---
title: Pulse Shape Comparison Scanning (PSCS)
url: https://www.emergentmind.com/topics/pulse-shape-comparison-scanning-pscs
type: topic
---

# Pulse Shape Comparison Scanning (PSCS)

Searching arXiv for PSCS and closely related pulse-shape comparison literature to ground the article.
arxiv_search.query({"search_query":"all:\"Pulse Shape Comparison Scanning\" OR all:PSCS", "max_results": 10})
Searching arXiv for AGATA PSCS and related pulse-shape comparison methods.
arxiv_search.query({"search_query":"all:AGATA AND (all:\"Pulse Shape Comparison Scanning\" OR all:PSCS OR all:\"pulse shape comparison\")", "max_results": 10})
Searching arXiv for hardware and template-based pulse-shape comparison methods relevant to PSCS.
arxiv_search.query({"search_query":"all:\"pulse shape discrimination\" AND (all:template OR all:comparison OR all:scan)", "max_results": 10})
Pulse Shape Comparison Scanning (PSCS) is a characterization procedure used in AGATA to construct an experimental pulse-shape database for pulse shape analysis (PSA) in segmented high-purity germanium crystals. In that usage, PSCS does not denote the online localization algorithm itself; rather, it denotes an offline scan-and-match procedure in which collimated measurements acquired in different geometries are compared at waveform level so that pulse sets associated with the same three-dimensional interaction point can be assembled into an experimental basis for later PSA reconstruction [2508.06545]. Outside AGATA, the exact term is not broadly standardized: a 2025 survey of nearly sixty pulse-shape-discrimination methods does not list a method explicitly called PSCS, although several template-comparison, likelihood-comparison, and waveform-scanning methods are methodologically adjacent [2508.02750].

## 1. Definition and canonical scope

In AGATA, PSA is the operational localization method that estimates the interaction position of a \(\gamma\) ray by comparing measured traces with a pre-calculated basis of traces associated with known detector locations. PSCS belongs to the preceding characterization stage: it is one of the experimental routes for producing that basis directly from real scans, alongside coincidence scanning and electronic-collimation scanning [2508.06545].

The reason PSCS is needed is that simulated bases are not exact. The cited AGATA work states that detector response depends on crystal geometry, electric fields, weighting potentials, electronics, and boundary effects that are difficult to model perfectly. Historically, AGATA bases have often been simulated with ADL and, more recently, AGATAGeFEM; PSCS exists because an experimentally derived basis can capture those detector-specific effects directly [2508.06545].

A common source of confusion is the distinction between PSCS and PSA. PSA is the reconstruction method used during detector operation. PSCS is a calibration and database-construction procedure used to generate the pulse library on which PSA can later operate. This distinction is explicit in the AGATA literature and is central to the term’s most concrete usage [2508.06545].

## 2. Conventional AGATA PSCS procedure

The conventional Strasbourg PSCS workflow uses two collimated two-dimensional scans of the same crystal, one with the detector placed horizontally and one with it placed vertically. The geometry constrains different coordinate pairs in the two configurations: the horizontal scan determines \((X,Z)\), while the vertical scan determines \((X,Y)\). Together, those partially labeled scans provide the information needed to construct a three-dimensional experimental basis [2508.06545].

The pulse-comparison step is the core of the method. A pulse from the vertical scan and a pulse from the horizontal scan are treated as candidates for the same physical interaction point if their multichannel detector responses are sufficiently similar. Similarity is quantified by the inter-scan statistic
\[
\chi^2 = \frac{\sum\limits_{ch=0}^{36} \sum\limits_{i=1}^{120} \left(V_{ch}(i) - H_{ch}(i)\right)^2}{\sigma^2 \cdot N},
\]
where \(V_{ch}(i)\) is the amplitude in channel \(ch\) at sample \(i\) for a vertical-scan event, \(H_{ch}(i)\) is the corresponding horizontal-scan amplitude, \(\sigma\) is the noise amplitude, and \(N=37\times120\) because all 37 channels and all 120 time samples are compared [2508.06545].

Pairs whose \(\chi^2\) falls below a threshold are taken to correspond to the intersection point of the two scans. The process is iterative: the selected signal set at each position is progressively refined so that the final basis is representative of that interaction point. In strict methodological terms, this is waveform-level pulse comparison across a scan geometry rather than feature-space classification [2508.06545].

## 3. Acquisition geometry and signal representation

The AGATA implementation described in the literature uses the Strasbourg scanning table with a collimated \(^{137}\)Cs source and a 1 mm diameter pinhole collimator. The scan pitch is \(2\times2\ \mathrm{mm}^2\). Two crystals are reported explicitly: S001, a non-standard prototype symmetric crystal, and A005, a standard AGATA type-A crystal previously used at GANIL [2508.06545].

The scan densities illustrate the practical scale of PSCS campaigns. For S001, the vertical and horizontal scans comprised 1351 and 2195 points, respectively. For A005, the corresponding counts were 1305 and 1858. These numbers matter because conventional PSCS compares pulse populations across all such scan points, which is one reason its computational burden is substantial [2508.06545].

Each raw event consists of digitized traces from 37 channels, namely 36 segment channels and one core channel, with 120 samples per trace and 10 ns per sample. Conventional PSCS operates on these recorded detector traces directly. Later machine-learning work derived a reduced “super-trace” representation by retaining 60 samples from each of the 37 channels, but that reduction belongs to the successor method rather than to conventional PSCS itself [2508.06545].

This signal representation clarifies what PSCS is comparing. It is not a scalar pulse descriptor, and it is not merely a segment-energy pattern. It is a full multichannel waveform object whose spatial information is distributed across core and segment time development.

## 4. Relation to adjacent pulse-shape methods

Outside AGATA, PSCS is best understood as part of a broader family of waveform-comparison methods, but the literature distinguishes it from several nearby approaches. The 2007 HPGe study implemented a self-calibrating real-time discriminator based on three scalar features—pulse width, asymmetry, and normalized moment—mapped into a 3D acceptance histogram. That method uses feature-space acceptance rather than waveform-to-template comparison, and the paper is explicit that it is not equivalent to PSCS [0712.0594].

Two other methods are much closer in spirit. In Si-based neutron detectors, a measured sequence \(V_m(t_i)\) is compared against a stored neutron reference pulse \(V_r(t_i)\) after amplitude scaling
\[
A_s = \frac{\sum_i V_m(t_i)}{\sum_i V_r(t_i)},
\qquad
S = \sum_i \left[V_m(t_i) - A_s V_r(t_i)\right]^2,
\]
with acceptance when \(S<S_o\). The paper does not name this PSCS, but it is explicitly a single-template residual-based pulse-shape comparison and can be interpreted as a partial PSCS analogue [1805.01261].

The n\_TOF pulse-processing framework is likewise PSCS-adjacent. It scans a template across a candidate interval, solves analytically for the best amplitude scale \(\alpha_i=C_i/P_i\), and evaluates a reduced-\(\chi^2\)-like fit quality at each candidate position. This is waveform scanning with template adjustment, even though the paper presents it as a detector-generic pulse-shape fitting routine rather than under the PSCS label [1601.04512].

A more compressed hardware approximation appears in the Citiroc1A ASIC work. There, two differently shaped outputs from the same pulse are compared through a ratio between low-gain and high-gain peak outputs, and all 49 combinations of the available peaking times are scanned to maximize neutron/gamma separation. The authors do not call this PSCS, and it is not full template matching, but the method is explicitly described as a reduced-feature pulse-shape comparison scheme in which the shaping constants are scanned over a discrete configuration space [2403.16927].

The 2025 survey helps delimit the terminology. It does not list a method explicitly called PSCS, but it identifies closely related statistical methods: Pattern Recognition, which compares a pulse vector to a reference vector by angular similarity; the Gatti Parameter, which compares an event to neutron and gamma templates through a weighted difference; and the Log-Likelihood Ratio, which compares a pulse against class-dependent pulse-shape probability models [2508.02750]. This suggests that “PSCS” is best treated as a specific calibration-and-matching practice within the wider domain of pulse-shape comparison, not as a universally standardized algorithmic label.

## 5. Limitations, ambiguities, and common misconceptions

The conventional AGATA implementation has several explicit limitations. The “optimal” \(\chi^2\) threshold is not uniform throughout the crystal; it varies with position. This makes automation harder and introduces location-dependent tuning. The process is also computationally heavy because many pulse pairs across the two scans must be compared [2508.06545].

The scan data themselves are imperfect labels. The AGATA study notes contamination from random coincidences, multiple-hit events, background, and events mislabeled with respect to nominal scan position. These effects are particularly problematic toward the back of the crystal in some geometries, especially in the vertical scan. In addition, even photopeak events can contain multiple interactions within the same segment, so the observed waveform may not correspond cleanly to the nominal collimated point [2508.06545].

A second misconception is to equate any pulse-shape discriminator with PSCS. The comparative literature argues against that simplification. Feature-space classifiers, tail-to-total ratios, and onboard shaping-ratio methods all exploit pulse-shape information, but they compress the waveform early and do not perform the inter-template or inter-scan matching that defines conventional PSCS in AGATA [0712.0594, 2403.16927].

A third misconception is that full waveform comparison necessarily implies a single preferred evaluation metric. The broad PSD benchmark literature argues that Figure of Merit can be misleading for comparison-style methods when the score distribution is non-Gaussian or threshold-sensitive, and recommends labeled metrics such as F1-score or ROC-AUC when class labels are available [2508.02750]. This does not alter the AGATA PSCS workflow directly, but it matters when PSCS-like methods are compared against alternative classifiers.

## 6. Later developments and current direction

The most explicit later development is the replacement of conventional PSCS processing by a neural method trained directly on Strasbourg scan-table data. In that work, the scan campaign is retained, but the conventional pulse-pair matching algorithm is replaced by an LSTM-based regression model with a masked Euclidean loss
\[
L = \frac{1}{N} \sum_{k=1}^{N} \sqrt{ M_x (x_k - \hat{x}_k)^2 + M_y (y_k - \hat{y}_k)^2 + M_z (z_k - \hat{z}_k)^2 } ,
\]
where the masks encode whether a sample comes from a vertical scan \((M_x,M_y,M_z)=(1,1,0)\) or a horizontal scan \((1,0,1)\). The method predicts all three coordinates even though each event is only partially labeled by the scan geometry [2508.06545].

The resulting experimental database was reported to outperform both simulated bases and, where available, the conventional PSCS-derived basis. For S001, the mean standard deviation metric over the full crystal was 7.8 a.u. for the neural database versus 10.6 a.u. for conventional PSCS; for a representative slice at \(Z=30\) mm, the values were 7.4 a.u. and 8.4 a.u. For A005, PSA with the neural basis achieved position resolutions around \(3.0\)–\(3.1\) mm FWHM relative to scan position, compared with \(4.8\)–\(6.2\) mm for AGATAGeFEM and \(5.2\)–\(8.1\) mm for ADL [2508.06545].

The same work also quantified the operational burden of conventional PSCS. It reports about 5 days for conventional PSCS, versus about 10 hours on 3 GPUs for model training and about 3 hours for inference for one crystal. The authors further showed that training with only one known coordinate per event was only marginally worse than the standard two-coordinate case. This suggests a shift in emphasis: future systems may continue to rely on experimental scan data, but the reconstruction of the 3D basis may move away from explicit pairwise pulse comparison toward learned mappings from partially labeled traces [2508.06545].

In that sense, PSCS remains important both as a historical method and as a reference concept. It defines the experimental basis-construction problem—recovering a three-dimensional pulse library from partially constrained scans—even where the solution method is no longer the original pairwise \(\chi^2\) comparison.

Source: https://www.emergentmind.com/topics/pulse-shape-comparison-scanning-pscs