Level1 Calorimeter Trigger Timing Issues 5 th July

Level-1 Calorimeter Trigger Timing Issues 5 th July, 2007 o Architecture o Detector-Like aspects o Trigger-Like aspects o Summary Level-1 Calorimeter Trigger Timing 1

Architecture Reminder L 1 Calo Latency (determines ATLAS latency? ) Digitized Energies Analogue Calorimeter signals (>7000) Preprocessor 124 modules Real-time Data Path Readout Data Cluster Processor 56 modules Merging 8 modules Jet/Energy Processor 32 modules Merging 4 modules Readout Driver (ROD) 14 modules Merged Results To CTP Region of Interest ROD 6 modules Region of Interest Data 5 th July, 2007 Level-1 Calorimeter Trigger Timing 2

General Comments o Many aspects similar to detectors: o Readout buffers o o Relatively large data size (multiple RODs) TTC system ‘client’ Some aspects different o Provide CTP input o o o Requires pipeline synchronization Many different buffers at different processing points Timing of calorimeter inputs varies: o 5 th July, 2007 Require timing, BCNum synchronization etc Beam, cosmics, calibration Level-1 Calorimeter Trigger Timing 3

Treatment of Timing Signals o Use standard TTCvi to TTCrx distribution o o L 1 A, BCR, ECR Treatment of BCR and L 1 A varies by module type o Processor modules count BCNum independently of TTCrx o o ROD uses TTCrx internal counter o o o 5 th July, 2007 Requires ORBIT signal to come before first bunch Tuned via single parameter in TTCvi BGo 0 setup Check is made for consistency in ROD o o Stable fudge factor to correct to real value and internally in some modules BCNumber reported in 0 -3563 range Level-1 Calorimeter Trigger Timing 4

Generic Readout Buffers o Buffers at least 128 deep o o o Read pointer adjustable to any depth wrt write pointer o Data corresponding to L 1 A Usually 1 -5 timeslices Up to 128 for input FADC data NB large depth probably unnecessary for L 1 Calo! o o o Read Pointer Readout window adjustable o o ie 3. 2 µs But useful for diagnostics VME readable for non-ROD diagnostics Pointer settings vary by module o o o Individual settings per module (>1) Different at different processing stages But broadly similar per module/buffer type 5 th July, 2007 Write Pointer Level-1 Calorimeter Trigger Timing 5

Readout Capabilities o ROD (and other modules) tested to full specification o o o More than 100 k. Hz instantaneous rate L 1 As separated by down to 5 ticks No restrictions beyond current CTP dead-time algorithms No other special requirements (ECRs, long gap etc) Bandwidth limitation with large number of slices o eg at 100 k. Hz o CPM cannot run with 5 slices FADC slices limited to 8 or 9 o More than 5 slices requires extra CTP deadtime after L 1 A o o o Wide readout foreseen only for timing in and initial running ECR mechanism implemented and tested stand-alone o o ROD is the only module that is ECR and L 1 ID aware Parasitically used to clear out bad events o 5 th July, 2007 Assumes 1 ms ECR isolation from L 1 A, as advertised Level-1 Calorimeter Trigger Timing 6

Event Counting o Nothing currently implemented beyond L 1 ID o o Rely on BCNum to spot missing events o o o And that’s only performed in ROD BCNum Formed in every module Internal problems flagged Subsystem-wide problem needs flagging downstream Testbeam experience showed BCNum comparison useful Could conceivably add event counting/timing event support o o Almost everywhere But requires firmware development and testing o o 5 th July, 2007 We’re a bit busy right now! Significant development if implemented in modules other than ROD Level-1 Calorimeter Trigger Timing 7

Timing in: Readout and Trigger o Two completely different processes o Timing in readout o o Getting the right readout pointers Multiples of 25 ns BC interval In principle relatively easy, but dependant upon first: Timing in trigger o o Far more difficult Each of 7000+ input channels requires individual timing o o o Input latency depends on cable lengths, detector position etc o o o 5 th July, 2007 At sub-25 ns level to optimize resolution At 25 ns level to synchronize trigger input to algorithms Varies by eg ~30 m cable length Probably up to 10 BC intervals Dependant on type of data (beam, cosmics etc) Level-1 Calorimeter Trigger Timing 8

Timing for different run types o Beam collisions o o Hope to have approximate timings before first beam Will require adjustments for best timing Widely spaced bunches will aid initial calibration Calibration runs o o Very useful for initial energy calibration At least indicative of true timing o o Cosmic runs o o 5 th July, 2007 Possible to calculate beam vs calibration residuals? Obtain initial timing from calorimeter calibration runs Timing again only indicative of beam timings o Muon trajectories cf beam collisions introduce 1 -2 BC shifts o But Tile. Cal laser pulser runs look interesting o Depends on calorimeter policy Signal size and rate makes individual tower calibration difficult L 1 A timing policy for cosmics: correct for muon paths? Level-1 Calorimeter Trigger Timing 9

Recursive Timing Problem: o 1) 2) 3) Approximately synchronize all triggers to latest input Set up readout pointers for all of ATLAS Discover a late L 1 Calo trigger tower 4) Go back to step 1). Remember L 1 Calo determines ATLAS latency o If this happens too often, L 1 Calo be very unpopular! o Solution: o Try to set up timing as soon as possible with final latency o o Establish ‘final’ ORBIT timing too (ORBIT in relation to BCNumber) Next Problem: We don’t know our latency o o 5 th July, 2007 ie delay all current inputs to expected latest trigger input timing Include safety margin (if there is room) Must measure it soon Ideally use L 1 Calo trigger and establish timing in M 4 Level-1 Calorimeter Trigger Timing 10

LHC Clock Phase Drift o We don’t like it o Phase drift of 3 ns from nominal is unpleasant for us o o o Must be worse for more precise detectors Loss of resolution Possible loss of small signals o More than 3 ns may require recalibration o Clock drift should be carefully monitored o Isn’t there a better way than a long fibre? 5 th July, 2007 Level-1 Calorimeter Trigger Timing 11

Summary o As a detector, we’re quite conventional o o o And reasonably well prepared No special requirements Ready for BCNum, ECR etc o As a trigger, we have much work to do o First goals: o Establish approximate timings using: o o 5 th July, 2007 Detector calibration systems Cosmic runs Time trigger output in CTP and measure latency Read out triggered cosmics in our data and calorimeters Level-1 Calorimeter Trigger Timing 12