2.3 Design Requirements
23
to configure a predictable order in which pad arrives before another is beneficial. As
the mapping of which pads are connected to which channel is not the same in all
positions of the detector, there would be a need for changing the predefined ordering
also.
2.3.4.4 Interfaces
The GBTx provides 10 serial uplinks at 320 Mbps when Front-End Card (FEC) is
applied, or 14 links when operating in the Wide Bus Mode where it is disabled. With
FEC, several errors in a packet can be recovered in case of corruption during transmission due to either noise or radiation. Using two GBTx chips provides a maximum
uplink bandwidth of 6.4 Gbps or 8.96 Gbps without forward error correction. Limiting to two GBTx chips per board will be done due to budgetary reasons. The new
front-end cards will replace the existing ones, which each connects to 160 front-end
channels and each front-end card will generate 16 Gbps of data. To fit all the data in
the allotted bandwidth, the compression with lossless encoding must be better than
2.5 and with zero suppression the occupancy must be lower than 40%.
2.3.5 TPC Specific Requirements, Revised Architecture
A well-known issue with GEM detectors is that they possess a common-mode effect,
which results in additional noise contributions if proper corrective baseline restoration has not been done. The effect results from the capacitive coupling between the
amplification structure and the readout pads. When an avalanche occurs in the GEM
foils, the current introduces a voltage drop on the GEM electrodes leading to induction of a correlated common-mode signal with opposite polarity on all anode pads
facing the GEM electrode [23].
The designed filtering structures present on the SAMPA were only found to partially restore the performance [23]. Better performance can be achieved if information
from a larger number of pads can be correlated, this correction would though need
to be done on the CRU with raw uncompressed data.
2.3.5.1 Direct Readout Serialization
The SAMPA has the possibility to send raw samples or lossless encoded samples in
the packet-based mode, but to be able to perform the common-mode correction on
the CRU the data from all channels need to be in parallel. As the SAMPA sends one
packet after another, this would require the need for a large amount of buffering on
the CRU. A CRU can be connected to up to 1600 channels, which if no compression
is used would require 16 Mbit of buffer memory. With lossless compression the size
could be reduced, but would instead require extra logic for the decoding.
Précédent

- 42/173

Suivant