status of the agata psa
DESCRIPTION
Status of the AGATA PSA. For the PSA team, P. Désesquelles (IPN Orsay). [email protected]. PSA formalization (1). X =. One segment. One « Meta-signal » : hit segment+4(or 8) neighbors. 0 0 E 1 0 0 E 2 0. Energy deposit in a voxel. S 1. MGS. …. …. T. S =. …. T -1 ?. - PowerPoint PPT PresentationTRANSCRIPT
22 Feb. 2005 AGATA week in Darmstadt 1
Status of the AGATA PSA
For the PSA team, P. Désesquelles (IPN Orsay)
22 Feb. 05 AGATA week in Darmstadt 2
PSA formalization (1)
X =
0
0E1
0
0E2
0
T
T-1 ?
S =
……
…
S1
…
One segment One « Meta-signal » : hit segment+4(or 8) neighbors
Energy deposit in a voxel
MGS
22 Feb. 05 AGATA week in Darmstadt 3
PSA formalization (2)
T X = S :
10
0
X
S1T
S1
…
about 50voxels/segment
Each column = MGS signal
10 ns bins
22 Feb. 05 AGATA week in Darmstadt 4
Tasks
Number of hits Folding algo. (Milano/Munchen) not adapted. Smoothing/derivation (Orsay) not adapted. Derivation/data base (Milano) >65% (→ PSA meeting). Acclivity (Darmstadt) in progress. Neural networks (Orsay) in progress. Discriminant Analysis (Strasbourg/Orsay) next.
22 Feb. 05 AGATA week in Darmstadt 5
Tasks
Location and energy Neural networks (Orsay/Munchen) not adapted. Multivariate Analysis (Strasbourg) not adapted. Genetic algo. (Legnaro/ Darmstadt) too slow → coupled
with grid search (→ PSA meeting). Wavelets (Darmstadt) in progress (→ PSA meeting). Wavelets + grid descent (Orsay+Saclay) in progress (→ PSA
meeting). Matrix Inversion (Orsay+Strasbourg) in progress (→ PSA meeting).
22 Feb. 05 AGATA week in Darmstadt 6
Thus…
Difficulties with A.I. methods.
Exp. info. must be used in an optimum way.
Math. before algo.
22 Feb. 05 AGATA week in Darmstadt 7
Difficulties (1)
“Sensitivity” = How much S is changed for a given X shift
shift shift
very large sensitivity range very low sensitivity zones
22 Feb. 05 AGATA week in Darmstadt 8
Difficulties (2)
signals mainly sensitive to c.m. of energy deposits
G
23
ill conditioned transform
22 Feb. 05 AGATA week in Darmstadt 9
Difficulties (3)
Multi hits True noise The signal does not belong to the base distance between the hits relative energies neighbor segments whole detector number of hits unknown sampling rate time
Treat the realistic case :
22 Feb. 05 AGATA week in Darmstadt 10
Grid to choice
advantages drawbacks
r, cst. values of t10-90…
cylindrical
not homogenous
√r, cst. values of t10-90…
homogen., cylindr.
not the same x/y accuracy
x,y,z homogenous
simple
not cylindrical
large distances to grid
hexagon, z cylindrical
compact
not compact in z
not homogenous
hexagonal
compact
cylindrical
maximum compacity
less “standard”
Adaptated grid
optimum conditioning of the problem
not homogenous
(we work with the last one)
22 Feb. 05 AGATA week in Darmstadt 11
A grid adapted to the sensitivity
2 between grid points > 2 min
Condition number divided by 4 to 10
22 Feb. 05 AGATA week in Darmstadt 12
Sampling time
One hit in each of two neighboring segments
resolution is not worsen up to 150 ns bins !
22 Feb. 05 AGATA week in Darmstadt 13
Performances for one segment
Location : 0.3 mm ! (1 hit) 2 mm (simple multi-hit)
Energy : 1% (1 hit) some % (simple multi-hit)
Time : ~ ms (1 hit) 0.1 s (simple multi-hit on 2.4 GHz Matlab)
22 Feb. 05 AGATA week in Darmstadt 14
Conclusions
The single-isolated hit PSA is solved → neural networks The front-end can include :
Signals preprocessing Single-isolated hit PSA Tagging of events → which algo to use
The multi-hit PSA is difficult ! The X → S transform is not well conditioned Large sensitivity range Multi hits at the same r, Juge an algo on realistic case We should include numerical analysis specialists in
our group
22 Feb. 05 AGATA week in Darmstadt 15
Thank you