Speaker
Description
Summary
The hadronic Tile Calorimeter of ATLAS (TileCal) is a sampling
calorimeter made out
of iron as absorber material and plastic scintillating plates as
active medium. The
light produced in the scintillating tiles is collected through
wavelength shifting
fibers and read-out by photomultipliers. The analogue
electrical signal of the
photomultipliers is digitized in 10-bit samples taken each 25
ns which is the latency
of an LHC bunch crossing. Therefore, the ~10000
photomultipliers of the calorimeter
produce up to ~164 Gbps of data. These data are read-out
and processed by the Read
Out Driver boards (RODs) of the TileCal back-end electronics.
The RODs are equipped with the TMS320C6414 Digital Signal
Processors (DSPs) of Texas
Instruments in order to execute energy reconstruction
algorithms in real time. The
aim of this is to provide information about energy deposition
in the calorimeter to
the second level trigger. The latency of the second level
trigger is ~10 microsec
thus the main requirements of the algorithm are the
processing time and the precision
on the reconstructed energy.
Besides the energy reconstruction algorithm the ROD
implements additional algorithms
to provide useful information to the second level trigger. We
propose in this paper a
fast algorithm to tag muons of low transverse momentum
(pT) inside the calorimeter.
The front-end electronics of TileCal is designed to produce
pulses with an amplitude
proportional to the energy deposited in the active medium.
The algorithm that we
propose for the energy reconstruction is the so called
Optimal Filtering (OF). OF
reconstructs the amplitude of the photomultiplier signal by
means of a weighted sum
of the digital samples. The weights are obtained from the
pulse shape of the
photomultipliers and the noise autocorrelation matrix. The
process to calculate them
minimizes the effect of the noise in the amplitude
reconstruction. Optimal Filtering
also reconstructs other magnitudes related to the signal
such as timing and pedestal
through different weighted sums of the samples.
Additionally to Optimal Filtering, a fast muon tagging
algorithm,the MuTag algorithm,
is also implemented. The aim of the MuTag algorithm is to
make up for the lack of
efficiency in the standalone muon spectrometer trigger for
soft muons (pT < 5 GeV/c)
which are characteristic of some interesting B-physics
channels. The MuTag algorithm
seeks energy deposition patterns inside a module of the
calorimeter using the energy
previously reconstructed with Optimal Filtering.
This paper explains the performance of both algorithms in
the ROD DSPs. The
TMS320C6414 DSP is a real time fixed-point processor which
performs up to 32-bit data
Multiply and Accumulate (MAC) instructions. We use low level
functions in oder to
reduce the processing time. We compare the processing
time obtained from the
simulation with the one measured in the laboratory. The
precision on the
reconstruction is also compared with the one obtained with
other algorithms, for
instance, the fit of the photomultiplier signal or the same
weighted sum calculated
in a floating-point processor. This paper also shows the
study of the processing time
of the MuTag algorithm as well as its efficiency and fake
rates calculated with Monte
Carlo data and its online performance with cosmic muons.