Stochastic resonance (SR) — improvement of a nonlinear system's response to a weak signal by added noise — is experimentally established as a weak-signal detection tool, but only within tuned regimes whose boundaries are found empirically. What remains missing is a general theory: necessary and sufficient conditions — in terms of signal class, noise statistics (including non-Gaussian and colored noise), detector parameters, and the chosen detection metric — under which adding or tuning noise improves detection performance, and a principled general rule for tuning noise intensity and system parameters. Later literature settles the existence question (noise-induced signal-to-noise-ratio gains exceeding unity) and gives rigorous noise-benefit conditions for static threshold detectors in statistical detection theory, but no analogous characterization exists for the dynamical bistable systems used in practice; tuning remains heuristic and platform-specific, and the optimal noise level is provably metric-dependent. The general characterization and tuning problem is open.

Background

Stochastic resonance (SR) is the counterintuitive phenomenon in which adding noise to a nonlinear system improves its response to a weak signal. The canonical setting is a bistable system — one with two stable states separated by a potential barrier, such as an overdamped particle in the quartic double-well potential V(x)=−ax2/2+bx4/4V(x)=-ax^{2}/2+bx^{4}/4 — driven by a subthreshold signal (one too weak to push the system across the barrier on its own) plus random noise. When the noise-induced hopping rate between wells matches the period of a weak periodic signal, the hopping becomes partially synchronized with the signal, and the output signal-to-noise ratio (SNR, the ratio of power in the signal's spectral line to the surrounding background noise power) attains a maximum at a finite, nonzero noise intensity. SR was proposed to explain the quasi-periodic recurrence of Earth's ice ages and first observed electronically in a Schmitt trigger — a comparator circuit with two switching thresholds that acts as an electronically convenient bistable element; the modern theory, including analytic SNR formulas for weakly periodic signals in white Gaussian noise under the two-state approximation (which reduces the dynamics to hops between two discrete states with noise-dependent rates), is systematized in the 1998 review of Gammaitoni, Hänggi, Jung and Marchesoni.

Early on it was unclear whether SR is genuinely useful for signal reception, i.e. whether the output SNR can exceed the input SNR (an SNR gain larger than one) rather than merely reshaping the output spectrum. Experiments on a bistable electronic circuit demonstrated SNR gains above one in tuned regimes (Gingl, Vajtai and Kiss, 2000); suprathreshold stochastic resonance — a noise benefit for signals strong enough to cross a single threshold, realized in arrays of parallel noisy threshold devices — showed such arrays can transmit information efficiently even above threshold (Stocks, 2000, 2001); and the SNR-gain enhancement can itself be maximized twice by combining SR with vibrational resonance, in which a high-frequency deterministic drive rather than noise plays the enhancing role (Duan, Chapeau-Blondeau and Abbott, 2014). The existence question — can SR improve weak-signal reception at all? — thus has a demonstrated affirmative answer in specific tuned regimes.

Three applied studies define what remains unresolved. Kim and Seo (2025) built a hardware bistable Schmitt-trigger detector and showed experimentally that SR-based frequency detection of damped sinusoidal pulses works at carrier frequencies above roughly 100 Hz but degrades sharply below, while amplitude estimation is blocked by an integral relation between switching statistics and the signal envelope that has no analytical inversion. Zheng, Li, Djouadi and Wang (2009) applied SR as a pre-filter of an energy detector — a receiver that decides signal presence by comparing the total received energy in a frequency band to a threshold — for spectrum sensing, i.e. detection by unlicensed secondary users of very weak licensed (primary-user) signals in cognitive radio, and found that the SR benefit is confined to very low input SNR (at −10-10 dB input SNR the SR-preprocessed detector underperforms an ordinary energy detector), with the system parameters aa, bb and the added-noise intensity selected by example rather than by any general rule. Wu, Yang, Sanjuán and Liu (2020) extended SR to unknown linear-frequency-modulated (LFM, or chirp) signals, whose sweeping instantaneous frequency defeats the conventional SR performance indices, introducing a performance index based on the fractional Fourier transform (a generalization of the ordinary Fourier transform whose basis functions are chirps, matched to an LFM signal's sweeping frequency) and a cloud-model genetic algorithm (a metaheuristic optimizer) for adaptation, but validated the method only for Gaussian white noise with empirically chosen control coefficients.

A parallel rigorous line exists in statistical detection theory, but only for static (memoryless) nonlinear devices: Kay (2000) showed that adding noise can improve the performance of suboptimal detectors; Chen, Varshney, Kay and Michels (2007, 2008) developed a theory of the SR effect in Neyman–Pearson detection — the framework that maximizes detection probability at a fixed false-alarm probability — for fixed and variable detectors; related analyses cover static nonlinear signal transmission (Chapeau-Blondeau and Godivier, 1997; Chapeau-Blondeau, 1999; Rousseau and Chapeau-Blondeau, 2005) and the limitations of SR for sine detection (Zozor and Amblard, 2002). No analogous necessary-and-sufficient characterization exists for the dynamical bistable systems used in the applied work. Crucially, any general answer must be indexed by the chosen detection observable: Kim, Kim, Park, Kim and Park (2026) demonstrate experimentally in an analog Schmitt trigger that the noise level maximizing spectral SNR is simultaneously a minimum of the input–output mutual information (the information-theoretic quantity measuring how much the output reveals about the input signal) for amplitude-encoded signals, and McDonnell and Abbott (2009) document that "stochastic resonance" itself has context- and metric-dependent, evolving definitions. The unresolved target is therefore not whether SR can help — it demonstrably can — but the general conditions under which it helps and a principled way to tune it.

Problem Statement

Characterize, in general, when stochastic resonance improves weak-signal detection performance, and how to tune it. No unifying theory is known for the general, metric-indexed characterization across signal classes, noise statistics, detector parameters, and detection observables, so the task poses one bounded, structured case of it, for which both parts (1) and (2) are required. Specifically, for nonlinear — in particular bistable dynamical — detectors of weak signals in additive noise:

  1. Conditions. For the bounded class of the canonical overdamped bistable detector — an overdamped particle in the quartic double-well potential V(x)=−ax2/2+bx4/4V(x)=-ax^{2}/2+bx^{4}/4 — driven by a periodic subthreshold signal in additive white Gaussian noise, give necessary and sufficient conditions, stated for a single fixed decision statistic on the device output — the correlation of that output with the known periodic waveform — in terms of its detection probability at a fixed false-alarm rate, under which adding or tuning noise improves that metric relative to the same detector operated with the tunable added-noise intensity set to zero but with its ambient input noise retained, so that the signal-absent hypothesis remains random.

  2. Tuning. For the same bounded detector and observable, provide a principled rule or algorithm for selecting the noise intensity and the system parameters aa and bb — replacing the example-based, calibration-curve, and evolutionary-heuristic selections of the applied literature — and state what optimality, if any, it guarantees and under which assumptions.

The result for this bounded class must account for its regime boundaries, stating how the optimal noise level is defined by the single observable it fixes. The general, multi-axis characterization remains open in its own right: the degradation of SR-tuned detectors at moderate input SNR, the failure of conventional SR performance indices for chirp signals, the metric-dependence of the optimal noise level, and robustness to non-Gaussian and colored noise are parts it still has to explain.

References

  1. Luca Gammaitoni, Peter Hänggi, Peter Jung, Fabio 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
  2. Yoonkang Kim, Donghyeok Seo, 'Stochastic resonance in Schmitt trigger and its application towards weak signal detection', arXiv:2501.10405 (2025), https://arxiv.org/abs/2501.10405 (also posted as Research Square preprint, DOI 10.21203/rs.3.rs-5821721/v1, https://doi.org/10.21203/rs.3.rs-5821721/v1)
  3. Kun Zheng, Husheng Li, Seddik M. Djouadi, Jun Wang, 'Spectrum Sensing in Low SNR Regime via Stochastic Resonance', arXiv:0906.0739 (2009), https://arxiv.org/abs/0906.0739
  4. Chengjin Wu, Jianhua Yang, Miguel A. F. Sanjuán, Houguang Liu, 'Stochastic resonance induced by an unknown linear frequency modulated signal in a strong noise background', Chaos 30, 043128 (2020). DOI 10.1063/5.0002134, https://doi.org/10.1063/5.0002134
  5. Z. Gingl, R. Vajtai, L. B. Kiss, 'Signal-to-noise ratio gain by stochastic resonance in a bistable system', Chaos, Solitons and Fractals 11, 1929-1932 (2000). DOI 10.1016/S0960-0779(99)00131-9, https://doi.org/10.1016/S0960-0779(99)00131-9
  6. 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
  7. N. G. Stocks, 'Information transmission in parallel threshold arrays: Suprathreshold stochastic resonance', Physical Review E 63, 041114 (2001). DOI 10.1103/PhysRevE.63.041114, https://doi.org/10.1103/PhysRevE.63.041114
  8. Fabing Duan, François Chapeau-Blondeau, Derek Abbott, 'Double-maximum enhancement of signal-to-noise ratio gain via stochastic resonance and vibrational resonance', Physical Review E 90, 022134 (2014). DOI 10.1103/PhysRevE.90.022134, https://doi.org/10.1103/PhysRevE.90.022134
  9. Steven Kay, 'Can detectability be improved by adding noise?', IEEE Signal Processing Letters 7, 8-10 (2000). DOI 10.1109/97.809511, https://doi.org/10.1109/97.809511
  10. Hao Chen, Pramod K. Varshney, Steven M. Kay, James H. Michels, 'Theory of the Stochastic Resonance Effect in Signal Detection: Part I—Fixed Detectors', IEEE Transactions on Signal Processing 55, 3172-3184 (2007). DOI 10.1109/TSP.2007.893757, https://doi.org/10.1109/TSP.2007.893757
  11. Hao Chen, Pramod K. Varshney, 'Theory of the Stochastic Resonance Effect in Signal Detection—Part II: Variable Detectors', IEEE Transactions on Signal Processing 56, 5031-5041 (2008). DOI 10.1109/TSP.2008.928509, https://doi.org/10.1109/TSP.2008.928509
  12. François Chapeau-Blondeau, Xavier Godivier, 'Theory of stochastic resonance in signal transmission by static nonlinear systems', Physical Review E 55, 1478-1495 (1997). DOI 10.1103/PhysRevE.55.1478, https://doi.org/10.1103/PhysRevE.55.1478
  13. Steeve Zozor, Pierre-Olivier Amblard, 'On the use of stochastic resonance in sine detection', Signal Processing 82, 353-367 (2002). DOI 10.1016/S0165-1684(01)00174-8, https://doi.org/10.1016/S0165-1684(01)00174-8
  14. François Chapeau-Blondeau, 'Stochastic Resonance and Optimal Detection of Pulse Trains by Threshold Devices', Digital Signal Processing 9, 162-177 (1999). DOI 10.1006/dspr.1999.0341, https://doi.org/10.1006/dspr.1999.0341
  15. David Rousseau, François Chapeau-Blondeau, 'Stochastic resonance and improvement by noise in optimal detection strategies', Digital Signal Processing 15, 19-32 (2005). DOI 10.1016/j.dsp.2004.09.006, https://doi.org/10.1016/j.dsp.2004.09.006
  16. Mark D. McDonnell, Derek Abbott, 'What Is Stochastic Resonance? Definitions, Misconceptions, Debates, and Its Relevance to Biology', PLoS Computational Biology 5, e1000348 (2009). DOI 10.1371/journal.pcbi.1000348, https://doi.org/10.1371/journal.pcbi.1000348
  17. Minwoo Kim, Heemin Kim, Seoyeon Park, Taek-min Kim, 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
  18. A. S. Asdi, A. H. Tewfik, 'Detection of weak signals using adaptive stochastic resonance', Proceedings of the 1995 International Conference on Acoustics, Speech, and Signal Processing (ICASSP 1995), vol. 2, pp. 1332-1335 (1995). DOI 10.1109/ICASSP.1995.480486, https://doi.org/10.1109/ICASSP.1995.480486
  19. Xingyu Chen, Jia Liu, Xu Zhang, Lijun Chen, 'ASR: Efficient and Adaptive Stochastic Resonance for Weak Signal Detection', IEEE INFOCOM 2023 (2023). DOI 10.1109/INFOCOM53939.2023.10228979, https://doi.org/10.1109/INFOCOM53939.2023.10228979
  20. Haitao Dong, Shilei Ma, Jian Suo, Zhigang Zhu, 'Matched Stochastic Resonance Enhanced Underwater Passive Sonar Detection under Non-Gaussian Impulsive Background Noise', Sensors 24, 2943 (2024). DOI 10.3390/s24092943, https://doi.org/10.3390/s24092943
  21. Haoyu Chen, Yongfeng Guo, Qingzeng Song, Qin Yu, 'Stochastic resonance of a fractional-order bistable system driven by corrected non-Gaussian noise and Gaussian noise', Journal of the Franklin Institute 362, 107593 (2025). DOI 10.1016/j.jfranklin.2025.107593, https://doi.org/10.1016/j.jfranklin.2025.107593