In modulation-based stochastic-resonance (SR) detection, a noisy measured signal is multiplied by a scanning cosine so that a weak high-frequency component appears at a small difference frequency that satisfies the adiabatic approximation of bistable SR theory, allowing its recovery from strong noise. For the 2017 source paper that formalized this scheme, every quantitative prediction - the Kramers escape rate, the output signal-to-noise ratio (SNR), and the choice of scanning frequency and circuit parameters - was derived only under the adiabatic small-parameter conditions for the symmetric quartic bistable system driven by Gaussian white noise, and validated only in simulations. The open problem is therefore to build the quantitative theory of modulation-based SR detection outside the adiabatic regime - output-SNR formulas valid for finite difference frequencies, amplitudes, and noise levels - together with a general predictive criterion for selecting the scanning frequency and system parameters, and a characterization of the method's performance limits, all within the source's model class: the symmetric quartic bistable system with additive Gaussian white noise.

Background

Stochastic resonance (SR) is the phenomenon in which adding noise to a nonlinear system enhances its response to a weak periodic signal; the modern theory is reviewed by Gammaitoni, Hänggi, Jung and Marchesoni (1998). The canonical setting is an overdamped particle in the symmetric quartic bistable potential U(x)=−μ2x2+14x4U(x)=-\frac{\mu}{2}x^{2}+\frac{1}{4}x^{4}, which has two stable wells at x=±μx=\pm\sqrt{\mu} separated by a barrier of height ΔU=μ2/4\Delta U=\mu^{2}/4. Under additive zero-mean Gaussian white noise of intensity DD, the particle hops between the wells with the Kramers escape rate Rk=2 μ2πexp⁡(−μ24D)R_{k}=\frac{\sqrt{2}\,\mu}{2\pi}\exp(-\frac{\mu^{2}}{4D}), obtained from the Fokker–Planck equation (the partial differential equation governing the probability density of the noisy state variable). SR occurs when the noise-induced hopping rate is matched to the frequency of a weak sinusoidal drive s(t)=Acos⁡Ωts(t)=A\cos\Omega t, so that noise energy is converted into a coherent output component at Ω\Omega.

The classical quantitative theory of SR detection rests on the adiabatic approximation, which requires the signal amplitude and angular frequency to be small (A≪1A\ll 1, Ω≪1\Omega\ll 1, with the further small-parameter condition Axm/D≪1Ax_{m}/D\ll 1 for expanding the escape rates, where xm=μx_{m}=\sqrt{\mu}). In this regime the dynamics reduce to a two-state model (a random telegraph process jumping between the two wells), whose Fourier analysis yields the output signal-to-noise ratio SNR=π(AxmD)2Rk\mathrm{SNR}=\pi(\frac{Ax_{m}}{D})^{2}R_{k}. Linear-response theory of SR (Dykman et al., 1993) shows that the response of bistable systems to simple sinusoidal drives can be treated beyond the strict adiabatic limit, but the detection-oriented literature has overwhelmingly stayed inside the adiabatic framework.

The practical restriction is severe: engineering signals - for example rolling-bearing fault signatures, whose characteristic harmonics lie at tens of hertz and above - violate Ω≪1\Omega\ll 1, so conventional adiabatic SR theory does not apply to them directly. Zhao, Wang and Fan (2017) addressed this with a modulation (frequency-mixing) scheme: multiplying the noisy measured signal xn(t)=a0+∑kAkcos⁡(kω0t+φk)+n(t)x_{n}(t)=a_{0}+\sum_{k}A_{k}\cos(k\omega_{0}t+\varphi_{k})+n(t), where n(t)n(t) is the corrupting measurement noise and kk indexes the harmonics of the fundamental frequency ω0\omega_{0}, by a scanning cosine y(t)=cos⁡(ωct)y(t)=\cos(\omega_{c}t) produces spectral components at the sum and difference frequencies kω0±ωck\omega_{0}\pm\omega_{c}; by choosing ωc\omega_{c} close to a target harmonic so that the difference frequency Δω=kω0−ωc≪1\Delta\omega=k\omega_{0}-\omega_{c}\ll 1, the weak large-frequency component is converted into a low-frequency drive that satisfies the adiabatic condition. They re-derived the Kramers rate and output SNR under these assumptions, designed an analog modulator circuit (mixer followed by integrator, adder, inverter, multipliers and gain blocks realizing the bistable nonlinearity) as a standard implementation of the canonical equation x˙=μx−x3+xn(t)y(t)\dot{x}=\mu x-x^{3}+x_{n}(t)y(t), in which the multiplier acts on the input branch and the gains k1,k2k_{1},k_{2} realize the cubic nonlinearity, and demonstrated detection of single- and multi-tone signals in simulations (e.g., μ=0.8\mu=0.8, A=0.6A=0.6, D=2D=2, a 10 Hz tone recovered using a 9.9 Hz scanning cosine). Related engineering strategies for large parameters include parameter-adjusted frequency shifting (Zhang, Wang and Hui, 2012) and the re-scaling/scale-transformation school (Lai and Leng, 2015; Liu, Leng, Lai and Fan, 2018), which map large signal parameters into the adiabatic window by variable substitution rather than extending the theory beyond it.

The unresolved issue left by this body of work is that the modulation scheme is a demonstrated trick, not a predictive theory. All closed-form predictions hold only inside the adiabatic small-parameter regime; the scanning frequency and system/circuit parameters are chosen by trial, simulation sweeps, or black-box optimization (for example the parameter-adaptive SR tuned by the whale optimization algorithm (a nature-inspired metaheuristic) of He, Huang, Wang, Yan and Dong, 2019, or the genetic-algorithm tuning used in the tristable and quad-stable variants of Zhang and co-workers, 2022–2025); and no analysis delimits when the method fails.

Problem Statement

For the frequency-mixing stochastic-resonance detection scheme - in which the noisy measured signal xn(t)=a0+∑kAkcos⁡(kω0t+φk)+n(t)x_{n}(t)=a_{0}+\sum_{k}A_{k}\cos(k\omega_{0}t+\varphi_{k})+n(t) (where n(t)n(t) is the corrupting measurement noise and kk indexes the harmonics of the fundamental frequency ω0\omega_{0}) is multiplied by a scanning cosine y(t)=cos⁡(ωct)y(t)=\cos(\omega_{c}t) and the mixed signal drives the overdamped bistable system x˙=μx−x3+xn(t)y(t)\dot{x}=\mu x-x^{3}+x_{n}(t)y(t) with U(x)=−μ2x2+14x4U(x)=-\frac{\mu}{2}x^{2}+\frac{1}{4}x^{4} - all existing quantitative predictions rest on the adiabatic small-parameter regime (A≪1A\ll 1, Ω≪1\Omega\ll 1, Axm/D≪1Ax_{m}/D\ll 1, difference frequency Δω=kω0−ωc≪1\Delta\omega=k\omega_{0}-\omega_{c}\ll 1 for a chosen target harmonic kk), within which the two-state/Kramers analysis yields the output SNR π(AxmD)2Rk\pi(\frac{Ax_{m}}{D})^{2}R_{k} with Rk=2 μ2πexp⁡(−μ24D)R_{k}=\frac{\sqrt{2}\,\mu}{2\pi}\exp(-\frac{\mu^{2}}{4D}).

Establish the quantitative theory of this detection scheme outside that regime, and derive from it operative design rules. The single scientific objective is to replace the simulation-demonstrated mixing trick with a predictive detection theory; three jointly supporting results serve it:

  1. Compute the output SNR (or an equivalent, defined detection-performance measure, with the spectral component at which it is measured identified) of the bistable detector under the modulated multi-frequency drive when the small-parameter conditions are violated - at minimum in explicitly identified regimes such as finite difference frequency Δω\Delta\omega, amplitudes comparable to the barrier scale, or noise intensities outside Axm/D≪1Ax_{m}/D\ll 1 - rather than only in the adiabatic limit.
  2. Obtain a general, checkable criterion for selecting the scanning frequency ωc\omega_{c} and the system/implementation parameters (μ\mu and the gains realizing the bistable nonlinearity) for a specified signal and noise environment, so that parameter choice is predicted from signal and noise characteristics rather than found by trial, simulation sweeps, or metaheuristic optimization.
  3. Characterize the performance limits of the method within this model class (the symmetric quartic bistable system with additive Gaussian white noise and the mixed drive xn(t)cos⁡(ωct)x_{n}(t)\cos(\omega_{c}t)) - the conditions under which modulation-based SR detection fails or is dominated by alternatives.

A result restricted to identified regimes, signal classes, or model systems must state the restriction explicitly, and the adiabatic window must not be silently identified with the method's full scope. The three results must be mutually consistent, and the performance formulas must reduce to the known adiabatic expressions in the appropriate limit.

References

  1. Wenli Zhao, Linze Wang, Jian Fan, 'Theory and method for weak signal detection in engineering practice based on stochastic resonance', International Journal of Modern Physics B 31(28), 1750212 (2017). DOI: 10.1142/S0217979217502125, https://doi.org/10.1142/S0217979217502125 (source paper)
  2. Luca Gammaitoni, Peter Hänggi, Peter Jung, Fabio Marchesoni, 'Stochastic resonance', Reviews of Modern Physics 70, 223 (1998). DOI: 10.1103/RevModPhys.70.223, https://doi.org/10.1103/RevModPhys.70.223
  3. M. I. Dykman, H. Haken, Gang Hu, D. G. Luchinsky, R. Mannella, P. V. E. McClintock, C. Z. Ning, N. D. Stein, N. G. Stocks, 'Linear response theory in stochastic resonance', Physics Letters A 180(4-5), 332-336 (1993). DOI: 10.1016/0375-9601(93)91186-9, https://doi.org/10.1016/0375-9601(93)91186-9
  4. Lizhu Zhang, Fuzhong Wang, Zhang Hui, 'Detecting High Frequency Weak Signal Using Parameter-Adjusted Frequency Shifting Based on Stochastic Resonance (SR)', Physics Procedia 24, 382-389 (2012). DOI: 10.1016/j.phpro.2012.02.056, https://doi.org/10.1016/j.phpro.2012.02.056
  5. Zhi-Hui Lai, Yong-Gang Leng, 'Generalized Parameter-Adjusted Stochastic Resonance of Duffing Oscillator and Its Application to Weak-Signal Detection', Sensors 15(9), 21327 (2015). DOI: 10.3390/s150921327, https://doi.org/10.3390/s150921327
  6. Jinjun Liu, Yonggang Leng, Zhihui Lai, Shengbo Fan, 'Multi-Frequency Signal Detection Based on Frequency Exchange and Re-Scaling Stochastic Resonance and Its Application to Weak Fault Diagnosis', Sensors 18(5), 1325 (2018). DOI: 10.3390/s18051325, https://doi.org/10.3390/s18051325
  7. Yunjiang Liu, Fuzhong Wang, Lu Liu, Yamin Zhu, 'Secondary signal-induced large-parameter stochastic resonance for feature extraction of mechanical faults', International Journal of Modern Physics B 33(15), 1950157 (2019). DOI: 10.1142/S0217979219501571, https://doi.org/10.1142/S0217979219501571
  8. Biao He, Yan Huang, Danyang Wang, Bing Yan, Dawei Dong, 'A parameter-adaptive stochastic resonance based on whale optimization algorithm for weak signal detection for rotating machinery', Measurement 136, 658–667 (2019). DOI: 10.1016/j.measurement.2019.01.017, https://doi.org/10.1016/j.measurement.2019.01.017
  9. Si-Hai Zhao, Jiang-Ye Xu, Yu-Xiao Liu, Ze-Xing Zhao, Zhong-Shun Qin, 'Performance Investigation of Stochastic Resonance in Three Types of Asymmetric Bistable System Driven by Trichotomous Noise', Complexity 2020, 8842122 (2020). DOI: 10.1155/2020/8842122, https://doi.org/10.1155/2020/8842122
  10. Gang Zhang, Yujie Zeng, Lifang He, 'Tri-stable stochastic resonance coupling system driven by dual-input signals and its application in bearing fault detection', Physica Scripta 97(4), 045202 (2022). DOI: 10.1088/1402-4896/ac5695, https://doi.org/10.1088/1402-4896/ac5695
  11. Gang Zhang, Xingyu Zhu, Tianqi Zhang, 'Research and application of a novel piecewise unsaturated asymmetric tristable stochastic resonance system', Measurement Science and Technology 33(5), 055902 (2022). DOI: 10.1088/1361-6501/ac5059, https://doi.org/10.1088/1361-6501/ac5059
  12. Gang Zhang, Longmei Cao, Mingjun Wu, Zhaorui Li, 'Stochastic Resonance Effect in Segmented Underdamped Asymmetric Tristable System and its Application in Weak Signal Detection Research', Fluctuation and Noise Letters 24(2), 2550001 (2024). DOI: 10.1142/S0219477525500014, https://doi.org/10.1142/S0219477525500014
  13. Qi Liang, Wenxin Yu, Yujie Deng, Qiumei Xiao, 'Weak Signal Processing Based on Stochastic Resonance and Its Circuits Implementation', International Journal of Circuit Theory and Applications 54(4), 2298-2313 (2025). DOI: 10.1002/cta.70098, https://doi.org/10.1002/cta.70098