Simulation of circuits demonstrating stochastic resonance
Introduction
Noise is usually considered a nuisance in communication and signal processing systems, but via a phenomenon known as stochastic resonance (SR) noise, can assist the detection of a signal. Using the signal-to-noise ratio (SNR) as a measure of the coherence of the output signal, the signature of SR is a sharp increase in the SNR followed by a gradual decrease as the noise is increased. The three main ingredients usually required to observe SR are noise (with correlation time much lower than that of the system), subthreshold periodic signal and a nonlinear system. The nonlinear system is essential since the output would be defined by linear response theory for a linear system, thus the SNR would be proportional at the input and output of such a system. The simplest way to provide a nonlinear system is by introducing a threshold element.
Since its emergence as an explanation for the periodic recurrences on the Earth's climate [1], [2], [3], [4] where the term SR was first coined, SR has traversed many disciplines. These range from electronic systems [5], [6], sensing neurons [7], [8], visual perception [9], [10], [11], [12], bidirectional ring lasers [13] and super conducting quantum loops (SQUIDS) [14] to name a few. For further background, Gammaitoni et al. have written an extensive review [15]. More recently, SR is believed to assist with hearing systems in the auditory nerve [16], [17], [18] while adaptive systems can learn to add the optimal amount of noise to some nonlinear feedback systems [19].
As an example of SR, we will consider the work by Simonotto et al. [9], which deals with the human visual system. This is closely related to the dithering effect [10]. Consider a system that is capable of transmitting single bits of information, each of which marks a threshold crossing. A visual realization of this is shown in Fig. 1 that was generated following the procedure in Ref. [9]. The original gray scale image shown in Fig. 1a is depressed beneath a threshold, white noise added to the gray value in each pixel, and the result compared to the threshold. Thus, the noise is incoherent with that in all other pixels. Pixels with a value above the threshold are made black and the others below are made white. Every pixel contains one bit of information, whether or not the threshold has been crossed. Fig. 1b–d shows the result of adding noise of three intensities, increasing from left to right. There is an optimal noise intensity in (b) that maximizes the information content. Additional improvement in perceived picture quality can be gained by varying the noise temporally [9]. The images in Fig. 1 have been averaged over five ensembles.
A limitation of SR is that it only considers periodic signals; this shortcoming has lead to the development of a method for characterizing SR with aperiodic stimuli [20], where the term aperiodic stochastic resonance (ASR) was coined. Most of the literature regarding ASR to date has considered neuronal models [20], [21], [22], [23], [24], [25], [26], [27].
In this paper, we first describe the types of nonlinear systems and noise that are used. This is followed by algorithms used for numerical simulations. The next two sections replicate SR and ASR, which includes applying ASR to the simplest nonlinear system.
Section snippets
Nonlinear systems and noise
We used noise given by the Ornstein–Uhlenbeck (OU) stochastic process of the formwhere ξ(t) is the white Gaussian noise with mean 〈ξ(t)〉=0 and autocorrelation 〈ξ(t)ξ(s)〉=2Dδ(t−s). The angled brackets 〈·〉 denote an ensemble average. The correlation time of the OU process is τc=λ−1 and the autocorrelation is given bywith a variance of D/τc. The OU process provides control over both noise intensity D, and correlation time τc.
In biological systems, SR has
Numerical simulations
This section explicitly details the algorithms used for numerical analysis. By showing the algorithms, it should enable the reader to easily replicate the simulations.
We can approximate Gaussian white noise by choosing τc equal to the integration step size. The Box–Muller algorithm [32] (, ) were used to generate normalized Gaussian random variables from uniform random variables, and the algorithm described by Eq. (7) was used to integrate the OU process to produce colored noise [33]. This has
Replicating SR
The most common way to quantify SR is through the SNR [36], which is the method used in this paper. The SNR is defined as the ratio of the signal power spectral density to the broadband background noise taken at the signal frequency and is given in decibels aswhere S and B are the signal and background noise at the fundamental frequency f0, respectively. The process used to calculate the SNR given the output signal from a system is shown in Fig. 4. If SR exists we expect
Replicating ASR
A large proportion of work in SR has been limited to systems with periodic stimulus. Although it has served useful in many areas (Section 1), the applicability of SR to in practice is limited. This is due to many real world stimuli being aperiodic.
This limitation leads to the concept of ASR, first coined by Collins et al. [20]. ASR introduces another hindrance, namely, how to measure it. Both the methods used for SR in Section 4 assess the coherence of the response from the system with the
Conclusion
We have explicitly provided the algorithms for the numerical simulations. This should enable the reader to easily replicate a system that exhibits ASR. Using these, it was shown that ASR is present in even the simplest nonlinear system—the LCC. This exposes a wide variety of systems where ASR can be used, not just neuron models. For the first time, we have explicitly demonstrated ASR in an LCC. This is of importance for motion detection models, for example, and for making a clearer performance
Acknowledgements
This work was funded by the Australian Research Council and the Sir Ross and Sir Keith Smith Fund.
References (45)
- et al.
Stochastic resonance in a bistable system
Phys. Lett.
(1983) - et al.
A stochastic resonator is able to greatly improve signal-to-noise ratio
Phys. Lett. A
(1996) - et al.
Noise-assisted signal transmission in a nonlinear electronic comparator: experiment and theory
Signal Proc.
(1997) - et al.
The mechanism of stochastic resonance
J. Phys. A
(1981) Stochastic aspects of climate transitions response to a periodic forcing
Tellus
(1982)- et al.
Stochastic resonance on climate change
Tellus
(1982) Staggering through the ice ages: What made the planet careen between climate extremes?
Science News
(1994)Schmitt trigger: a solvable model of stochastic resonance
Phys. Rev. E
(1993)Stochastic resonance in neuron models
J. Stat. Phys.
(1993)- et al.
Noise enhancement of information transfer in crayfish mechanoreceptors by stochastic resonance
Nature
(1993)
Visual perception of stochastic resonance
Phys. Rev. Lett.
Stochastic resonance and the dithering effect in threshold physical systems
Phys. Rev. E
Stochastic resonance in the perceptual interpretation of ambiguous figures: a neural network model
Phys. Rev. Lett.
Periodic perturbation of ambiguous figure: a neural-network model and a nonsimulated experiment
Nuovo. Cimento.
Observation of stochastic resonance in a ring laser
Phys. Rev. Lett.
Stochastic resonance and the benefits of noise: from ice ages to crayfish and SQUIDs
Nature
Stochastic resonance
Rev. Mod. Phys.
A stochastic model of the electrically stimulated auditory nerve: single-pulse response
IEEE Trans. Biomedical Engng
A stochastic model of the electrically stimulated auditory nerve: pulse-train response
IEEE Trans. Biomedical Engng
The effects of stochastic neural activity in a model predicting intensity perception with cochlear implants: low-rate stimulation
IEEE Trans. Biomedical Engng
Adaptive stochastic resonance
Proc. IEEE
Aperiodic stochastic resonance in excitable systems
Phys. Rev. E
Cited by (20)
Experimental and numerical study of noise effects in a FitzHugh-Nagumo system driven by a biharmonic signal
2013, Chaos, Solitons and FractalsCitation Excerpt :Both the numerical and the experimental results were performed by averaging 200 realizations. Experimentally, the noise was generated with the soundboard of a computer [23,50,54], while for the numerical simulations, the algorithm detailed in [22] was used. To experimentally produce noise, a Matlab routine was developed by using the randn () function to produce gaussian random samples in a “for” loop.
Signal detection for frequency-shift keying via short-time stochastic resonance
2005, Physics Letters, Section A: General, Atomic and Solid State PhysicsCitation Excerpt :Stochastic resonance (SR) is now a well established phenomenon wherein the response of a nonlinear system to a subthreshold periodic input signal can be enhanced by the assistance of noise [1–7]. Since a single-frequency sinusoidal input conveys little information content, this effect has been extended to aperiodic (i.e., broadband) input signals, leading to the term: aperiodic stochastic resonance (ASR) [8–16]. From the point of view of information transmission, an aperiodic information-bearing signal might be associated with analog (amplitude and frequency) modulated signals [17–21], or digitally modulated signals [10,12,20–22,24–29] within the context of SR effects.
An analysis of noise enhanced information transmission in an array of comparators
2002, Microelectronics JournalNoise induced breather generation in a sine-Gordon chain
2009, Journal of Statistical Mechanics: Theory and ExperimentAlgorithm used to detect weak signals covered by noise in pind
2019, International Journal of Aerospace Engineering