FUZZY PROCESSING FOR DATA REDUCTION OR ANALYSIS FROM SILICON DRIFT DETECTORS DETECTOR

Paper: 182
Session: A (poster)
Presenter: Petta, Catia, University of Catania
Keywords: analysis, neural networks, simulation, electronic data management, expert systems


FUZZY PROCESSING FOR DATA REDUCTION OR ANALYSIS FROM
SILICON DRIFT DETECTORS DETECTOR
C. Petta (1) , M. Russo (2), G. V. Russo (1); B. Batyunya (3), C. Caligiore
(4), A. Gabrielli) (5), E. Gandolfi (5), A. Insolia (1), D. Lo Presti (1), M.
Masetti (5), S. Panebianco (1), N. Randazzo (1), S. Reito (1) , A. Zinchenko(3)
(1) Dipartimento di Fisica dell'UniversitÓ and Sezione INFN, Catania, Italy
(2) Istituto di Ing. El. e Telecom. dell'UniversitÓ and Sezione INFN, Catania,
Italy
(3) Joint Institute for Nuclear Researchs, Dubna, Russia
(4) Centro Siciliano di Fisica Nucleare e Struttura della Materia and Sezione
INFN, Catania, Italy
(5) Dipartimento di Fisica dell'UniversitÓ and Sezione INFN, Bologna, Italy
Abstract
The ALICE's Inner Tracker System (ITS) Sikicon Drift Detectors (SDD's) data,
rightly processed, allows the reconstruction of the impact point coordinates
and the energy released by a particle crossing the detector. To obtain the
impact point reconstruction it is necessary to go back to the centre-of-mass of
a suitable set of data with a pseudo-gaussian shape. This goal is accomplished
by means of a proper fitting procedure. The energy is calculated with a simple
sum of a group of data. This is the traditional method of analysis. A new
procedure for ALICE's ITS SDD's is proposed.
We suggest the use. Instead of traditional analysis methods, Artificial
Intelligence techniques, as for instance Fuzzy Logic (FL). The spatial
resolution that we can obtain with the FL method is comparable with the
expected one obtained using traditional methods, but in absence of noise. When
noise is relevant, on the contrary, it is better FL methods because of its
substantial noise immunity.
An estimation of attainable precision using FL was obtained for ALICE's ITS's
SDDs. SHAKER code was used to simulate central collisions in Pb + Pb at 6.3
>TeV/N with density of charged particles dN/dy = 8000 at zero rapidity. The
results of these simulations was used with a special GEANT 3.21 based, ITS
simulatin program that I will send.
has produced the anodes signals from SDD. At the present only single tracks
from the central detector of the inner layer were taken into account. The
anodes data, together with original SHAKER data for position, energy loss,
direction of track and so on, were utilised to build a collection of 30,000
tracks. A portion of them were used to learn and to find rules using an
original genetic FuGeNeSys algorithm. The remaining part was used to cheeck
learned rules.
We got about 20 micron of spatial resolution for the impact point co-ordinates
and about 23 % for the total charge released in the SDD, in presence of the
noise foreseen with the front-end electronics for such detector. We must
highlight that these results can be achieved with a very low number of rules in
the range 4 - 12, according to the cases.
The main conclusion are:
1 - one can foresee that FL analysis of SDD's data can be achieved in time much
shorter than the traditional one, because of the low computational complexity
of FL, and since the limited number of rules;
2 - a substantial noise immunity is achieved without using special methods;
3 - the hypothesis of implementing an hardware system for ON-LINE data
reduction with a suitable dedicate Fuzzy Processor wich uses the rules found is
valid and promising. In this way one can avoid to transfer and store an
excessive amount of data. One can achieve a compression in data volume by a
factor of 400 with respect to raw data. Moreover detector cabling is simpler,
silicon area is smaller and power dissipation is lower. We are designing this
hardware implementation. The front-end amplifier and the Fuzzy Processor were
already realised producing satisfactory results. At the moment we are working
on the Analog Memory.