Quantitative weak-signal detection with stochastic resonance in Schmitt-trigger detector arrays: numerical inversion of the last-transition-time statistic, calibration with minimal detector counts, and low-frequency extension

A 2025 preprint by Kim and Seo demonstrated stochastic resonance (SR, the noise-enhanced detection of weak sub-threshold signals) in a custom bistable Schmitt-trigger circuit and proposed a detector array for estimating the parameters of weak damped sinusoidal pulses, but left its central inversion step unsolved: the calibration curve relating the mean last-transition time $\langle t_0 \rangle$ of the trigger output to the noise standard deviation $\sigma$ is given only as an analytically intractable expression (Eq. 9 of the source, the mean of a discretized last-transition-time distribution constructed from the normal cumulative distribution function), so amplitude and decay-constant recovery from measured $\langle t_0 \rangle$ values was not achieved. This record poses the problem of making the scheme quantitative. Its open core is (a) a validated numerical inversion of the $\langle t_0 \rangle$ – $\sigma$ relation yielding amplitude and decay-constant estimates with stated stability and uniqueness properties; extensions discussed by the source or natural within the same framework are (b) a calibration protocol reducing the number of parallel detectors needed for non-replayable, real-time signals (the source estimates dozens per point, hundreds for curve fitting, but notes that a single detector with sequential repetitions suffices when the signal can be replayed); (c) extension of frequency recovery further into the sub-100 Hz regime where FFT peak picking breaks down; and (d) characterization and compensation of op-amp nonidealities. No later work is known to close these components.

Background

Stochastic resonance (SR) is a phenomenon in which added noise, rather than degrading detection of a weak signal, enhances it: in a bistable system (one with two stable output states separated by switching thresholds), noise of the right amplitude synchronizes state transitions with a weak periodic or pulsed input, maximizing output signal-to-noise ratio. It was first demonstrated experimentally in a bistable optical system by Fauve and Heslot [2], developed into a solvable Schmitt-trigger model by Melnikov [3], and reviewed comprehensively by Gammaitoni et al. [4]. Stocks later showed that arrays of threshold detectors with independent noise can extend the benefit to suprathreshold signals [5]. A Schmitt trigger is an operational-amplifier comparator with hysteresis (two distinct switching thresholds), making it a natural electronic bistable system; its threshold $V_{th}$ is set by a resistive voltage divider. Kim and Seo [1] built such a circuit (threshold $V_{th} \approx 0.045$ V per branch at the $\pm 1$ V supply used for the SNR, hysteresis, and frequency experiments; $V_{th} = 0.199$ V at the 4 V supply used for the $\langle t_0 \rangle$ – $\sigma$ calibration data), confirmed bistability via hysteresis curves, reproduced the classic bell-shaped SNR-versus-noise- $\sigma$ curve, and then applied the device to detecting weak damped sinusoidal pulses $f(t)$ superimposed on Gaussian white noise of standard deviation $\sigma$ . Their key statistic is $\langle t_0 \rangle$ , the mean time of the last output transition before the decaying signal dies out: for deterministic input component $v(t) = V_0 - f(t)$ , with $V_0$ the trigger threshold, the source derives $\langle t_0 \rangle$ (its Eq. 9) as the mean of a discretized last-transition-time distribution built from the normal cumulative distribution function $\Phi$ evaluated at $v(t)/\sigma$ , a smooth but analytically unsolvable functional of $\sigma$ and the waveform parameters (amplitude, decay constant). Over its measured $\langle t_0 \rangle$ – $\sigma$ data the source fitted a two-parameter sigmoid-shape empirical relation (its Eq. 11), whose dependence on the decay constant (its Eq. 12) the source could not clarify for the same reason. The source envisions an array of parallel Schmitt triggers driven by noise of different standard deviations (its Fig. 16), using the measured $\langle t_0 \rangle$ – $\sigma$ curve to invert for the unknown amplitude and decay constant of the pulse. In parallel, adjacent literature has advanced the information-theoretic side of Schmitt-trigger SR — a 2026 European Journal of Physics study of mutual-information transmission and Kramers-rate analysis (the thermally activated, noise-driven switching rate between the two stable states) in a Schmitt-trigger circuit [6] — and the quantum side (quantum stochastic resonance in an a.c.-driven single-electron quantum dot [7] and in a single-photon emitter [8]), but none of it addresses the inversion and calibration problems of this detection scheme.

Problem Statement

Given a parallel array of bistable Schmitt-trigger detectors (comparators with hysteresis) driven by the same weak damped sinusoidal pulse $f(t)$ , of the form $f(t) = A e^{-bt}\sin(2\pi f t)$ with amplitude $A$ , decay constant $b$ , and carrier frequency $f$ , superimposed on independent Gaussian white noise of standard deviation $\sigma$ (different $\sigma$ for each detector, or sequential repetitions of a replayable signal at different $\sigma$ ), the statistic $\langle t_0 \rangle$ — the mean time of the last output transition before the decaying signal dies out — is given by the source's Eq. 9: the mean of a discretized last-transition-time distribution $P(t = t_0)$ constructed in the source's Eqs. 3–7 from the normal cumulative distribution function $\Phi$ evaluated at $v(t)/\sigma$ , where $v(t) = V_0 - f(t)$ with $V_0$ the trigger threshold, a transition occurs when $v(t)$ falls at or below the instantaneous noise value, and the time discretization satisfies $t_0 + N\,\Delta t = T$ over the measurement window $[0,T]$ . The source shows that $\langle t_0 \rangle$ is thereby a function of $\sigma$ and $v(t)$ , but that Eq. 9 cannot be solved analytically. For the source's $\langle t_0 \rangle$ – $\sigma$ calibration data set [1] invoked for validation, the trigger operated at $V_{dc} = 4$ V, giving $V_{th} = 0.199$ V via the source's linear fit $V_{th} = 0.051\,V_{dc} - 0.005$ (its Fig. 8), with a factor-2 input voltage divider (the double of $V_{th}$ is required at the input to switch the output); the value $V_{th} \approx 0.045$ V corresponds instead to the $\pm 1$ V SNR, hysteresis, and frequency experiments. The problem is to make this SR-based detection scheme quantitative:

Inversion. Construct and validate a numerical inverse of the $\langle t_0 \rangle$ – $\sigma$ calibration map defined by the source's Eq. 9, recovering the amplitude and decay constant of an unknown damped sinusoidal pulse from a measured set of $\langle t_0 \rangle(\sigma)$ values (across parallel detectors at different $\sigma$ , or sequential repetitions of a replayable signal), with an analysis of the stability and uniqueness of the inversion — in particular whether the two-parameter family (amplitude, decay constant) is identifiable from the curve, and how inversion error scales with measurement noise in $\langle t_0 \rangle$ (finite samples per $\sigma$ ).
Calibration efficiency (extension, non-replayable signals). For signals that cannot be replayed — where the source's own mitigation (a single detector with repeated measurements, applicable when its signal generator reproduces the same input) does not apply — reduce the source's stated requirement of dozens of detectors per $\sigma$ value (hundreds for a reliable fitted curve) via a principled calibration protocol.
Low-frequency extension. Extend reliable frequency recovery further into the sub-100 Hz regime, where fast-Fourier-transform (FFT) spectral peaks smear: in the source's simulations at non-optimal noise, frequency-estimation errors reach ~10% below roughly 100 Hz, while at the SNR-optimal noise level $\sigma = 0.03$ V the error rate at 50 Hz already falls to 0.28% and no FFT peak is detected at 10 Hz.
Nonideality accounting. Characterize and, where possible, compensate the dominant op-amp nonidealities of the physical circuit — finite slew rate, input offset, and the observed instability ('shakes') of both saturated output values — quantifying their effect on the threshold $V_{th}$ , on Eq. 9's validity, and on the inversion error.

These components share one scientific objective: converting the source's qualitative SR demonstration into a quantitative, calibrated weak-signal detection scheme keyed to the last-transition-time statistic. The source's explicitly open core is component 1: a validated numerical inversion of Eq. 9 that recovers the decay constant, and hence the envelope amplitude, of an unknown pulse from measured $\langle t_0 \rangle(\sigma)$ values, with quantified stability and uniqueness, validated on simulated pulses and, ideally, the source's experimental parameter set, would constitute a resolution. Components 2–4 lie beyond the completion boundary: the detector-count burden of component 2 concerns only non-replayable, real-time signals, since the source itself notes that a single detector with sequential repetitions suffices when the signal can be replayed; components 3–4 are natural extensions within the same framework. The quantum-dot translation floated by the source is speculative outlook and is likewise not required for resolution.

References

Yoonkang Kim and Donghyeok Seo, "Stochastic resonance in Schmitt trigger and its application towards weak signal detection", arXiv:2501.10405 (2025), also posted as DOI 10.21203/rs.3.rs-5821721/v1, https://arxiv.org/abs/2501.10405
S. Fauve and F. Heslot, "Stochastic resonance in a bistable system", Physics Letters A 97, 5–7 (1983), DOI 10.1016/0375-9601(83)90086-5, https://doi.org/10.1016/0375-9601(83)90086-5
V. I. Melnikov, "Schmitt trigger: A solvable model of stochastic resonance", Physical Review E 48, 2481–2489 (1993), DOI 10.1103/PhysRevE.48.2481, https://doi.org/10.1103/PhysRevE.48.2481
L. Gammaitoni, P. Hänggi, P. Jung, and F. Marchesoni, "Stochastic resonance", Reviews of Modern Physics 70, 223–287 (1998), DOI 10.1103/RevModPhys.70.223, https://doi.org/10.1103/RevModPhys.70.223
N. G. Stocks, "Suprathreshold Stochastic Resonance in Multilevel Threshold Systems", Physical Review Letters 84, 2310–2313 (2000), DOI 10.1103/PhysRevLett.84.2310, https://doi.org/10.1103/PhysRevLett.84.2310
Minwoo Kim, Heemin Kim, Seoyeon Park, Taek-min Kim, and Yun Daniel Park, "Stochastic resonance in a Schmitt-trigger circuit: divergent optimal noise levels for signal detection and image reconstruction", European Journal of Physics 47, 035106 (2026), DOI 10.1088/1361-6404/ae7eb2, https://doi.org/10.1088/1361-6404/ae7eb2
Timo Wagner, Peter Talkner, Johannes C. Bayer, Eddy P. Rugeramigabo, Peter Hänggi, and Rolf J. Haug, "Quantum stochastic resonance in an a.c.-driven single-electron quantum dot", Nature Physics 15, 330–334 (2019), DOI 10.1038/s41567-018-0412-5, https://doi.org/10.1038/s41567-018-0412-5
H. Mannel, J. Zöllner, E. Kleinherbers, M. Zöllner, N. Schwarz, F. Rimek, A. D. Wieck, A. Ludwig, A. Lorke, J. König, and M. Geller, "Quantum stochastic resonance in a single-photon emitter", Communications Physics 8, 404 (2025), DOI 10.1038/s42005-025-02334-4, https://doi.org/10.1038/s42005-025-02334-4 (preprint arXiv:2505.14221, https://arxiv.org/abs/2505.14221)