Ph. Busson (Ecole Polytechnique, Palaiseau)
W. Lustermann (ETH, Zurich)
T. Monteiro (CERN) 
J. C. Silva (LIP, Lisbon)
C. Tully (CERN and Princeton)
J. Varela (CERN and LIP, Lisbon)
!!
The CMS ECAL readout card is designed to handle 100 crystals.
Envisioned as a 9U VME module, the 1700 crystals, which form an ECAL
barrel supermodule, are handled by 17 such cards, effectively filling
a single VME crate. The data from these 17 cards is concatenated in 
each crate, on an event basis, by the Data Concentrator Card (DCC), 
and sent to the DAQ system.
The ECAL generates 2.4MBytes of data per Level-1 trigger (L1A). To 
reduce the data volume down to 120kBytes per L1A, as required by the 
central CMS DAQ system, a data reduction factor of 20 must be 
achieved. Since the actual number of ECAL crystals containing 
information of use in reconstructing any given triggered event is 
small, a substantial reduction in data volume is possible, without 
any loss of information meaningful to the physics.
Algorithmic data reduction in the ECAL is based on principles of 
calorimetric energy measurement. The energy of a showering particle 
is contained primarily within a compact array of crystals. Therefore, 
the energy reading in a compact array of crystals is a means to base 
a selection of crystals to be readout in an event.
The region of crystals to be readout needs to be large enough to 
achieve the limiting angular and energy resolution of the ECAL, to 
perform shower shape analyses for particle ID and to measure the 
proximity of neighboring showers or particles within jets. The 
readings of all neighboring crystals, at least down to the 
noise-levels of the electronics, are therefore equally important to 
physics analyses which use the ECAL data.
The baseline for the data reduction architecture uses the energy 
sums in trigger towers to generate readout information. An energy 
reading above a programmable threshold will force the readout of the 
channels in a 3x3 array of trigger towers centered on the trigger 
tower of interest. Complementary, zero suppression is applied on a 
channel basis with a threshold close to the noise level, 
independently of the selective readout algorithm.
The effectiveness of selectively reading out regions of the ECAL, as 
well as the impact on physics, was studied with a full-detector Monte 
Carlo simulation for several different physics triggers in the 
presence of pile-up. 
In the present design, the energy of each trigger tower is compared 
with two (programmable) thresholds, and a 2-bit status word is set. 
This calculation is executed as part of the trigger logic and runs 
synchronously at 40MHz. These bit pairs are transferred on a 
dedicated bus to the corresponding DCC. The DCC acts as a local 
"reflector", receiving the 2-bit words from boards within the crate 
and from neighboring DCC's, and transferring back to the boards the 
results of the tower selective readout decision. By utilizing the CMS 
Level-1 processing latency to exchange this SR information, readout 
may commence immediately upon the receipt of a L1A signal. 
Another possibility, currently under investigation, is to implement 
a dedicated Centralised Selective Readout Processor (CSRP) in order 
to facilitate the information exchange across the hardware 
boundaries. The DCC's will identify the seed regions and communicate 
them to the CSRP where a map is created of the channels to be read. 
This information is returned to the DCC's where the selective readout 
is performed.


!!
Algorithms suitable to reduce the volume of ECAL data passed
to the CMS DAQ system are investigated in terms of its physics
performance. Various implementation scenarios are analyzed which
demonstrate the feasibility of the selective readout techniques 
proposed.

!!