilder.conf EventMerger.conf TrackingFilter.conf CrystalPositionLookUpTable BasicAFC.conf

Binary spectra Simple C-style multidimensional (max 6) arrays written mostly in binary format For historical reasons the format is not recorded in the file. Often written as part of the file name: Prod__ UI__Ampli.spec Is a file dump of an array defined as unsigned integer Ampli[4][38][32768] containing 4*38 = 152 spectra of channels No difference between spectra and matrices; the type is only an hint to how to interpret them The viewers TkT and Mat can decode and interpret the format. The user can always override the program. Other programs (e.g. RecalEnergy) can interpret the spectrum length and type but the user has to specify the number of spectra to act upon.

Some Useful programs TkT spectrum viewer (& al) tailored to composite detectors A nightmare concerning programming-style, but contains ~all I need(ed) to analyze AGATA data. Virtually no documentation. RecalEnergy Analysis of spectra looking for peaks xTalkSort xTalkInvert (xTalkMake SortCapsule) Sort and analysis of Agata events without traces SortFFT Fourier spectrum from “Long Traces” SortPsaHits Sort of PSA hits (special format) to determine neutron damage correction parameters gen_conf.py Unified procedure to produce configuration files for all actors solveTT.py Optimize time alignment of “equal” detectors

Local Actors Producers Filters Consumer CrystalProducer
Readout of electronics, or get raw data from file Local event builder Save original data to be able to replay experiment Raw projections Filters PreprocessingFilter  PreprocessingFilterPSA Energy calibrations Retrigger and Time calibration Cross talk correction Amplitude calibration and time alignment of traces Improved pile-up rejection PSAFilter  PSAFilterGridSearch Decomposition of calibrated experimental traces by comparison with a calculated signal basis In principle more than one algorithm available but only one used in practice. PostPSAFilter Neutron deficit corrections Recalibrations of energy and time Smearing of positions Consumer Save PSA hits for “Global Level”-only processing

Global Actors Event Builder Filters Consumer
Essentially a producer with multiple inputs Assemble event fragments from Ge detectors (Builder) and Auxiliary detectors (Merger) Filters TrackingFilter Global histograms (time, energy …) Grouping and further filtering Format data as needed by the specific Tracking Filter Histograms of tracked gammas Consumer Write tracked data

CrystalProducer Reads the data from:
The PCI express driver connected to the ATCA electronics Raw data files (event_mezzdata.bdat) for the offline Acts as a local event builder to check and merge the sub-events coming from the ATCA readout (or from the raw data file) For the online has to work in nonblocking-mode to avoid blocking the ACQ This feature was implemented with boost::threads Damiano has now implemented the “select” and threads have been removed The local event builder and the management of the mezzanine data should be completely rewritten … Can write the original data with the format of the ATCA electronics, and various other formats (e.g. only the energies for calibrations) Prepares data:crystal frames, without using adf (version using adf exists, maybe slower) Not much to do for the users both for Data Acquisition and Data Replay

What does it read and write
Raw event: data from 7 mezzanines hosted in two ATCA Carrier Boards Mezzanine length (for 100 samples traces): 16+(8+100)*6 = 664 short (2 bytes) words Event length: 7*664 = 9296 bytes Local event builder to assemble data from 2 readout ports, according to mapping specified in CrystalProducerATCA.conf contains the mapping Write original data (event built, compressed 3.6 kB) data files typically split after 1 M events ~ 3.6 GB event_mezzdata.cdat.0000 event_mezzdata.cdat.0001 … Generate raw spectra for amplitudes and baselines Format data into an data:crystal adf frame and send it to the data flow.

Raw data from EDAQ CC High Gain Low Gain A6 A5 A4 A3 A2 A1
B B B B B B1 C C C C C C1 D D D D D D1 E E E E E E1 F F F F F F1

Digitizers 100 MS/s max frequency correctly handled is 50 MHz (Nyquist) 14 bits Effective number of bits is ~12.5 (SNR ~75 db) 2 cores (one board) and 36 segments (in 6 boards) Core 2 ranges 5, 20 MeV nominal Segments taken either at high gain (5 MeV) or at low gain (20 MeV) The digitized samples are serialized and transmitted via optical fibre (1/channel) to the preprocessing electronics. Transmission rate is 2 Gbit/s on each of the 38 fibres. Power consumption 240 W Weight 35 kg New digitizers and preprocessing electronics with same specs, much lower Power consumption, weight and cost have been developed (for AGATA and GALILEO) and are under test.

Range of digitizers 1 5R 3R 2R 4R 1R 5G 3G 2G 4G 1G 5B 3B 2B 4B 1B
Some cores and segments saturate below 3 MeV on the high-gain range Gain in the digitizers should be reduced by ~30% test5atc_110603/run_000

Range of Digitizers Values derived from the energy calibration coefficients , with offsets adjusted to have the “zero”of preamp signals at ¼ of the conversion range Digitizer modified before the GANIL phase to better match the ranges of 6 and 20 MeV

Range of digitizers 2 Operation at high counting rates 1 ms
1R CC high gain 1R CC low gain 1R B1 (low gain) CC ~30 kHz segment B1 ~7 kHz

Prod__4-38-32768-UI__Ampli.spec [1]-38-32768 used for energy calibrations
A B C D E F1 A B C D E F2 A B C D E F3 A B C D E F4 A B C D E F5 A B C D E F6 CC0 High Gain CC1 Low Gain

Prod__ UI__TstDiff.spec ms

PreprocessingFilter Performs
Energy calibrations and cross talk corrections 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 as data:ccrystal frames After Preprocessing: energies are stored in units of keV times are in units of samples (10 ns) (but time calibration parameters are in ns) positions are given in mm, when they show up after the PSA

Calibrations Energy Cross-talk Time Response function of the system
Gain-only, no offset coefficient needed because of the way the amplitude is generated in the preprocessing electronics. Challenged by the Differential Non Linearity (DNL) of the ADC Cross-talk 36*36=1296 coefficients to correct capacitive coupling correlations between segments and core Used also to recover up to one broken or missing segment per crystal Time Local  36 coefficients to alignment of segment to the core great influence on the performance of PSA Global  alignment of crystals and other detectors important to reduce random coincidences Response function of the system To match the PSA calculated signals to the real data Neutron damage Recover energy resolution of the segments using info from PSA.

PreprocessingFilterPSA.conf

Cross-Talk Proportional Differential
Affects energy spectra of multi-segment events xTalkSort … Differential Affects rise-time of signals and therefore PSA Not yet fully characterized

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] For the last part of my talk, I would like to draw the attention to a problem common to any segmented detector : namely cross talk. This creates strong energy shifts in the segment sum energy, and this proportional to the hit multiplicity: Doubles are shifted compared to singles. Three folds even more, and so on. >> If one analyzes the doubles, one observes that it does not simply consists of a single peak, but: By sorting the two folds into all possible twofold combinations, one observes that the centroids of these peaks are individually shifting according to a regular patten. The same happens in the core,however there is no such offset observed as is in the segments. There is thus real physics behind this phenomenon. >>

Cross talk correction: Results

Cross talk

Correction of negative energy in non-hit segments
D1 D1 C1 C1 B1 B1 keV keV A1 A1

Missing/Broken segments
[1332.5, ] 1332.5 [1312.9, 366.2] Broken FET SumSegs Core Core Core Exploiting the redundancy Segs==Core Compensate loss of energy in core Assign energy to missing segment Emissing = Ecore -SumOther Remove ghost peaks in neighbours and determine energy of missing segment as an extreme case of cross-talk Final energy of missing segment using again SumSegs==Core Recovery of broken segment A5 in Crystal 2G (LNL 2010)

Time calibrations at Local Level
Need to align the traces because the hardware does not compensate all time lags Need to re-trigger the event of a whole inside the capture time. Important is the absolute time of the first sample (timestamp) Straight line fit or Digital CFD of the signal rise-time using info from core and segments.

Time alignment of segments re Core

PSAFilter Signal decomposition
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

Pulse Shape Analysis concept
B3 B4 B5 C3 C4 C5 CORE measured 791 keV deposited in segment B4 Courtesy Francesco Recchia

B3 B4 B5 (10,10,46) C3 C4 C5 y C4 B4 CORE measured calculated D4 x A4 E4 F4 791 keV deposited in segment B4 z = 46 mm

Result of Grid Search Algorithm B3 B4 B5 (10,25,46) C3 C4 C5 y C4 B4 CORE measured calculated D4 x A4 E4 F4 791 keV deposited in segment B4 z = 46 mm

Pulse Shape Analysis Algorithms
8 Singular Value Decomposition Adaptive Grid Search  The standard AGATA PSA, so far 6 Artificial Intelligence (PSO, SA, ANN, ...) Full Grid Search Position resolution (mm FWHM) now Genetic algorithm 4 Wavelet method Least square methods 2 ms s hr Computation Time/event/detector

The Grid Search algorithm implemented in Narval as PSAFilterGridSearch
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) The number of used neighbours is a compile time parameter (usually 2 as Manhattan distance) 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 Using ADL bases (Bart Bruyneel) with the neutron-damage correction model (the n-damage correction is actually performed in the PostPSAFilter) 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 Search To speedup execution, parallelism has been implemented (using boost:threads) with blocks of ~100 events passed to N parallel threads of execution

Adaptive Grid Search in action
A B C D E F CC

Adaptive Grid Search in action
6 Final Result Event with 3 net-charge segments (D1, D2, E3) 5 Final residuals 4 Smallest net-charge segment searched after removing result of the other two 1 Initial residuals, after removing a hit at the center of the net charge segments 2 Largest net-charge segment passed to the search 3 Second net-charge segment searched after removing result of largest one A B C D E F CC

Decomposition with 2 hits/segment
How Speed Clearly an improvement for some events Overall effect on tracking seems rather small More by Waely ?

Response function and PSA
Fast and slow signals belong to specific parts of segments If response function not correct signals appear as too fast or too slow, ending up in the corresponding (possibly wrong) regions Same effect if time alignment not correct (but time is a “free parameter” in PSA) Parameters Bandwith of the preamplifier Fall time of preamplifier (~1% on amplitude) Fourier transform of “empty” traces

Effect of time alignment
5 10 15 20 25 30 22+Tfit Effect of time position of the experimental trace Similar behaviour when scanning over the risetime of the preamplifier, because a too-fast/too-slow response sees the experimental traces as slow/fast or late/early

Best time alignment and fit of T0 of signal
Still a lot of clustering Difficult to get better results Is it Problem of: Algorithm ? Calculated basis ? Preparation of data ? with a different algorithm and a completely different way of preparing the data GRETINA has similar effects Could be an intrinsic limit of the method

Cross talk and PSA Logically, cross talk is part of response function of the system Proportional cross-talk applied to signal basis Differential cross-talk applied in the same way but clearly not done well. Question of proportionality between the two. Example

Response function and cross-talk applied to the signal basis
Average over segment C2 Red experimental signal Green raw basis signals Black adapted basis Nearest neighbours particularly bad Maybe a problem of differential cross-talk

Speed-ups Independent local level processing for each detector:
Normal operation in online because each detector is served by its own workstation For replay distribute the work on different machines or on the grid/cloud For each of these processors the grid search algorithm can be distributed on more than one core (essentially transparent, using the threads API provided by the Multiplatform Boost C++ libraries (not yet tried but should be easily ported to std::treads in C++11) ) But: quality is more important than speed and that’s where we are still suffering.

PostPSAFilter Final energy and time calibrations
Recovery (partial) of neutron damage using info from PSA Impact of neutron damage on: energy resolution Signal shape and PSA

Correction of Radiation Damage
Line shape of the segments of an AGATA detector at the end of the experimental campaign at Legnaro (red) The correction based on a charge trapping model that uses the positions of the hits provided by the PSA restores the a Gaussian line shape (blue) B.Bruyneel et al, Eur. Phys. J. A 49 (2013) 61

Radiation damage from fast neutrons Shape of the 1332 keV line
6 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)

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 crystall depth (z) for interactions 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 (Bart Bruyneel, IKP Köln)

EventBuilder/EventMerger
Work on data filtered by the real-time digital trigger The software builders use the timestamp of the data coming from the detectors Could also reconstruct using the eventnumber but this is given with only 24 bits and rolls over every 16M events  modulo electronics The online Narval builders had a hard life in keeping up with the variability of data flow and sometimes lost synchronization. A solution was to have very large timeouts in the reconstruction algorithms but this slowed too much down the stop procedures The latest improvements by Xavier Grave have not yet been tested The offline builders have a simpler life because data flow from files is (?) more smooth (and the timeouts in the reconstruction algorithms can be very large)

TrackingFilter This is the first point where we can check integrity of global events and analyse them As AGATA only Build global spectra: energy, time and Tgg Adjust gain of CC (and SG, not implemented)  should be moved to postPSA Force sum energy of segments to energy of CC  small loss of resolution but removes extra left tail As AGATA + Ancillary Spectra of Ge in coincidence with ancillary Gate on TGe-Tanc (still timestamp only) to select the data for tracking Input of tracking can be written in mgt-style for offline tracking with OFT or MGT Tacking algorihm TrackingFilterOFT Waely’s, parameter-free cluster tracking algo Has been a bottleneck, but now can achieve 20 kHz Problem: Timing not reported well Output of Tracking written as adf or root tree Both outputs keep the input hits (and the Ancillary data if present), adding the tracked gammas and the Ancillary filter Timing is not yet reported properly

Global Actions Check that all detectors ok Better time windows
Inspect spectra of cores and segments Inspect g-g patterns of segments Global time alignment Better time windows In special cases use groups and conditions on groups

gg patterns for (36·N)2 segments
14A B C 07A Missing segment 28 (E5) Should be fixed in PreprocessingFilter. (36·21) = 756 segments xTalk problems in neighbours

Global time alignment Alignment of GTS tree only up to GTS mezzanine, not down to the digitizers  expect to have fixed (as long as electronics is not reset) time latencies between detectors. Local level processing not affected Local level processing changes time of the event Fixed latencies calibrated at the global level (TrackingFilter) and compensated in PostPsaFilter Integer part of event position inside captured trace is absorbed into timestamp in PostPSAFilter

Time Alignment example with 9 detectors
Peak Positions i \ j 1 2 3 4 5 6 7 8 501.20 477.18 485.80 503.14 482.81 495.61 487.06 481.42 498.83 476.46 482.24 502.04 483.06 495.15 489.47 480.16 522.99 523.68 506.89 524.30 506.82 516.20 506.01 503.37 514.21 517.78 493.41 517.00 498.45 507.93 503.30 498.88 497.32 498.99 476.36 483.21 479.85 488.34 483.19 479.90 517.95 516.96 493.77 501.57 520.27 508.58 502.35 496.64 505.08 506.93 484.54 492.18 511.97 492.17 494.80 489.09 514.30 514.10 494.58 497.26 520.19 497.88 505.35 495.66 519.41 521.74 502.11 520.42 503.65 511.01 504.60 average 500.34 stdev 13.29 T0 None corr chi2[i] 3R 0.86 -23.16 -14.54 2.80 -17.53 -4.72 -13.28 -18.92 3G -1.51 -23.88 -18.10 1.70 -17.28 -5.19 -10.87 -20.18 3B 22.65 23.34 6.56 23.96 6.48 15.86 5.67 3.03 2R 13.87 17.44 -6.93 16.66 -1.89 7.59 2.96 -1.46 894.42 2G -3.02 -1.34 -23.98 -17.13 -20.49 -12.00 -17.15 -20.44 2B 17.61 16.62 -6.57 1.23 19.93 8.24 2.01 -3.70 1R 4.75 6.60 -15.80 -8.16 11.63 -8.17 -5.54 -11.25 741.33 1G 13.96 13.76 -5.75 -3.08 19.85 -2.46 5.01 -4.68 874.05 1B 19.07 21.40 1.77 20.08 3.31 10.67 4.26 chi2[j] 955.91 709.31 682.44 0.00 Re-first -0.65 -0.51 -0.67 -0.22 0.08 0.03 0.68 0.15 1.67 1.51 0.28 -2.72 0.19 1.84 1.07 4.60 0.40 33.33 -22.65 -0.82 -2.22 -1.71 1.44 -2.04 -0.55 24.19 -13.87 2.06 1.85 -0.23 -1.53 3.05 3.74 36.77 3.02 0.17 1.69 -0.24 0.14 -4.23 -0.17 1.65 23.58 -17.61 -2.50 -2.51 -0.70 -4.62 -1.64 -2.24 44.40 -4.75 0.34 2.10 0.96 3.86 4.69 3.67 3.07 65.31 -13.96 2.94 -3.17 2.87 1.19 -4.20 0.43 49.08 -19.07 0.82 0.50 -3.43 -2.01 -3.65 -0.85 34.19 15.33 22.23 41.89 30.75 35.76 77.78 56.93 31.83 312.51 Analytical 9.77 -0.15 -1.35 0.77 -0.33 2.19 0.30 -0.53 9.81 10.78 -0.50 -1.06 -1.78 -0.42 -0.20 2.73 3.72 -0.78 26.61 -12.04 0.84 0.52 0.06 -0.98 0.74 -2.56 -0.39 10.11 -1.44 1.12 -0.43 -1.13 -0.81 1.62 12.73 12.9 0.11 0.78 1.31 -1.29 -1.96 -0.44 1.08 -6.3 1.54 -0.46 -0.83 0.47 0.73 -0.92 -0.48 -1.38 7.00 2.86 -2.16 -1.32 -0.90 0.24 1.59 0.99 1.13 0.23 12.16 -3.81 0.38 2.48 3.14 -1.66 0.13 21.44 -8.62 2.00 -1.31 10.31 10.70 8.82 11.77 9.31 19.42 7.60 21.87 23.97 6.81 120.28 1.30 None Relative to one of the detectors Best Fit

Time Alignment Time shift (Si) each detector so as to minimize the distance of the TAC peaks for all pairs Pij-Pji Matrix is singular  system solved by SVD Coefficients determined up to a constant (fixed by Si=0) Use RecalEnergy to fit position of time peaks Use solveTT.py to determine Si Example for 4 detectors

Global time alignment GSI 2014 SourceRun run_0005 60Co
FTHM of common time spectrum before alignment 44 ns after alignment 34 ns (26 ns when gating on EB detector at 1332 keV and full range in the AGATA crystal)

Hands On SourceRun to determine the performance of AGATA for the GSI campaign: 60Co, 152Eu, 133Ba, 166Ho, 56Co use an external germanium detector (Euroball) as trigger run_ Co 21 AGATA Crystals, 1 EB detector Calorimetric efficiency using the sum peak of the 2 60Co lines

EGAN Workshop, GSI, 2014

23 crystals (seen from target)
AGATA layout at GSI X Y Z 23 crystals (seen from target) 21 01C is an Single detector used for triggering

Files in the example directory daq@galileo01:/mnt/gv-galileo/EGAN-School2014/AgataSource_run_0005
Conf/ configurations, read during initialization Data/ part of original data and full PSA hits for run_0005 Out/ created by gen_conf.py ADL/ calculated signal basis for A001 B001 C usually this is a centralized repository ADF.conf definition of adf frames used for this analysis gen_conf.py TopologyProd.conf TopologyPrep.conf TopologyPSA.conf TopologyGlobal.conf using Data/*/psa_0000.adf

Efficiency from Cascades of 2 transitions
Each transition can be: detected fully eP  P detected partly eB  B detected eT = eP + eB  T not detected eV = 1- eT  V (eT = eP / PT) P1 B1 V1 P2 B2 V2 [P1 B1 V1] [P2 B2 V2] P1 P2 P1 B2 P1 V2 B1 P2 B1 B2 B1 V2 V1 P2 V1 B2 V1 V2 Peaks in Sum Spectrum of 60Co Which Count Energy Order P1 V e1 V1 P e1 P1 P e2 R2 = Area(P1P2)/Area(P1) = P1P2/P1V2 = P2/V2 = P2/(1-T2) = P2/(1-P2/PT2) eP2 = R2/(1+R2/PT2) The approximation R2 ≈ eP2 is only valid if the total efficiency is very small. For AGATA at GSI eT~ 10%  must use full expression. Should also correct for angular correlation and random coincidences.

Dino Bazzacco INFN Padova

Similar presentations

Presentation on theme: "Dino Bazzacco INFN Padova"— Presentation transcript:

Similar presentations

About project

Feedback

Log in

Auth with social network:

Dino Bazzacco INFN Padova

Similar presentations

Presentation on theme: "Dino Bazzacco INFN Padova"— Presentation transcript:

Similar presentations

About project

Feedback