GEM-Rec: Reconstruction in Gas Electron Multipliers

• GEM-Rec is the integrated reconstruction system for Gas Electron Multipliers, combining detector design, calibration, and data processing into a unified discipline.
• It leverages both charge centroid and micro-TPC methods to achieve precise spatial resolution and adapt to varying detector conditions.
• GEM-Rec addresses online clustering, calibration stability, and system-level applications such as CMS triggering and high-resolution imaging.

“GEM-Rec” (Editor’s term) usefully denotes the reconstruction-centered domain of Gas Electron Multiplier research: detector geometries and readout topologies, cluster formation, charge- and time-based hit estimation, online data reduction, gain and uniformity calibration, and the downstream use of GEM observables in tracking, triggering, and imaging. In the arXiv literature, these functions are not treated as separable layers. Reconstruction quality is repeatedly tied to foil fabrication, gas composition, gap structure, electronics, environmental control, and field configuration, rather than to a single estimator in isolation (Alexeev et al., 2019, Wu et al., 2017, Ahmed et al., 2022).

1. Scope, usage, and bibliographic setting

The literature associated with GEM-Rec spans several experimental regimes. In CMS forward muon upgrades, GEM detectors are proposed for the high-$|\eta|$ endcap region to provide precision tracking, fast trigger information, improved muon trigger performance, improved muon momentum resolution, and missing redundancy in the high-$\eta$ region (Abbaneo et al., 2012). In SHiP, GEMs are evaluated as time-stamping electronic trackers coupled to emulsion targets, with explicit emphasis on position resolution as a function of incident angle and magnetic field (Alexandrov et al., 2017). In laboratory and instrumentation studies, GEM-Rec includes gain scans, gas-flow optimization, environmental normalization, foil qualification, ion-backflow suppression, and X-ray image formation (Patra et al., 2015, Greene et al., 2020, Ahmed et al., 2022).

This breadth matters because a common simplification treats GEM reconstruction as only a local hit-position problem. The published record instead associates reconstruction performance with the full detector-response chain: gain stability, cluster morphology, charge sharing, timing extraction, field-dependent transport, and uniformity correction. This suggests that GEM-Rec is best understood as a system discipline rather than a single algorithmic module.

A bibliographic caveat also exists. The arXiv record for “A GEM Detector System for an Upgrade of the High-eta Muon Endcap Stations GE1/1 + ME1/1 in CMS” contains no PDF and no source in the supplied record, so detector design, trigger logic, reconstruction, electronics, and performance statements cannot be extracted paper-faithfully from that record alone (Abbaneo et al., 2012).

2. Detector architectures and readout topologies

The detector configurations used in GEM-Rec studies range from small bench prototypes to full-scale CMS chambers. Their reconstruction implications are immediate because strip pitch, gap sequence, drift length, and segmentation determine cluster size, timing leverage, and achievable spatial precision.

System	Geometry and gaps	Readout emphasis
Small triple-GEM prototype (Patra et al., 2015)	$10 \times 10\ \mathrm{cm^2}$, $3/2/2/2\ \mathrm{mm}$	XY PCB, $256$ X + $256$ Y tracks, summed outputs
Micropack-foil triple-GEM (Gola et al., 2018)	$10 \times 10\ \mathrm{cm^2}$, $3/1/2/1\ \mathrm{mm}$	128-strip plane
Large-area self-stretched triple-GEM (You et al., 2014)	$30 \times 30\ \mathrm{cm^2}$, “3-2-2-2”	$6 \times 6$ sectorized gain readout
Full-scale CMS prototype (Abbaneo et al., 2012)	trapezoid $\eta$0, $\eta$1	$\eta$2 $\eta$3-partitions, $\eta$4 strips each
GEM-emulsion tracker (Alexandrov et al., 2017)	$\eta$5, $\eta$6 drift gap	XY strips, $\eta$7 pitch

The small NISER–IoP detector is a conventional triple-GEM using standard stretched single-mask GEM foils from CERN, a resistor-based voltage divider network, and an XY board whose many strip signals are summed into single outputs for early detector characterization rather than spatially resolved reconstruction (Patra et al., 2015). By contrast, the University of Delhi detector built with commercially manufactured Micropack foils uses a standard CMS-style small-gap configuration, a strip plane large enough to collect the full charge cluster for the reported measurements, and additional $\eta$8 protection resistors between the divider and the top electrode of each GEM foil (Gola et al., 2018).

Large-area engineering work changed the practical definition of GEM-Rec by making assembly and maintainability part of reconstruction readiness. The improved self-stretch technique derived from CERN’s NS2 concept replaces glue-based permanent mounting with mechanically stretched foils fixed on inner frames, eliminating spacers in the active area and allowing a $\eta$9 detector to be assembled in about $10 \times 10\ \mathrm{cm^2}$0 hour (You et al., 2014). A related fast self-stretching method was used in the CMS full-scale prototype program, where total assembly time was reported as less than two hours (Abbaneo et al., 2012). These assembly choices are not merely mechanical: they act on gap uniformity, dead regions, serviceability, and ultimately response uniformity.

The CMS full-scale geometry is especially reconstruction-specific. Its readout board is divided into $10 \times 10\ \mathrm{cm^2}$1 $10 \times 10\ \mathrm{cm^2}$2-partitions, each with $10 \times 10\ \mathrm{cm^2}$3 radially oriented strips; strip pitch varies from $10 \times 10\ \mathrm{cm^2}$4 to $10 \times 10\ \mathrm{cm^2}$5, and each partition is subdivided in $10 \times 10\ \mathrm{cm^2}$6 into $10 \times 10\ \mathrm{cm^2}$7 readout sectors of $10 \times 10\ \mathrm{cm^2}$8 strips each (Abbaneo et al., 2012). The chamber therefore measures the bending-sensitive coordinate with explicitly nonuniform local granularity, a fact that any chamber model or local-reconstruction software must preserve.

3. Local hit reconstruction: charge centroid, $10 \times 10\ \mathrm{cm^2}$9TPC, and field-angle coupling

The core algorithmic literature on GEM-Rec centers on two complementary local estimators: charge centroid (CC) and micro-Time-Projection-Chamber ($3/2/2/2\ \mathrm{mm}$0TPC) reconstruction. The CC method uses a weighted average of strip charges,

$3/2/2/2\ \mathrm{mm}$1

and performs best when the induced charge profile is compact and symmetric (Alexeev et al., 2019). The $3/2/2/2\ \mathrm{mm}$2TPC method converts strip time into depth,

$3/2/2/2\ \mathrm{mm}$3

assigns $3/2/2/2\ \mathrm{mm}$4 points to the strips in a cluster, and fits a local track segment inside the drift gap (Alexeev et al., 2019).

Beam tests at the CERN SPS H4 line showed why both methods are needed. In planar $3/2/2/2\ \mathrm{mm}$5 triple-GEM prototypes operated up to $3/2/2/2\ \mathrm{mm}$6, CC gave very strong performance for orthogonal tracks and no magnetic field, with spatial resolution below $3/2/2/2\ \mathrm{mm}$7 in favorable conditions and more generally well below $3/2/2/2\ \mathrm{mm}$8. As magnetic field increased, CC degraded approximately linearly with $3/2/2/2\ \mathrm{mm}$9 because Lorentz drift broadened and deformed the avalanche footprint. The $256$0TPC method, especially with a $256$1 drift gap, recovered performance in inclined-track or strong-field regimes, and the combination of CC and $256$2TPC guaranteed resolution better than $256$3 up to $256$4; at $256$5, the combined method gave a nearly flat $256$6 resolution for track angles from $256$7 to $256$8 (Alexeev et al., 2019).

The physical mechanism is the relation between track angle and Lorentz angle. In one SPS study with $256$9, $256$0, and $256$1, Garfield/Magboltz yielded $256$2 and $256$3 (Alexeev et al., 2019). When $256$4, the detector enters a focusing configuration favorable to CC; when the geometry is defocusing, $256$5TPC gains leverage.

The GEM-emulsion hybrid tracker for SHiP provides an independent measurement of the same phenomenon with an ultra-precise reference. Using a $256$6 triple-GEM with a $256$7 drift gap and emulsion reference tracks, the measured CC resolution was $256$8 at $256$9, $10 \times 10\ \mathrm{cm^2}$0, degraded to $10 \times 10\ \mathrm{cm^2}$1 at $10 \times 10\ \mathrm{cm^2}$2 without field, and reached $10 \times 10\ \mathrm{cm^2}$3 at $10 \times 10\ \mathrm{cm^2}$4, $10 \times 10\ \mathrm{cm^2}$5, where the reported Lorentz angle was about $10 \times 10\ \mathrm{cm^2}$6 for that chamber configuration (Alexandrov et al., 2017). The hybrid study therefore confirms, with a different apparatus, that charge-centroid performance is highly geometry-dependent and can be restored near the focusing condition.

A recurrent misconception is that GEM spatial resolution in magnetic field can be treated as a charge-centroid problem with small perturbative corrections. The beam data do not support that view. They show instead that field-angle coupling changes the qualitative shape of the charge distribution and makes dual-mode reconstruction structurally advantageous.

4. Online cluster reconstruction and streaming data reduction

A separate branch of GEM-Rec concerns online cluster formation inside the DAQ path. The FPGA study on on-line cluster reconstruction of GEM detectors implements a serial streaming algorithm whose purpose is to compress raw detector readout in real time so that only cluster-level information is transmitted and stored (Wu et al., 2017).

The detector used for demonstration was a two-dimensional positive-sensitive triple GEM with $10 \times 10\ \mathrm{cm^2}$7 sensitive area, a $10 \times 10\ \mathrm{cm^2}$8 drift gap, two $10 \times 10\ \mathrm{cm^2}$9 transfer gaps, a $3/1/2/1\ \mathrm{mm}$0 induction gap, and $3/1/2/1\ \mathrm{mm}$1 at volume ratio $3/1/2/1\ \mathrm{mm}$2. The readout board had $3/1/2/1\ \mathrm{mm}$3 strips per axis at $3/1/2/1\ \mathrm{mm}$4 pitch. Readout values entering the FPGA were $3/1/2/1\ \mathrm{mm}$5-bit ADC data, and the reconstruction logic was tested on an Altera DK-DEV-2AGX125N board (Wu et al., 2017).

The algorithm begins by thresholding each $3/1/2/1\ \mathrm{mm}$6-bit channel value to a $3/1/2/1\ \mathrm{mm}$7-bit occupancy,

$3/1/2/1\ \mathrm{mm}$8

then processes the stream from left to right. A fired element with no already-checked fired neighbors starts a new cluster; one with fired neighbors inherits that label; and a fired element that connects two previously distinct labels records a later merge. During streaming, the FPGA accumulates

$3/1/2/1\ \mathrm{mm}$9

which are then written to FIFO after deferred re-merging (Wu et al., 2017). The retained quantities support center-of-gravity estimation,

$30 \times 30\ \mathrm{cm^2}$0

while discarding most empty-channel payload.

The implementation is explicitly hardware-oriented rather than track-oriented. It binarizes amplitudes for clustering decisions, uses only already-checked neighbors because of scan order, and permanently deletes raw data after cluster summarization. The study validated the method on synthetic square, strip, and X-shaped cluster topologies and on real X-ray imaging data, where the FPGA-compressed outputs still reproduced a clear image of the Lanzhou University badge (Wu et al., 2017). The timing study further reported that, for a non-extreme occupancy of about $30 \times 30\ \mathrm{cm^2}$1, the ratio of reconstruction time to total data input time was about $30 \times 30\ \mathrm{cm^2}$2, implying a need for large-capacity FIFO buffering.

Within GEM-Rec, this work defines the online boundary condition: cluster finding can be moved into firmware, but only by reducing detector observables to a hardware-friendly connected-component labeling problem. It is therefore a compression architecture, not a substitute for high-level track reconstruction.

5. Calibration, stability, uniformity, and transport control

The calibration literature shows that GEM-Rec depends critically on operating-point control. In a $30 \times 30\ \mathrm{cm^2}$3 triple-GEM operated in $30 \times 30\ \mathrm{cm^2}$4, count-rate-versus-flow measurements with $30 \times 30\ \mathrm{cm^2}$5Co, $30 \times 30\ \mathrm{cm^2}$6Cs, and $30 \times 30\ \mathrm{cm^2}$7Sr showed a maximum at about $30 \times 30\ \mathrm{cm^2}$8 for all three sources and for all applied voltages tested (Patra et al., 2015). The same study logged temperature, pressure, and relative humidity continuously and fitted the source-induced anode current with

$30 \times 30\ \mathrm{cm^2}$9

with $6 \times 6$0 and $6 \times 6$1. After normalization, the detector showed no ageing observed over about $6 \times 6$2, up to $6 \times 6$3, with normalized current distributed around mean $6 \times 6$4 and $6 \times 6$5 (Patra et al., 2015).

Foil qualification studies push the same point further upstream. For three Techtra $6 \times 6$6 single-mask foils, optical scans over $6 \times 6$7 cells yielded top and bottom copper outer diameters of $6 \times 6$8 and $6 \times 6$9, top and bottom Kapton inner diameters of $\eta$00 and $\eta$01, and pitch $\eta$02 (Ahmed et al., 2022). Electrical tests reported impedance $\eta$03 and leakage current $\eta$04 at $\eta$05, with leakage current under $\eta$06 up to $\eta$07 at $\eta$08 and RH $\eta$09, and no visible discharges (Ahmed et al., 2022). The same work reported that leakage current increases linearly with temperature and exponentially with RH, while capacitance is strongly correlated with RH; it recommended leakage current less than $\eta$10, corresponding to a temperature of roughly $\eta$11 and a relative humidity of around $\eta$12 for better performance and stability of the detector (Ahmed et al., 2022).

Detector-level uniformity and startup evolution were likewise quantified. In the Techtra-based $\eta$13 triple-GEM, gain at fixed divider current $\eta$14 rose by $\eta$15 in the first $\eta$16, then by another $\eta$17 in the next $\eta$18, and became stable in the last $\eta$19, which the authors attributed to charging-up and polarization (Ahmed et al., 2022). Sector-by-sector gain mapping over $\eta$20 sectors at $\eta$21 gave

$\eta$22

This suggests that flat-field correction is not optional when quantitative image intensity or uniform cluster response is required.

Commercial-foil benchmarking reached similar operational conclusions. The Micropack-foil detector showed linear I–V behavior with effective resistance $\eta$23, no sparks or HV trips up to $\eta$24 in pure $\eta$25, maximum spurious rate $\eta$26 at about $\eta$27, maximum effective gain about $\eta$28 at $\eta$29, energy resolution about $\eta$30, and a defective-hole fraction of $\eta$31 without measurable degradation of operation (Gola et al., 2018).

Transport optimization beyond gain is addressed explicitly in the triple-versus-quadruple GEM study. Effective gain was defined as

$\eta$32

and ion backflow as

$\eta$33

At baseline symmetric operation in $\eta$34, triple GEM had IBF about $\eta$35 at gain $\eta$36, while quadruple GEM had IBF about $\eta$37 at gain $\eta$38. With optimized fields, triple-GEM IBF was reduced to about $\eta$39 and quadruple-GEM IBF to about $\eta$40 while maintaining approximately constant gain (Greene et al., 2020). A recurring consequence is that gain alone is not an adequate operational figure of merit. The papers instead tie reconstruction-relevant detector quality to a joint control of gain, transparency, charging-up, and ion transport.

6. System-level applications: CMS triggering, cost-reduced readout, and imaging

In CMS, GEM-Rec is tied directly to endcap trigger and tracking performance. The full-scale beam-tested trapezoidal prototypes for the first muon endcap station inner ring were operated in $\eta$41, stably up to gas gains of $\eta$42, with low detector noise allowing threshold $\eta$43 and a $\eta$44 efficiency plateau at gain $\eta$45 (Abbaneo et al., 2012). Using synchronous $\eta$46 timing logic, approximately $\eta$47 of hits were contained within a single $\eta$48 bunch crossing at $\eta$49 (Abbaneo et al., 2012). In an $\eta$50-section with strip pitch approximately $\eta$51, the measured residual width was $\eta$52, close to the binary expectation

$\eta$53

The result is that full-scale binary-strip operation reached essentially the strip-limited spatial precision expected from geometry.

The same CMS beam program also showed that coarse strip pitch does not imply coarse resolution if charge sharing is engineered into the readout. Small triple-GEM prototypes with $\eta$54 zigzag strips of $\eta$55 pitch, read out with SRS and APV25, achieved plateau efficiency $\eta$56 and a single-detector spatial resolution of $\eta$57, inferred from an inter-detector position-difference RMS of $\eta$58 via

$\eta$59

The paper explicitly states that such zigzag readout could reduce the number of readout strips by roughly a factor of three (Abbaneo et al., 2012). A common misconception is therefore incorrect: large pitch does not automatically force binary-scale resolution.

Imaging studies extend GEM-Rec into projection radiography. In the Techtra-based $\eta$60 triple-GEM, imaging was performed in $\eta$61 at $\eta$62, with detector gain set to approximately $\eta$63, raw data acquired at $\eta$64 samples/s, and reconstruction requiring additional data processing (Ahmed et al., 2022). One image was reconstructed from $\eta$65 events, corresponding to $\eta$66 million subevents, with effective bin size

$\eta$67

Normalized hit densities distinguished materials of different density and mass thickness: for example, FR4 gave $\eta$68, a 316-steel key $\eta$69, and copper $\eta$70 at $\eta$71 (Ahmed et al., 2022). The authors also reported that a bin wise fitting correction improved dimensional fidelity, reducing the nut outer-diameter error from $\eta$72 to $\eta$73 and the nut inner-diameter error from $\eta$74 to $\eta$75 (Ahmed et al., 2022). This establishes a distinct imaging branch of GEM-Rec in which flat-fielding, counting statistics, and offline geometric correction are as important as local charge collection.

Taken together, the literature presents GEM-Rec as a layered technical program. At one end are detector architectures, foil geometries, and serviceable assembly methods; in the middle are cluster formation, charge- and time-based estimators, and calibration against gas-flow, $\eta$76, gain drift, and ion transport; at the far end are application-specific outcomes in CMS triggering, high-resolution tracking, and transmission imaging. The published record therefore supports a system definition of GEM-Rec: not merely reconstruction on GEM data, but reconstruction conditioned by how GEM data are produced.