Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
David López Mateos, Harvard University, for the NSW trigger electronics group, nSW Design Review, February 13th, 2015
MicroMegas Trigger Processor Interface and Trigger Algorithms
‣ Interface to the detector
‣ Trigger algorithms
‣ Performance
‣ Trigger processor testing and integration
‣ Conclusions
nSW Design Review, February 2015 D. Lopez Mateos
Trigger Processor Interfaces
2
Level-1 Accepted events
algo params
TTC
BC clock
to Sector Logic
(2 streams of candidates)
6.4Gb/s each
ROD
Configurationprocessor
Monitor
processor
DCS
Sector
Logic
32 fibers(4 fibers per layer
8 layers)MM: 4.8G
sTGC: 5.28G
data tagged by BCID at source
FELIX
exceptionsstatisticssampled events
TTC
FPGA
configLL_TrigProcContext_V01
Busy
ATCA
carrier
Trigger
Processor
FPGA
on ATCA
mezzanine
card
Ethernet
ATCA
shelf mgr
IPMI
Ethernettemps
voltages
duplex fiberto FELIX
Note:
7 copies of each
stream are produced
NSW Trigger Processor context
E-links
nSW Design Review, February 2015 D. Lopez Mateos
Interfaces to the Detector
3
Level-1 Accepted events
algo params
TTC
BC clock
to Sector Logic
(2 streams of candidates)
6.4Gb/s each
ROD
Configurationprocessor
Monitor
processor
DCS
Sector
Logic
32 fibers(4 fibers per layer
8 layers)MM: 4.8G
sTGC: 5.28G
data tagged by BCID at source
FELIX
exceptionsstatisticssampled events
TTC
FPGA
configLL_TrigProcContext_V01
Busy
ATCA
carrier
Trigger
Processor
FPGA
on ATCA
mezzanine
card
Ethernet
ATCA
shelf mgr
IPMI
Ethernettemps
voltages
duplex fiberto FELIX
Note:
7 copies of each
stream are produced
NSW Trigger Processor context
E-links ‣ 64 strips per VMM
‣ Up to 1 trigger hit per VMM (earliest arrival hit)
‣ 32 VMMs per ADDC
‣ Up to 8 hits per ADDC per BC (1/4 of the sector)
‣ 32 fibers and ADDCs per sector
‣ 1 trigger processor per sector
nSW Design Review, February 2015 D. Lopez Mateos
Data Format
4
‣ Data transmitted from the ADDC to the trigger processor using the GBT architecture and protocol in a low-latency widebus mode (4.8 Gb/s, no forward error correction)
‣ Data packet of 112 bits provided every bunch crossing, will contain up to 8 hits per ADDC (32 VMMs), no encoding
‣ Two options considered:– VMM IDs for each trigger provided– Hit map provided (VMM ID encoding performed in the trigger processor
FPGA)
‣ The two options need to be implemented to understand performance
nSW Design Review, February 2015 D. Lopez Mateos
Data Format
5
‣ BCID: bunch crossing
‣ ART_DATA: strip number within one VMM that was hit
‣ ARTDATA_PARITY: parity of each ART_DATA 6-bit address (for error checking)
‣ HIT_CNT: Number of hits
‣ VMMID: Unique number for VMM that was hit for this ADDC
‣ HIT_LIST: single-bit list of hits (1 bit per VMM)
‣ ERR_FLAGS: Additional room for error reporting (e.g. more than 8 hits)
nSW Design Review, February 2015 D. Lopez Mateos
Fit-Based Trigger Algorithm
6
track
to IP
slope road
𝜽 (or 𝜽global)
𝛥𝜽
1. Translation of hardware addresses to global slopes, fixed to the IP
2. Determination of the presence of a multi-plane coincidence
3. Parallel calculation of average global slopes with horizontal and stereo strips, and local fit to 𝜽 with horizontal strips
4. Use of global and local slope from horizontal strips to calculate 𝛥𝜽, and global slopes from horizontal and stereo strips to calculate position (𝜽 and 𝝓)
➡ Used in most results discussed in what follows
nSW Design Review, February 2015 D. Lopez Mateos
Fit-Based Trigger Algorithm
7
‣ Incoming hits are converted to slope values using a look-up table, stored in a “finder” circular buffer
‣ Each slope road buffer is checked for coincidence every bunch crossing
‣ Upon coincidence, information is sent to the “fitters”
track
to IP
slope road
𝜽 (or 𝜽global)
𝛥𝜽
nSW Design Review, February 2015 D. Lopez Mateos
Fit-Based Trigger Algorithm
8
‣ Fitters calculate in parallel:– Average global slope (𝜽 position at entrance of spectrometer, 𝜽 w.r.t IP)
– Average local slope (𝜽 direction at entrance of spectrometer)
– Stereo slopes (for 𝝓 calculation)‣ Filter stereo planes if there are inconsistencies
track
to IP
slope road
𝜽 (or 𝜽global)
𝛥𝜽
nSW Design Review, February 2015 D. Lopez Mateos
Fit-Based Trigger Algorithm
9
‣ Use 𝜽 with respect to the IP and 𝜽 at the entrance of the spectrometer to
calculate 𝛥𝜽
‣ Use horizontal and stereo positions to calculate cartesian position, which gets turned into 𝜽, 𝝓 using a look up table
track
to IP
slope road
𝜽 (or 𝜽global)
𝛥𝜽
Trigger Processor
Fitter
Calc ROICalc Mu Global(u plane slopes)
ADDC Packet DecodeCalc strip Number
Calc strip Slope(one per fiber)
FinderCollect hits in slope road
(One per “region”)
ADDC ART DataTrack Data
Track StrobeHit Data
Hit Data includes:Slope,Strip#,BXID
For each hitHit DataAHit DataB
Calc Mx Global(x plane slopes)
Mx Global
Mu Global
Track DataTrack Strobe
Track DataTrack Strobe
Calc Mv Global(v plane slopes)
Mv GlobalTrack DataTrack Strobe
Calc Mx Local
Mx LocallTrack DataTrack Strobe
Calc Delta Theta
ROI
Track DataMx GlobalMx Local
Mx GlobalMu GlobalMv Global
DTheta
Track data includes: Hit Data for each member
of a “found” track
3 Ticks 3 Ticks
2 Ticks
2 Ticks
3 Ticks
5 Ticks
4 Ticks5 Ticks
3 Ticks
Critical path total 18 clock ticks
Track data from many ‘Finders’ are multiplexed into
one ‘Fitter’
nSW Design Review, February 2015 D. Lopez Mateos
Fit-Based Trigger Algorithm
10
track
to IP
slope road
𝜽 (or 𝜽global)
𝛥𝜽
‣ 18 clock ticks, or about 56 ns‣ 70% of resources used in a Xilinx V7
485 chip‣ Desirable upgrade to pin-compatible
Xilinx V7 690 for additional resources
nSW Design Review, February 2015 D. Lopez Mateos
Look-up Table Algorithm
11
1. Define 6 pairings of layers, use incoming hits to determine through a LUT slope of each pair
2. Slopes compared to each other and track candidate selected if enough pairs are consistent
3. Slopes averaged to obtain 𝜽 direction
of the track and similarly to obtain 𝜽 position
4. 𝝓 position calculated similarly to the previous algorithm
➡ Steps 1-3 timed to 59 ns using an FPGA implementation
1" 5" 2" 6" 3" 7" 4" 8"
Layer"Pair"Track"Candidate"
X"plane"U"plane"V"plane"
1" 6" 2" 5"
X0" X1" X2" X3"U" V"1 5 2 6 3 7 4 8 1 6 2 5
6.7$cm$
9.9$cm$
15.3$cm$
1$ 2$ 3$ 4$ 5$ 6$ 7$ 8$MM$$Layers$
Mul7plet$ Mul7plet$
18.5$cm$
DPAIR=$12.6$cm$
BC1A%
BC0A%
Layer%A%BC%hits%Shi1%
Register%
BC2@40%MHz%
BC1B%
BC0B%
Layer%B%BC%hits%Shi1%
Register%
BC2@40%MHz%
BC1A:BC1B%BC1A:BC0B%BC0A:BC1B%BC0A:BC0B%
READ%@320%MHz%
SLOPE%#0%
SLOPE%SELECT%
Hit%A%
Hit%B%
SLOPE%#63%
Hit%A%
Hit%B%
64%Slope%
CalculaHon%&%
Pre:selecHon%%
MAX%8%Hits/BC%
MATCH/SLOPE%STORAGE%
Shi1%Registers%Layer%A/Layer%B%RAW%DATA%
x%Number%of%Layer%Pairs%(6)% Slope%Select%Logic%
ROI%&%Track%Angle%%
Selected track(s)
4x31%Match%slope%bit%
BC1A:BC1B%BC1A:BC0B%BC0A:BC1B%BC0A:BC0B%
4x64%slopes%
Mi:%Match%slope%value%i%
M1% M31%
Slope%#0%
Slope%#255%
M1:M31% M1:M31%6%Layer%Pairs%
M1:M31% M1:M31%
4%BCA:BCB%combinaHons%
Mor%
OR% OR%
AND% Mor%
Selected%Slopes%
Data%ID%
Data%pointers%
Data%access%
3%Hcks% A%5%(Storage)%+%4%(Accumulate)%Hcks%
B%
1%Hcks%+1%per%track%
4%Hcks%
2%Hcks%
19%Hcks@320%MHz%A% B%
59,4%ns%From%% A%to% B%
Designed%(VHDL)%&%Simulated%XST%Synthesis%&%P&R%Virtex:7%xc7v485t%:2%speed%grade%
Slice%Registers%%%%%9%%Slice%LUTs%%%%%%%%%%%%14%%%%%%%%logic%%%%%%%%%%%%%%%%10%%%%%%%%shi1%Register%1%%%%%%%%route:thrus%%%%3%%Occupied%Slices%%%18%%%%%%%%%%%%%%%For%3%BCs%27%%%
Tunable%Logic%+/:%2%slope%bins%
x4%PR%
Match%Reset%
A
B
A B A B
nSW Design Review, February 2015 D. Lopez Mateos
Look-up Table Algorithm
12
truCT
2X,1UV 2X,2UV 3X,2UV 3X,3UV 4X,4UV
pass
/nfitn
0.97
0.98
0.99
1
1.01ηE=200 GeV, 2X,1UV, all
CT
2X,1UV 2X,2UV 3X,2UV 3X,3UV 4X,4UV
pass
/nfitn
0.92
0.94
0.96
0.98
1ηE=200 GeV, 2X,1UV, all
CT
2X,1UV 2X,2UV 3X,2UV 3X,3UV 4X,4UV
ent
/nfitn
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1 ηE=200 GeV, all
nSW Design Review, February 2015 D. Lopez Mateos
Ideal Efficiencies : Fit-based
13
‣ What fraction of events with some truth-level hits are fit with different coincidences?➡ Purely algorithmic‣ For all particles hitting the NSW, what fraction is fit with at least N number of hits?➡ Includes detector effects‣ For all particles creating at least 6 hits, what fraction are reconstructed with at least
N number of hits?➡Includes detector effects but only for “good” tracks, similar in LUT algorithm
‣ Simulations suggest 2X,2UV mode is background-resilient, but more realistic simulation required to establish operation mode
4(X,X)&+&(&1(U,U)&A
ND&1(V,V)&)&
3(X,X)&+&(&1(U,U)&A
ND&1(V,V)&)&
2(X,X)&+&(&1(U,U)&A
ND&1(V,V)&)&
4(X,X)&+&(&1(U,U)&O
R&1(V,V)&)&
3(X,X)&+&(&1(U,U)&O
R&1(V,V)&)&
1(X,X)&+&(&1(U,U)&O
R&1(V,V)&)&
2(X,X)&+&(&1(U,U)&O
R&1(V,V)&)&
nSW Design Review, February 2015 D. Lopez Mateos
Ideal Efficiencies: Look-up Table
14
‣ Inefficiencies at regions between modules caused by different look up tables used for different sector modules (need to devise way of recovering them)
‣ No dependence of efficiency on pT (but some loss from secondary showers expected at very high pT, from studies with fitter algorithm and E=1 TeV)
‣ Higher efficiency obtained for loser coincidence requirements
nSW Design Review, February 2015 D. Lopez Mateos
𝛥𝜽 Distributions
15
[rad]fitθ∆-15 -10 -5 0 5 10 15
-310×
track
s
0
2
4
6
8
10
12
14
16
18 rms=0.00432201E=50 GeV, 4X4UV
[rad]fitθ∆-15 -10 -5 0 5 10 15
-310×tra
cks
0
100
200
300
400
500
600
rms=0.00312018E=200 GeV, 4X4UV
[rad]fitθ∆-15 -10 -5 0 5 10 15
-310×
track
s
0
10
20
30
40
50
60
70
80rms=0.00318164
E=1000 GeV, 4X4UV
pT(𝜼=1.3)=25 GeV
pT(𝜼=2.7)=7 GeVpT(𝜼=1.3)=100 GeV
pT(𝜼=2.7)=27 GeV
pT(𝜼=1.3)=500 GeV
pT(𝜼=2.7)=135 GeV
‣ These include effects on the track upstream of the nSW
‣ Helpful to understand range of values necessary
‣ ±15 mrad seems enough to cover vast majority of events
‣ More systematic study (vs pT and 𝜼) necessary
[rad]truθ∆-fitθ∆-20 -15 -10 -5 0 5 10 15 20
-310×
norm
. are
a
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
mean=4.35e-05rms=0.00272fit mean=8.85e-06fit rms=0.00157
E=200 GeV, 4X4UV
ηall
[rad]truθ-fitθ-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-310×
norm
. are
a
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16mean=8.68e-05rms=0.000329fit mean=9.15e-05fit rms=0.000272
E=200 GeV, 4X4UV
ηall
[rad]truφ-
fitφ
-40 -30 -20 -10 0 10 20 30 40-310×
norm
. are
a
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18 mean=0.000715rms=0.00402fit mean=0.000382fit rms=0.00232
E=200 GeV, 4X4UV
ηall
nSW Design Review, February 2015 D. Lopez Mateos
Ideal Resolution
16
‣ Resolution in 𝝓 of about 2.3 mrad, degraded significantly (to 5 mrad) when adding incoherent background
‣ 𝜽 position resolution is very high, 0.3 mrad, not affected significantly when adding incoherent background
‣ 𝛥𝜽 resolution of about 1.6 mrad, degrades to 1.7 mrad in the presence of incoherent background, confirmed with the Look-up Table algorithm
nSW Design Review, February 2015 D. Lopez Mateos
Handling Misalignments
17
‣ Five types of misalignments studied with the fit-based algorithm
‣ a, b, d and e can be corrected with extra storage of constants in step 1 (no extra CPU)
‣ c requires some additional study of necessary resources (not done yet)
‣ Missing one rotation in these studies, to be studied in the near future
nSW Design Review, February 2015 D. Lopez Mateos
Technically Handling Misalignments
18
‣ Shifts along radial direction can be corrected in original translation of addresses to slope values
nSW Design Review, February 2015 D. Lopez Mateos
Technically Handling Misalignments
19
‣ z-shifts need also updates in constants necessary for local slope
‣ Tilts can be fixed in the same way (16 constants per layer)
nSW Design Review, February 2015 D. Lopez Mateos
Effects of Misalignments
20
‣ Effects of misalignments studied corresponding to up to 5 mm mistakes in relative placement of detector planes
‣ Impact on efficiency is negligible (but biases in 𝜽, 𝝓 can impact matching to Big Wheel)
‣ Impact on theta and phi position resolution is small for all misalignments (never bigger than 4%)
‣ For 𝛥𝜽, degradation can be large, up to 80% for the rotation for which no correction has been studied yet (making 𝛥𝜽 of about 3 mrad)
[mrad]backφ ∆
0 0.15 0.3 0.45 0.6 0.75 0.9 1.05 1.2 1.35 1.5
no m
isal
σ/truθ
∆-fitθ
∆σ
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9E=200 GeV, 4X4UV
ηall
nSW Design Review, February 2015 D. Lopez Mateos
Current Test Set-up
21
‣ Work on MM trigger processor firmware using evaluation board with Xilinx Virtex 7 485T and ADDC emulator
‣ Using same ART data as used for simulations, and an implementation of the GBT deserialization in the FPGA
‣ Currently working on the integration of the BNL ADDC with an ART pattern generator
ART Pattern Generator(VC707 Development Board)
Trigger Processor(VC707 Development Board)
ADDC ART Data(GBT Fiber)
Ethernet
BNL ADDCVMM ART Data(LVDS Mini SAS)
ART Pattern Generator(VC707 Development Board)
Trigger Processor(VC707 Development Board)
ADDC ART Data(GBT Fiber)
Ethernet
BNL ADDCVMM ART Data(LVDS Mini SAS)
nSW Design Review, February 2015 D. Lopez Mateos
Current Test Set-up
22
‣ Mezzanine card developed to connect FMC output connectors from development board to MiniSAS in ADDC card
ART pattern generator
!"#$#%""&'(")**""
+"#$#%,"&'("#-."
""""""""""
#-."/"
012'"34'56"
72899"#$#%""""
7:(1&" 012'"
7'(1891;'("
<"
<"
012'"34'56"
72899"#$#%""""
7:(1&" 012'"
7'(1891;'("
<"
<"
012'"34'56"
72899"#$#%""""
7:(1&" 012'"
7'(1891;'("
<"
<"
<<"
*1=1"7>7":?">@@3,"
""""""""""
#-."A"
012'"34'56"
72899"#$#%""""
7:(1&" 012'"
7'(1891;'("
<"
<"
012'"34'56"
72899"#$#%""""
7:(1&" 012'"
7'(1891;'("
<"
<"
012'"34'56"
72899"#$#%""""
7:(1&" 012'"
7'(1891;'("
<"
<"
<<"
*1=1"7>7":?">@@3,"
<<"
B/"C1:"5?D=:'("E"FB/"*G;"
F"
-:4'(=':" H'='(15"1=:'(I85'"
>JJ(',,"J'5?J'("
K3"
#KH>"
>:4'=8",12D98L?="
J8:8"
#?(28M'J"1=&D:"N9'""
O,:(1&"PQ"L2'Q""#$#%"8JJ(',,R"
@>S"$=:'(I85'"I?("5?22D=158L?="
nSW Design Review, February 2015 D. Lopez Mateos
Future Test Set-ups
23
‣ Cosmic ray set up at Harvard:– Provides muons of above 0.8 GeV with 1.5 ns time resolution– Not subject to test-beam schedules and allows for quick development iterations– Can integrate trigger testing boards easily since development happens on site
‣ MSW:– Test the full chain with prototype electronics at 40 MHz and low background levels– ART data recorded, can be later fed to pattern generator
‣ Vertical slice and test beams atCERN (see Joao’s talk)
nSW Design Review, February 2015 D. Lopez Mateos
Conclusions
24
‣ Good progress on definition of algorithms, and testing with evaluation boards
‣ Efficiency is very high and position resolution also
‣ 𝛥𝜽 resolution is closer to 2 mrad than 1 mrad (just from geometry considerations)
‣ Misalignments have a large impact on 𝛥𝜽 resolution, corrections for most types do not impact latency
‣ For certain rotations, corrections and impact on resources need to be studied
‣ Background effects are not too significant, but more realistic background simulation is highly desirable
‣ FPGA implementation is evolving with the algorithm development, but a large focus of the current effort is on integration and testing
‣ More sophisticated testing is planned as a natural evolution of current testing efforts
BACK-UP SLIDES
nSW Design Review, February 2015 D. Lopez Mateos
Trigger Processor and ADDC
26
nSW Design Review, February 2015 D. Lopez Mateos
Timing of Algorithm
27
Trigger Processor
Fitter
Calc ROICalc Mu Global(u plane slopes)
ADDC Packet DecodeCalc strip Number
Calc strip Slope(one per fiber)
FinderCollect hits in slope road
(One per “region”)
ADDC ART DataTrack Data
Track StrobeHit Data
Hit Data includes:Slope,Strip#,BXID
For each hitHit DataAHit DataB
Calc Mx Global(x plane slopes)
Mx Global
Mu Global
Track DataTrack Strobe
Track DataTrack Strobe
Calc Mv Global(v plane slopes)
Mv GlobalTrack DataTrack Strobe
Calc Mx Local
Mx LocallTrack DataTrack Strobe
Calc Delta Theta
ROI
Track DataMx GlobalMx Local
Mx GlobalMu GlobalMv Global
DTheta
Track data includes: Hit Data for each member
of a “found” track
3 Ticks 3 Ticks
2 Ticks
2 Ticks
3 Ticks
5 Ticks
4 Ticks5 Ticks
3 Ticks
Critical path total 18 clock ticks
Track data from many ‘Finders’ are multiplexed into
one ‘Fitter’
A : Incoming hits are converted to slope values
J : Calculate delta theta
K : Calculate cartesian
slopes
I: Filter background from stereo hits
E : Calculate local slope
F : Calculate global X slope
C : Check buffer for coincidence
B : Store hits in a buffer where address = (slope, plane, BC)cycles every BC, 2 BC deep
G : Calculate global U slope
H : Calculate global V slope
X strips X slopes U and V slopesV slopesU slopes
global U slope global V slope stereo validation
stereo global slopesglobal X slope
K : Lookup ,
(m_x,m_y)
D: Read Track from buffer and send components
Dtheta cut?
delta theta
local X slope
M: Abandon Fit
hit = (strip number, plane, slope)
slope road index to read
Incoming hit signals
L : Trigger Signal Output
local slope <0
delta theta too large
could cut if low
(not implemented)
block_detailed_v02
nSW Design Review, February 2015 D. Lopez Mateos
Misalignment Studies
28
nSW Design Review, February 2015 D. Lopez Mateos
Block Diagram
29
nSW Design Review, February 2015 D. Lopez Mateos
Slope Roads
30
nSW Design Review, February 2015 D. Lopez Mateos
Local Slope Calculation
31
θ
“Local” implies that the fit is not influenced by the IP
Note: Given 4 X-planes, there are 11 hit combinations, assuming N>1
Pre-calculated and stored in a reference table
nSW Design Review, February 2015 D. Lopez Mateos
𝛥𝜽 Calculation
32
θ θ0
We can now cut on delta_theta to filter out background tracks that are not originating from the vicinity of the IP
The error introduced by the approximation is <4%
nSW Design Review, February 2015 D. Lopez Mateos
Cartesian Slope Calculation
33
average
Note: Table (or mesh grid of cartesian slopes superimposed on detector) will be tuned to specified ROI resolution required
Reference table
ROI
A=csc(1.5°)B=cot(1.5°)
nSW Design Review, February 2015 D. Lopez Mateos
Look-up Table Algorithm
34
PR##1#
PR##2#
PR##3#
PR##0#
Z#
8#layers#11°#
30°#
1# 5#
Layer#Pair#
1#strip#
#Z#angle#aperture#(α<#θ)#
#
D0#
D2#D1#
DPAIR#
θ#
α#
D2=DPAIR(tanα#–#tanθ)##
1 5
MM""Layers"
Interac-on"Point"
z"
PR"#1"
Projec-ve"PR"#1"
PR"#1"
PR"#2"
PR"#3"
PR"#0"
)θ
)θ
SLOPE&SELECT&
#strip&Layer&&5&
1& 5&
Layer&Pair&
1&strip&
D2&
DPAIR&
#strip&Layer&&1&
Aperture&&&Offset&(common&to&a&Panel&Region)&
LUT&MATCH&SLOPE&
MATCH&
Hit&Offset&
1 5
nSW Design Review, February 2015 D. Lopez Mateos
Look-up Table Algorithm Performance
35
rms=1.7 mrad
nSW Design Review, February 2015 D. Lopez Mateos
Monitoring and Offline Information
36
Level-1 Accepted events
algo params
TTC
BC clock
to Sector Logic
(2 streams of candidates)
6.4Gb/s each
ROD
Configurationprocessor
Monitor
processor
DCS
Sector
Logic
32 fibers(4 fibers per layer
8 layers)MM: 4.8G
sTGC: 5.28G
data tagged by BCID at source
FELIX
exceptionsstatisticssampled events
TTC
FPGA
configLL_TrigProcContext_V01
Busy
ATCA
carrier
Trigger
Processor
FPGA
on ATCA
mezzanine
card
Ethernet
ATCA
shelf mgr
IPMI
Ethernettemps
voltages
duplex fiberto FELIX
Note:
7 copies of each
stream are produced
NSW Trigger Processor context
E-links
‣ ART data and processor results also sent to ROD upon L1 acceptance, buffer to keep this necessary for the L1 latency
‣ Additional ancillary functions implemented for monitoring purposes (see Lorne’s talk)
nSW Design Review, February 2015 D. Lopez Mateos
Configuration
37
Level-1 Accepted events
algo params
TTC
BC clock
to Sector Logic
(2 streams of candidates)
6.4Gb/s each
ROD
Configurationprocessor
Monitor
processor
DCS
Sector
Logic
32 fibers(4 fibers per layer
8 layers)MM: 4.8G
sTGC: 5.28G
data tagged by BCID at source
FELIX
exceptionsstatisticssampled events
TTC
FPGA
configLL_TrigProcContext_V01
Busy
ATCA
carrier
Trigger
Processor
FPGA
on ATCA
mezzanine
card
Ethernet
ATCA
shelf mgr
IPMI
Ethernettemps
voltages
duplex fiberto FELIX
Note:
7 copies of each
stream are produced
NSW Trigger Processor context
E-links
‣ Several configuration parameters for trigger processor and algorithm
‣ Configured through Ethernet or an e-link from FELIX
‣ More information in Lorne’s talk
nSW Design Review, February 2015 D. Lopez Mateos
Background Simulation
38
‣ Important work on background simulation ongoing (see D. Denys et al. talk)
‣ For performance studies, using non-coherent background extrapolated from Phase-1 LOI
‣ Other limitations exist in the simulation (e.g. implementation of neutron response in current and upgrade detectors)
5000
10000
15000
Hit
Rat
e [H
z/cm
2 ]
100 150 200 250 300 r[cm]
[*] ATL-U
PGR
AD
E-INT-2014-001
nSW Design Review, February 2015 D. Lopez Mateos
MM Trigger Algorithms
39
‣ Finders are dynamically allocated and pre-select track candidates
‣ Global slope calculated from average θ of horizontal strips, local slope from a fit
‣ Δθ comes from fitter, ROI from LUT using stereo strip global slopes
[*] Harvard, ATL-UPGRADE-INT-2014-001
[*] Saclay, see S. Hassani's talk for details
‣ Calculate all pair-wise slope combinations using LUT
‣ Compare these and their compatibility with IP to select track candidates
‣ Use output to calculate ROI and Δθ
nSW Design Review, February 2015 D. Lopez Mateos
MM Trigger Processor Emulation
40
MM Trigger processor
GBT
des
eria
lizer
Athena singlemuon events
‣ Work on MM trigger processor firmware using Xilinx Virtex 7 evaluation board
‣ Athena events used in performance studies also used to test firmware
‣ Latency studies described in detail in Joao’s talk
‣ Impact of misalignments on the latency need to be evaluated and will cause algorithmic modifications
nSW Design Review, February 2015 D. Lopez Mateos
MM Trigger Processor Emulation
41
ART Pattern Generator MM Trigger processor
GBT
GBT
des
eria
lizer
Athena singlemuon events
‣ Working on local integration at Harvard using an ART pattern generator (see T. Lazovich talk)
‣ Ready for testing as soon as ADDC is available‣ Will, in parallel, use an evaluation board to save test beam trigger data with ADDC