AtlasPublicTopicHeader.png

L0MuonTriggerPublicResults

Introduction

Approved plots that can be shown by ATLAS speakers at conferences and similar events. Please do not add figures on your own. Contact the responsible project leader in case of questions and/or suggestions. Follow the guidelines on the trigger public results page. (Chrome & Firefox on MacOS have a problem to show the plots and it works for Safari.)

Level-0 MUON Trigger with multiple systems

L0Muon trigger efficiency as a function of the truth muon pT (in the barrel region): ATL-COM-DAQ-2019-128 (14 Aug 2019)

Expected trigger efficiency for the Level-0 muon trigger with inclusions of MDT information (red) and without MDT (orange), compared with offline efficiency (green). The efficiency values relative to muons as trigger performance are obtained for muon tracks in the Large sectors in the barrel. The efficiency values are relative to a transverse momentum (pT) trigger threshold of 20 GeV. The values are obtained from single muon MC samples with no pile-up. Background level for the "Phase-II RPC" option is significantly higher than that for the "Phase-II RPC & MDT" option. .pdf
pdf


contact: Davide Cieri

Level-0 RPC Trigger

Convolutional Neural Network on FPGA for real time reconstruction of muons for the ATLAS Phase-II barrel trigger: ATL-COM-DAQ-2019-189 (17 Oct 2019)

An example input to a Convolutional Neural Network (CNN) for the Phase-2 ATLAS Level-0 muon trigger, implemented on a FPGA, is shown for a 13 GeV muon plus background. Resistive Plate Chambers (RPC) hits of a fixed $\phi$ sector are arranged in a matrix-like object. Each bin of the $y$-axis corresponds to a detector layer (3 detector layers for inner station, 4 for the middle and 2 for the outer station). The $x$-axis maps the $\eta$ coordinates of each physics RPC strip: for the $i$-th strip $\eta_{bin}^{i} = 384 \frac{\eta^{i}-\eta^{min}} {\eta^{max}-\eta^{min}}$, where $\eta^{max}$ and $\eta^{min}$ are respectively the maximum ($\eta^{max}=0.95$) and the minimum ($\eta^{min}=0.07$) $\eta$ values for the barrel RPC strips chosen to prevent muons to fall outside any layer of a specified sector; 384 is a realistic number of strips per layer. This particular choice has been taken in order to evaluate ML algorithm performances, without any geometrical acceptance effect. Random background has been added. The background rate has been evaluated from minimum bias events. Events used in the training phase of the CNN can also contain two or more muons in the same sector. Events with more than one muon are built superimposing one muon images with no background, which is then included. The CNN output is set to evaluate transverse momentum and $\eta$ of the leading and sub-leading muons (if the latter exists) in the sector and returns also a flag for events that contain more than 2 muons.

.pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
An example input to a Convolutional Neural Network (CNN) for the Phase-2 ATLAS Level-0 muon trigger, implemented on a FPGA, is shown for background only. Resistive Plate Chambers (RPC) hits of a fixed $\phi$ sector are arranged in a matrix-like object. Each bin of the $y$-axis corresponds to a detector layer (3 detector layers for inner station, 4 for the middle and 2 for the outer station). The $x$-axis maps the $\eta$ coordinates of each physics RPC strip: for the $i$-th strip $\eta_{bin}^{i} = 384 \frac{\eta^{i}-\eta^{min}} {\eta^{max}-\eta^{min}}$, where $\eta^{max}$ and $\eta^{min}$ are respectively the maximum ($\eta^{max}=0.95$) and the minimum ($\eta^{min}=0.07$) $\eta$ values for the barrel RPC strips chosen to prevent muons to fall outside any layer of a specified sector; 384 is a realistic number of strips per layer. This particular choice has been taken in order to evaluate ML algorithm performances, without any geometrical acceptance effect. Random background has been added. The background rate has been evaluated from minimum bias events. Events used in the training phase of the CNN can also contain two or more muons in the same sector. Events with more than one muon are built superimposing one muon images with no background, which is then included. The CNN output is set to evaluate transverse momentum and $\eta$ of the leading and sub-leading muons (if the latter exists) in the sector and returns also a flag for events that contain more than 2 muons. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
An example input to a Convolutional Neural Network (CNN) for the Phase-2 ATLAS Level-0 muon trigger, implemented on a FPGA, is shown for a 4 GeV muon without background. Resistive Plate Chambers (RPC) hits of a fixed $\phi$ sector are arranged in a matrix-like object. Each bin of the $y$-axis corresponds to a detector layer (3 detector layers for inner station, 4 for the middle and 2 for the outer station). The $x$-axis maps the $\eta$ coordinates of each physics RPC strip: for the $i$-th strip $\eta_{bin}^{i} = 384 \frac{\eta^{i}-\eta^{min}} {\eta^{max}-\eta^{min}}$, where $\eta^{max}$ and $\eta^{min}$ are respectively the maximum ($\eta^{max}=0.95$) and the minimum ($\eta^{min}=0.07$) $\eta$ values for the barrel RPC strips chosen to prevent muons to fall outside any layer of a specified sector; 384 is a realistic number of strips per layer. This particular choice has been taken in order to evaluate ML algorithm performances, without any geometrical acceptance effect. Random background has been added. The background rate has been evaluated from minimum bias events. Events used in the training phase of the CNN can also contain two or more muons in the same sector. Events with more than one muon are built superimposing one muon images with no background, which is then included. The CNN output is set to evaluate transverse momentum and $\eta$ of the leading and sub-leading muons (if the latter exists) in the sector and returns also a flag for events that contain more than 2 muons. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
An example input to a Convolutional Neural Network (CNN) for the Phase-2 ATLAS Level-0 muon trigger, implemented on a FPGA, is shown for a 12 GeV and 15 GeV muons without background. Resistive Plate Chambers (RPC) hits of a fixed $\phi$ sector are arranged in a matrix-like object. Each bin of the $y$-axis corresponds to a detector layer (3 detector layers for inner station, 4 for the middle and 2 for the outer station). The $x$-axis maps the $\eta$ coordinates of each physics RPC strip: for the $i$-th strip $\eta_{bin}^{i} = 384 \frac{\eta^{i}-\eta^{min}} {\eta^{max}-\eta^{min}}$, where $\eta^{max}$ and $\eta^{min}$ are respectively the maximum ($\eta^{max}=0.95$) and the minimum ($\eta^{min}=0.07$) $\eta$ values for the barrel RPC strips chosen to prevent muons to fall outside any layer of a specified sector; 384 is a realistic number of strips per layer. This particular choice has been taken in order to evaluate ML algorithm performances, without any geometrical acceptance effect. Random background has been added. The background rate has been evaluated from minimum bias events. Events used in the training phase of the CNN can also contain two or more muons in the same sector. Events with more than one muon are built superimposing one muon images with no background, which is then included. The CNN output is set to evaluate transverse momentum and $\eta$ of the leading and sub-leading muons (if the latter exists) in the sector and returns also a flag for events that contain more than 2 muons. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
An example input to a Convolutional Neural Network (CNN) for the Phase-2 ATLAS Level-0 muon trigger, implemented on a FPGA, is shown for a 13 GeV and 17 GeV muons with background. Resistive Plate Chambers (RPC) hits of a fixed $\phi$ sector are arranged in a matrix-like object. Each bin of the $y$-axis corresponds to a detector layer (3 detector layers for inner station, 4 for the middle and 2 for the outer station). The $x$-axis maps the $\eta$ coordinates of each physics RPC strip: for the $i$-th strip $\eta_{bin}^{i} = 384 \frac{\eta^{i}-\eta^{min}} {\eta^{max}-\eta^{min}}$, where $\eta^{max}$ and $\eta^{min}$ are respectively the maximum ($\eta^{max}=0.95$) and the minimum ($\eta^{min}=0.07$) $\eta$ values for the barrel RPC strips chosen to prevent muons to fall outside any layer of a specified sector; 384 is a realistic number of strips per layer. This particular choice has been taken in order to evaluate ML algorithm performances, without any geometrical acceptance effect. Random background has been added. The background rate has been evaluated from minimum bias events. Events used in the training phase of the CNN can also contain two or more muons in the same sector. Events with more than one muon are built superimposing one muon images with no background, which is then included. The CNN output is set to evaluate transverse momentum and $\eta$ of the leading and sub-leading muons (if the latter exists) in the sector and returns also a flag for events that contain more than 2 muons. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
An example input to a Convolutional Neural Network (CNN) for the Phase-2 ATLAS Level-0 muon trigger, implemented on a FPGA, is shown. This example contains three muons without background. Resistive Plate Chambers (RPC) hits of a fixed $\phi$ sector are arranged in a matrix-like object. Each bin of the $y$-axis corresponds to a detector layer (3 detector layers for inner station, 4 for the middle and 2 for the outer station). The $x$-axis maps the $\eta$ coordinates of each physics RPC strip: for the $i$-th strip $\eta_{bin}^{i} = 384 \frac{\eta^{i}-\eta^{min}} {\eta^{max}-\eta^{min}}$, where $\eta^{max}$ and $\eta^{min}$ are respectively the maximum ($\eta^{max}=0.95$) and the minimum ($\eta^{min}=0.07$) $\eta$ values for the barrel RPC strips chosen to prevent muons to fall outside any layer of a specified sector; 384 is a realistic number of strips per layer. This particular choice has been taken in order to evaluate ML algorithm performances, without any geometrical acceptance effect. Random background has been added. The background rate has been evaluated from minimum bias events. Events used in the training phase of the CNN can also contain two or more muons in the same sector. Events with more than one muon are built superimposing one muon images with no background, which is then included. The CNN output is set to evaluate transverse momentum and $\eta$ of the leading and sub-leading muons (if the latter exists) in the sector and returns also a flag for events that contain more than 2 muons. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
An example input to a Convolutional Neural Network (CNN) for the Phase-2 ATLAS Level-0 muon trigger, implemented on a FPGA, is shown. This example contains three muons with background. Resistive Plate Chambers (RPC) hits of a fixed $\phi$ sector are arranged in a matrix-like object. Each bin of the $y$-axis corresponds to a detector layer (3 detector layers for inner station, 4 for the middle and 2 for the outer station). The $x$-axis maps the $\eta$ coordinates of each physics RPC strip: for the $i$-th strip $\eta_{bin}^{i} = 384 \frac{\eta^{i}-\eta^{min}} {\eta^{max}-\eta^{min}}$, where $\eta^{max}$ and $\eta^{min}$ are respectively the maximum ($\eta^{max}=0.95$) and the minimum ($\eta^{min}=0.07$) $\eta$ values for the barrel RPC strips chosen to prevent muons to fall outside any layer of a specified sector; 384 is a realistic number of strips per layer. This particular choice has been taken in order to evaluate ML algorithm performances, without any geometrical acceptance effect. Random background has been added. The background rate has been evaluated from minimum bias events. Events used in the training phase of the CNN can also contain two or more muons in the same sector. Events with more than one muon are built superimposing one muon images with no background, which is then included. The CNN output is set to evaluate transverse momentum and $\eta$ of the leading and sub-leading muons (if the latter exists) in the sector and returns also a flag for events that contain more than 2 muons. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
A Convolutional Neural Network (CNN) implementation for the Phase-2 ATLAS Level-0 muon trigger with floating point weights has been set up and trained. The inputs to this network are all the strips of all the detector layers of a sector. The predicted leading muon transverse momentum ($p_T$) is shown as a function of the leading muon $p_T$ after detector simulation (true $p_T$ ). The reconstructed transverse momentum is used for the network training phase. The columns of the histogram are normalized to unity, since leading muon $p_T$ distribution is not flat. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
A ternary Convolutional Neural Network (tCNN, arXiv:1605.04711) implementation for the Phase-2 ATLAS Level-0 muon trigger has been set up and trained. The inputs to this network are all the strips of all the detector layers of a sector. The predicted leading muon transverse momentum ($p_T$ ) is shown as a function of the leading muon $p_T$ after detector simulation (true $p_T$ ). Ternary networks have weights made of just 2 bits. For this reason, tCNNs represent an optimal solution for FPGA synthesis and implementation, since the resource occupancy can be reduced up to a factor of 16, compared to a same-architecture non-ternary network. The columns of the histogram are normalized to unity. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
A ternary Convolutional Neural Network (tCNN, arXiv:1605.04711) implementation for the Phase-2 ATLAS Level-0 muon trigger has been set up and trained. The inputs to this network are all the strips of all the detector layers of a sector. The predicted number of muon is shown as a function of the numbers of muons obtained from truth information after detector simulation. Ternary networks have weights made of just 2 bits. For this reason, tCNNs represent an optimal solution for FPGA synthesis and implementation, since the resource occupancy can be reduced up to a factor of 16, compared to a same-architecture non-ternary network. The columns of the histogram are normalized to unity. No $p_T$ selection is applied to muon candidates. Fake muons rate is at the level of 0.6% but is furthermore reduced requiring reasonable $p_T$ thresholds (great fraction of fakes is predicted at very low momentum). .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
A comparison of the efficiencies of the Phase-2 ATLAS Level-0 standard muon trigger algorithm (cyan), the Convolutional Neural Network (red, CNN) and the ternary Convolutional Neural Network (blue, tCNN) is shown as a function of the transverse momentum $p_T$ . For the standard algorithm only the efficiency obtained with the configuration which requires at least 3 spatial coincidences over the 4 Resistive Plate Chambers stations is shown. All the efficiency curves take the single-muon images without background as input, and are tuned so that the efficiency reached at nominal threshold is the same (82%) to ease the comparison. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato
The efficiencies of the Phase-2 ATLAS Level-0 standard muon trigger algorithm (cyan) and the Convolutional Neural Networks with ternary weights (tCNN) and varying input sizes is shown as a function of the transverse momentum $p_T$ . For the standard algorithm only the efficiency obtained with the configuration which requires at least 3 spatial coincidences over the 4 Resistive Plate Chambers stations is shown. The network which takes as input all the strips of the images and has the best performance (blue) is shown. In magenta, a tCNN which takes as input just a section of the whole image at a time, and operate scanning the image, is shown. Passing a smaller input to the tCNN widely reduces the total number of multiplication, therefore making it possible to increase the parallelization and to fullfill the latency requirement. All the efficiency curves take the single-muon images without background as input, and are tuned so that the efficiency reached at nominal threshold is the same (82%) to ease the comparison. .pdf
pdf
contact: Luigi Sabetta & Federica Riti & Simone Francescato

Level-0 TGC Trigger

Performance of trigger algorithms in software and firmware implementations: ATL-COM-DAQ-2019-081 (25 Jul 2019)

Expected efficiency for the Level-0 muon trigger with HL-LHC scheme (red) and with LHC Run-2 scheme (blue) . The efficiency values relative to offline muons as trigger performance are obtained for a pseudorapidity range 1.05 < |η| < 2.4 and a transverse momentum (pT) threshold of 20 GeV with a single muon Monte Carlo simulation sample for the HL-LHC scheme, and data taken in 2018 for the Run-2 scheme. The HL-LHC scheme is based on Thin Gap Chamber (TGC), Tile Calorimeter (TileCal) and New Small Wheel (NSW). The HL-LHC scheme assumes track segment reconstruction with TGC hits by a pattern matching algorithm combined with the TileCal and NSW information for the determination of pT. The segments from Monitored Drift Tube are used to emulate the NSW segments. The HL-LHC scheme provides a higher efficiency in the plateau region with better rejection of low pT muons. The high efficiency at the plateau in the HL-LHC scheme comes from a loose condition for the coincidence among the TGC layers in the pattern matching algorithm, which requires 5 hits in 7 layers. The better pT resolution in the turn-on curve is from improvements in the TGC track reconstruction algorithm. .pdf
pdf
contact: Yuya Mino & Toshi Sumida
Estimated rate of the Level-0 single muon trigger at HL-LHC based on Thin Gap Chamber, Tile Calorimeter, and New Small Wheel for a pseudorapidity range 1.05 < |η| < 2.4 and a transverse momentum threshold of 20 GeV. The segments from Monitored Drift Tube are used to emulate the NSW segments. Events in the Run-2 data sample taken in 2016 with the zero-bias trigger are used for the point with 1x1034 cm-2s-1 and overlaid to account for higher luminosity points. The number of the simultaneous interactions in each bunch crossing ("pileup") in the original data is 27 in average. Pileup conditions at higher luminosity values are simulated for each event by overlaying a number of zero-bias events drawn from the expected distribution for that luminosity. The solid line shows a fitting result by a linear function crossing the origin. The dashed lines show the trigger rate at the luminosity in the HL-LHC (7.5x1034 cm-2s-1), calculated from a nominal number of pileup (200). .png
pdf
contact: Yuya Mino & Toshi Sumida
Distribution of the polar angle difference Δθ between a track segment reconstructed by a pattern matching algorithm with Thin Gap Chamber (TGC) hits assumed to be used in the Level-0 muon trigger at HL-LHC and a track segment reconstructed by an offline algorithm based on Monitored Drift Tube (MDT) hits. Precise measurement of the track segment angle is profited to determine muon transverse momentum and obtain sharper efficiency turn-on curves. The TGC pattern matching algorithm is implemented in a Virtex UltraScale+ FPGA XCVU9P on an evaluation kit. Test vectors of TGC hits are obtained from Monte-Carlo (MC) sample and used as the FPGA inputs. The FPGA outputs are compared with the offline track segments based on MDT hits. The offline track segment is considered to be a reference, thus the width of Δθ distribution is regarded as resolution of the TGC pattern matching algorithm for muon tracks used in this study. The MC sample includes a muon in each event, and the events are selected by requiring exactly seven TGC hits on the seven layers, i.e. neither missing hits nor cross talks, in a pseudorapidity range 2.13 < η < 2.16. Memory usage of the algorithm in this study corresponds to about one third of the total memory resource of XCVU9P when a full η range of the endcap is included. .png
pdf
contact: Haruka Asada & Yasuyuki Horii

Level-0 MDT Trigger


Major updates:
-- MasayaIshino - 2019-07-24

Responsible: MasayaIshino
Subject: Phase2 Upgrade Trigger

Topic attachments
I Attachment History Action Size Date Who Comment
PDFpdf ATL-COM-DAQ-2019-081-fig1.pdf r1 manage 36.4 K 2019-09-20 - 17:36 MasayaIshino  
PDFpdf ATL-COM-DAQ-2019-081-fig2.pdf r1 manage 28.9 K 2019-09-20 - 17:36 MasayaIshino  
PDFpdf ATL-COM-DAQ-2019-081-fig3.pdf r1 manage 24.9 K 2019-09-20 - 17:36 MasayaIshino  
PDFpdf ATL-COM-DAQ-2019-128-fig1.pdf r1 manage 16.8 K 2019-09-20 - 17:49 MasayaIshino  
PDFpdf ATL-COM-DAQ-2019-189-fig1.pdf r1 manage 173.2 K 2019-11-08 - 10:57 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig10.pdf r1 manage 71.7 K 2019-11-08 - 11:16 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig11.pdf r1 manage 108.3 K 2019-11-08 - 11:16 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig12.pdf r1 manage 108.7 K 2019-11-08 - 11:16 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig2.pdf r1 manage 172.5 K 2019-11-08 - 11:01 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig3.pdf r1 manage 172.3 K 2019-11-08 - 11:01 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig4.pdf r1 manage 172.5 K 2019-11-08 - 11:01 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig5.pdf r1 manage 172.9 K 2019-11-08 - 11:16 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig6.pdf r1 manage 172.7 K 2019-11-08 - 11:16 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig7.pdf r1 manage 173.2 K 2019-11-08 - 11:16 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig8.pdf r1 manage 78.6 K 2019-11-08 - 11:16 RiccardoVari  
PDFpdf ATL-COM-DAQ-2019-189-fig9.pdf r1 manage 78.4 K 2019-11-08 - 11:16 RiccardoVari  
Edit | Attach | Watch | Print version | History: r5 < r4 < r3 < r2 < r1 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r5 - 2019-11-20 - LuigiSabetta
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Atlas All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2020 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback