1: \documentclass[doublecol]{epl2}
2:
3: \title{Stochastic resonance with weak monochromatic driving: gains above unity induced by high-frequency signals}
4: %\shorttitle{SR with weak monochromatic driving etc} %Insert here a short version of the title if it exceeds 70 characters
5:
6: \author{Jes\'us Casado-Pascual\inst{1} \and David Cubero\inst{1} \and Jos\'e Pablo Baltan\'as\inst{2}}
7: %\shortauthor{J. Casado-Pascual \and } %Insert here the first author if authors exceed 70 characters
8:
9: \institute{
10: \inst{1} F\'{\i}sica Te\'orica, Universidad de Sevilla - Apartado de Correos
11: 1065, Sevilla 41080, Spain\\
12: \inst{2} Departamento de F\'{\i}sica Aplicada II, Universidad de Sevilla - Avda.~Reina Mercedes, 2, Sevilla 41012, Spain
13: }
14: \pacs{05.40.-a}{Fluctuation phenomena, random processes, noise, and Brownian motion}
15: \pacs{05.10.Gg}{Stochastic analysis methods (Fokker-Planck, Langevin, etc.)}
16: \pacs{02.50.-r}{Probability theory, stochastic processes, and statistics}
17:
18:
19:
20: \abstract{
21: We study the effects of a high-frequency (HF) signal on the response of a noisy
22: bistable system to a low-frequency subthreshold sinusoidal signal. We show that, by conveniently choosing the ratio of the amplitude of the HF signal to its frequency,
23: stochastic resonance gains greater than unity can be measured at the low-frequency value. Thus, the addition of the HF signal can entail an improvement in the detection of weak monochromatic signals. The results are explained in terms of an effective model and illustrated by means of numerical simulations.
24: }
25:
26:
27: \begin{document}
28:
29: \maketitle
30:
31: The phenomenon of stochastic resonance (SR) has been studied with
32: growing interest during the last three decades, being found
33: to be of relevance in a great variety of phenomena in physics,
34: chemistry, and the life sciences~\cite{gamjun98}. Roughly speaking, SR
35: consists in the amplification of a weak, time-dependent signal of
36: interest by the concerted actions of noise and the nonlinearity of the
37: system. Several quantifiers have been used to characterise the SR
38: response of noisy systems in the presence of periodic signals. In
39: particular, the nonmonotonic behaviour of the output signal-to-noise
40: ratio (SNR) with the strength of the noise is a widely accepted
41: signature of SR. In addition, a dimensionless quantity known as the SR
42: gain is usually defined as the ratio of the output SNR over the input
43: SNR. The SNR measures the ``quality'' of the signal, in terms of the
44: ratio of its ``coherent'' (periodic) component over its ``incoherent''
45: (noisy) component. In turn, the SR gain compares the ``qualities'' of the
46: output and the input signals. In general, obtaining high output
47: SNRs and SR gains greater than unity would be desirable when using SR
48: as an amplification mechanism. Analog~\cite{loergin96,ginmak01} and
49: numerical~\cite{casdenk03,casgom_pre_03} simulations have shown that
50: noisy bistable systems driven by subthreshold multifrequency forces
51: can display SR gains greater than unity when the parameters of the
52: problem are properly chosen. Moreover, in~\cite{casgom03}, a
53: two-state model of SR has been used to explain these results
54: analytically. By contrast, to the best of our knowledge, there is no
55: evidence of SR gains greater than unity when a subthreshold monochromatic
56: signal drives an isolated bistable system. However, these unusual
57: large gains have been reported in the case of a suprathreshold
58: sinusoidal signal for an isolated bistable system~\cite{hanin00} and in
59: the case of a subthreshold sinusoidal signal for coupled
60: bistable systems~\cite{casgom06}.
61:
62:
63: As it has been already pointed out in the literature~\cite{hanin00},
64: an improvement of the response of a nonlinear system to a subthreshold
65: signal of interest embedded in noise can be achieved better by
66: lowering the threshold value rather than by increasing the noise
67: strength. However, in most cases of practical interest, threshold
68: lowering is usually a difficult task. Recently, it has been shown
69: that the effect of a strong high-frequency (HF) monochromatic force on the
70: overdamped dynamics of a Brownian particle in a bistable potential can
71: be described in terms of an effective potential whose characteristics
72: depend on the parameters of the HF field~\cite{casbal04}. In particular, the barrier height of this effective
73: potential is smaller than the original one. Thus, an HF
74: force provides a mechanism to achieve an effective threshold
75: lowering. As shown in~\cite{chigia05} for a {\em
76: square-wave} signal of interest, it is possible to take advantage of
77: this effective threshold lowering to improve the SR gain values
78: evaluated at the fundamental frequency of the square wave signal. More
79: precisely, in~\cite{chigia05}, SR gains greater than unity are
80: found for a wide range of values of the noise strength. However, as
81: mentioned in the same reference, this positive effect of the
82: HF force seems to be absent when the signal of interest is
83: {\em sinusoidal}. It is pertinent to point out that the effect of HF fields on nonlinear stochastic systems is not necessarily positive
84: (see, for instance,~\cite{bormar05} for a ratchet model and~\cite{cubero06} for an excitable model). Nevertheless, in ratchet models positive effects have been also found~\cite{bormar06}.
85:
86: In this work we show that, by using a strong HF signal, SR
87: gains greater than unity can be also observed in a bistable noisy
88: system when the signal of interest is {\em subthreshold} and {\em
89: sinusoidal}. In order to do that, let us consider a stochastic system characterised by a
90: single degree of freedom $x(t)$, whose dynamics (in dimensionless
91: units) is described by the stochastic differential equation
92: \begin{equation}
93: \dot{x}(t)=-U^{\prime}[x(t)]+F(t)+\xi(t)\,.
94: \label{eq:lang}
95: \end{equation}
96: Herein, $\xi(t)$ is a Gaussian white noise of zero mean and
97: autocorrelation function $\langle \xi(t)\xi(s)\rangle=2D\delta(t-s)$,
98: $D$ being the noise strength, $F(t)$ is the sinusoidal driving
99: $F(t)=A\cos(\Omega t)$,
100: and $U^{\prime}(x)$ denotes the derivative with respect to $x$ of the
101: quartic potential $U(x)=x^4/4-x^2/2$.
102: Henceforth, we will restrict ourselves to subthreshold
103: driving signals. More precisely, we will assume that $A\le
104: A_{\mathrm{th}}=2/\sqrt{27}$, where $A_{\mathrm{th}}$ is the static
105: threshold value. In this case, the time dependent potential
106: $U(x)-A\,x\cos(\Omega t)$ possesses two local minima separated by a local maximum
107: for any instant of time $t$. Notice that the dynamical threshold value
108: (defined as the maximum value of $A$ such that no interwell
109: transitions are possible in the absence of noise) always exceeds this
110: adiabatic threshold $A_{\mathrm{th}}$.
111:
112: Equation~(\ref{eq:lang}) describes the overdamped
113: motion of a particle in a symmetric double well potential driven by a
114: sinusoidal force and noise. Alternatively,
115: $x^{({\mathrm{in}})}(t)=F(t)+\xi(t)$ may be interpreted as an input signal
116: consisting of a noisy term $\xi(t)$ and a periodic term $F(t)$ (signal
117: of interest). This input signal is transformed into an output signal
118: $x^{({\mathrm{out}})}(t)=x(t)$ after being processed by the nonlinear
119: device characterised by $U(x)$. In this context, SR can be understood as a mechanism in
120: which noise plays a positive role in the optimal detection of some
121: features of the signal of interest $F(t)$ in the output signal $x(t)$.
122:
123:
124: Several quantifiers have been used to characterise the SR
125: phenomenon. Here, we will only consider those quantifiers involving
126: the power spectral density (PSD) of the input and output signals,
127: namely, the SNR of the output signal and the SR gain. According to the
128: generalisation of the Wiener-Khinchine theorem to a periodically
129: driven stochastic process, the PSD of the output signal $x(t)$ is
130: given by~\cite{jung93}
131: \begin{equation}
132: S^{(\mathrm{out})}(\omega)=\frac{2}{\pi}\int_{0}^{\infty} \mathrm{d}\tau\,
133: C(\tau)\cos(\omega\tau),
134: \label{eq:sout}
135: \end{equation}
136: with $\omega\ge 0$, where $C(\tau)=T^{-1}\int_{0}^{T} \mathrm{d}t\,\langle
137: x(t+\tau)x(t)\rangle_{\infty}$, $\langle\cdots\rangle_{\infty}$
138: representing the average over the realizations of the noise after a
139: relaxation transient stage. It can be proved that
140: $S^{(\mathrm{out})}(\omega)$ consists of a series of deltalike spikes
141: located at the odd harmonics of the fundamental frequency $\Omega$,
142: superimposed on a background PSD \cite{gamjun98}. This background PSD (also called
143: incoherent part of the PSD) is given by
144: \begin{equation}
145: S^{(\mathrm{out})}_{\mathrm{incoh}}(\omega)=\frac{2}{\pi}\int_{0}^{\infty}
146: \mathrm{d}\tau\, C_\mathrm{incoh}(\tau)\cos(\omega\tau),
147: \label{eq:sincoh}
148: \end{equation}
149: where $C_\mathrm{incoh}(\tau)=C(\tau)-C_\mathrm{coh}(\tau)$, with
150: $C_\mathrm{coh}(\tau)=T^{-1}\int_{0}^{T} \mathrm{d}t\,\langle
151: x(t+\tau)\rangle_\infty\langle x(t)\rangle_{\infty}$ (see, for
152: instance,~\cite{casgom05}). Then, the output SNR is defined as \cite{gamjun98,casgom05}
153: \begin{equation}
154: R^\mathrm{(out)}=\lim_{\epsilon\rightarrow
155: 0^+}\frac{\int_{\Omega-\epsilon}^{\Omega+\epsilon}\mathrm{d}\omega\,
156: S^{(\mathrm{out})}(\omega)}{S^{(\mathrm{out})}_\mathrm{incoh}(\Omega)}\,.
157: \label{eq:snrout}
158: \end{equation}
159:
160: The input PSD $S^{(\mathrm{in})}(\omega)$, as well as
161: $S^{(\mathrm{in})}_{\mathrm{incoh}}(\omega)$ and $R^\mathrm{(in)}$,
162: can be obtained from the above expressions by replacing $x(t)$ by
163: $x^{(\mathrm{in})}(t)$. In particular, it is easy to prove that the input
164: SNR reads
165: \begin{equation}
166: R^\mathrm{(in)}=\frac{\pi A^2}{4D}.
167: \label{eq:snrin}
168: \end{equation}
169: Finally, the SR gain is defined as
170: \begin{equation}
171: G=\frac{R^\mathrm{(out)}}{R^\mathrm{(in)}}\,.
172: \label{eq:gain}
173: \end{equation}
174: As mentioned before, obtaining SR gains greater than unity would be desirable. For this purpose, let us introduce a strong HF signal of the form:
175: \begin{equation}
176: Y(t)=N\Omega r \cos(N \Omega t+\varphi)\,,
177: \label{eq:cs}
178: \end{equation}
179: where the parameter $N$ is a positive integer, $r$ is the ratio of the
180: amplitude of $Y(t)$ to its frequency, and $\varphi$ is an arbitrary
181: initial phase. We are interested in situations in which the parameters
182: $N\Omega r$ and $N\Omega$ appearing in this monochromatic force are
183: much larger than the rest of the parameters in the problem. That is
184: why $Y(t)$ is called a strong HF signal. This situation
185: can be formally achieved by taking the limit $N\rightarrow\infty$,
186: with the ratio $r$ kept fixed. The HF signal $Y(t)$ must
187: be summed to the right hand side of eq.~(\ref{eq:lang}), so that now
188: $x(t)$ fullfils the Langevin equation
189: \begin{equation}
190: \dot{x}(t)=-U^{\prime}[x(t)]+F(t)+Y(t)+\xi(t)\,.
191: \label{eq:langHF}
192: \end{equation}
193: As a consequence, the input signal in our previous discussion must be
194: replaced by $x^{({\mathrm{in}})}(t)=F(t)+Y(t)+\xi(t)$. It is easy to
195: prove that the input PSD in the presence of the HF signal
196: [$S^{(\mathrm{in})}(\omega;r)$ with $r\ne 0$] can be expressed in
197: terms of that obtained in its absence [$S^{(\mathrm{in})}(\omega;0)$]
198: as
199: \begin{equation}
200: S^{(\mathrm{in})}(\omega;r)=S^{(\mathrm{in})}(\omega;0)+
201: \frac{N^{2}r^2\Omega^2}{2}\delta(\omega-N\Omega)\,.
202: \label{eq:sx_input}
203: \end{equation}
204: It should be emphasised that we are interested in the evaluation of
205: the input and output SNRs at the frequency $\Omega$ [frequency of the
206: signal of interest $F(t)$], $Y(t)$ being just a tool introduced to
207: improve the response features of the system at that frequency. In
208: addition, since in the definition of $R^{(\mathrm{in})}$ only the PSD
209: around the frequency $\Omega$ plays a role, it is clear from
210: eq.~(\ref{eq:sx_input}) that the input SNR is given by
211: eq.~(\ref{eq:snrin}) even in the presence of the HF signal
212: $Y(t)$, as in this case $N\gg 1$.
213:
214: For reasons that will be clarified later, it is convenient to define
215: the new stochastic process $z(t)=x(t)-r\sin(N\Omega t+\varphi)$. After replacing this definition in eq.~(\ref{eq:sout}),
216: it can be proved that the output PSD, $S^{(\mathrm{out})}(\omega)$,
217: and the PSD of the process $z(t)$, $S_z(\omega)$, are related
218: according to
219: \begin{equation}
220: S^{(\mathrm{out})}(\omega)=S_{z}(\omega)+
221: 2r\left[\frac{r}{4}-\mathrm{Im}(e^{-i\varphi}Z_N)\right]
222: \delta(\omega-N\Omega)\,,
223: \label{eq:sx_sxhat}
224: \end{equation}
225: where $Z_N$ is the Fourier component of the process $z(t)$ at the
226: frequency $N\Omega$, i.e., $Z_{N}=T^{-1}\int_{0}^{T} \mathrm{d}t\,\langle
227: z(t)\rangle_{\infty}e^{-i N \Omega t}$. From eq.~(\ref{eq:sx_sxhat})
228: it is clear that, in order to evaluate quantities which only depend on
229: the behaviour of the output PSD around the frequency $\Omega$ [as for
230: instance $R^{(\mathrm{out})}$] both processes, $x(t)$ and $z(t)$, are
231: equivalent as long as $N\ne 1$.
232:
233: The dynamics of the stochastic process $z(t)$ is described by
234: the equation
235: \begin{equation}
236: \dot{z}(t)=-U^{\prime}[z(t)+r\sin(N\Omega
237: t+\varphi)]+F(t)+\xi(t)\, .
238: \label{eq:lang3}
239: \end{equation}
240: As we have mentioned before, we are interested in the limit
241: $N\rightarrow \infty$, with $r$ kept fixed. From eq.~(\ref{eq:lang3}),
242: it follows that the time derivative of the process $z(t)$ is at most
243: of order $1$ as $N\rightarrow\infty$. In this sense, we will say that
244: this process is {\em slow}. In contrast, according to the definition of $z(t)$,
245: it is clear that, in this limit, the process $x(t)$ is
246: {\em fast}, in the sense that it is a highly oscillating process
247: around $z(t)$. As the process $z(t)$ is {\em slow}, a large number of
248: oscillations of the function $r\sin(N\Omega t+\varphi)$ appearing in
249: $U^{\prime}[z(t)+r\sin(N\Omega t+\varphi)]$ takes place before a
250: significant change in $z(t)$ occurs. Therefore, for $N\gg 1$, $z(t)$
251: must be almost independent of the phase $\varphi$ and, consequently,
252: $z(t,\varphi)\simeq\bar{z}(t)\mathrel{\mathop:}=(2\pi)^{-1}
253: \int_{0}^{2\pi}\mathrm{d}\varphi^{\prime}\,z(t,\varphi^{\prime})$, where we have written
254: explicitly the dependence of $z(t)$ with respect to $\varphi$. Taking
255: into account this property and carrying out the phase average in
256: eq.~(\ref{eq:lang3}), one obtains
257: \begin{equation}
258: \dot{\bar{z}}(t)=-U_{\mathrm{eff}}^{\prime}[\bar{z}(t)]+F(t)+\xi(t)\, ,
259: \label{eq:eff}
260: \end{equation}
261: where we have introduced the effective potential
262: \begin{eqnarray}
263: U_{\mathrm{eff}}[\bar{z}]&\mathrel{\mathop:}=&
264: \frac{1}{2\pi}\int_0^{2\pi}\mathrm{d}\varphi\,
265: U[\bar{z}+r\sin(N\Omega t+\varphi)]\nonumber \\
266: &=&\frac{\bar{z}^4}{4}-c(r)\frac{\bar{z}^2}{2}+\frac{r^2}{4}\left[\frac{3 r^2}{8}-1\right]\, ,
267: \label{eq:eff_pot}
268: \end{eqnarray}
269: with $c(r)=1-3r^2/2$. From this result, it is clear that the stability
270: of this effective potential depends on the ratio $r$. More precisely,
271: if $r<\sqrt{2/3}$ the potential is bistable, whereas if
272: $r\ge\sqrt{2/3}$ is monostable. Thus, an increase in $r$ leads to a
273: decrease in the effective barrier height and, eventually, to its
274: disappearance.
275:
276: Henceforth, we will restrict our study to the case $r<\sqrt{2/3}$,
277: which corresponds to $0<c(r)\le 1$ (bistable potential). In order to
278: reduce the effective Langevin equation [eq.~(\ref{eq:eff})] to the standard form [eq.~(\ref{eq:lang})], it is convenient to introduce the following time and space scale changes:
279: \begin{eqnarray}
280: \tilde{t}&=&t\,c(r)\, , \label{eq:t_resc}\\
281: \tilde{z}(\tilde{t})&=&\bar{z}
282: \left[\tilde{t}/c(r)\right]/\sqrt{c(r)}\, \label{eq:z_resc}.
283: \label{eq:xscale}
284: \end{eqnarray}
285: Then, it is easy to prove that the stochastic process $\tilde{z}(\tilde{t})$ fullfils the Langevin equation (\ref{eq:lang}) but for the rescaled values of the amplitude, noise strength, and frequency given by $A/c(r)^{3/2}$, $D/c(r)^{2}$, and $\Omega/ c(r)$, respectively. We will denote by $f_r$ the mapping which assigns to each set of parameter values $(A,D,\Omega)$ the corresponding rescaled values,
286: and by $f_r^{-1}$ its inverse mapping, i.e.,
287: \begin{eqnarray}
288: \label{mapping}
289: f_r(A,D,\Omega)&=&\left[A/c(r)^{3/2},D/c(r)^{2},\Omega/c(r)\right], \\
290: \label{inversemapping}
291: f_r^{-1}(A,D,\Omega)&=&\left[A c(r)^{3/2},D c(r)^{2},\Omega c(r)\right].
292: \end{eqnarray}
293: In fig.~\ref{fig:scaling} we depict how two regions of the parameter plane $D-A$, $R_1$ and $R_2$, are transformed into each other by the mappings $f_r$ and $f_r^{-1}$.
294: \begin{figure}
295: \onefigure[scale=0.30]{f1.eps}
296: \caption{Two regions, $R_1$ and $R_2$, of the parameter plane $D-A$ are transformed into each other by the mappings
297: $f_r$ and $f_r^{-1}$. $R_1$ is the subthreshold region delimited by the four vertices
298: $(D,A/A_{\mathrm{th}})=(0.005,0.1)$, $(0.015,0.1)$, $(0.015,0.3)$, and $(0.005,0.3)$, and $R_2$ is obtained
299: from $R_1$ by applying $f_r$ with $r=0.800529$. This last parameter value has been chosen so
300: that $R_2$ lies within a region wherein SR gains greater than
301: unity have been observed in the case of a bistable system in the
302: presence of a suprathreshold sinusoidal signal with frequency
303: $\Omega=0.1$ (see fig.~2 in~\cite{hanin00}).}
304: \label{fig:scaling}
305: \end{figure}
306: \begin{figure}
307: \onefigure[scale=0.30]{f2.eps}
308: \caption{Dependence on the noise strength $D$ of the SR gain $G$ in a bistable noisy system driven by a sinusoidal subthreshold signal, both in the absence (empty diamonds) and presence (full diamonds) of an HF signal. The parameter values of the sinusoidal subthreshold signal $F(t)$ and the HF signal $Y(t)$ are: $A=0.1$, $\Omega=0.005$, $r=0.797$, $N=4000$, and $\varphi=0$. The SR gain obtained from the numerical solution of the effective dynamics in eq.~(\ref{eq:eff}) is plotted with solid line. The horizontal dashed line marks the unity for the SR gain.}
309: \label{fig:sin}
310: \end{figure}
311:
312: As mentioned before, for $N\ne 1$, the output SNR of the
313: processes $x(t)$ and $z(t)$ coincide exactly. Furthermore, for $N\gg
314: 1$, the processes $z(t)$ and $\bar{z}(t)$ give rise to the same value
315: of this quantifier with a great deal of approximation. Finally, the
316: process $\bar{z}(t)$ is related to the process $\tilde{z}(\tilde{t})$
317: [which is the process $x(t)$ but in the absence of the
318: strong HF signal and for the above mentioned rescaled parameter values] by means of the transformations (\ref{eq:t_resc}) and (\ref{eq:z_resc}). Then, by using the
319: definition of the output SNR, one obtains that
320: \begin{equation}
321: R^{(\mathrm{out})}\left[(A,D,\Omega);r\right]\simeq c(r)R^{(\mathrm{out})}\left[f_r(A,D,\Omega);0\right]
322: \label{eq:rout_resc}
323: \end{equation}
324: for $N\gg 1$, as expected from the change of scale in eq.~(\ref{eq:t_resc}) and the fact that the SNR has dimensions of time$^{-1}$.
325: The above expression relates the output SNR in the presence of the
326: HF signal, $R^{(\mathrm{out})}[(A,D,\Omega);r]$, with that
327: corresponding to a new set of parameter values but in the absence of
328: the HF signal, $R^{(\mathrm{out})}\left[f_r(A,D,\Omega);0\right]$.
329:
330: Analogously, taking into account
331: the mapping in eq.~(\ref{mapping}), as well as
332: the fact that the input SNR at the frequency of interest $\Omega$ is
333: given by eq.~(\ref{eq:snrin}) even in the presence of the
334: HF signal [see the paragraph below
335: eq.~(\ref{eq:sx_input})], it is also clear that
336: \begin{eqnarray}
337: R^{(\mathrm{in})}[(A,D,\Omega);r]&=&R^{(\mathrm{in})}[(A,D,\Omega);0]\nonumber\\ &=&c(r)R^{(\mathrm{in})}\left[f_r(A,D,\Omega);0\right]\, .
338: \label{eq:rin_resc}
339: \end{eqnarray}
340: Finally, from eqs.~(\ref{eq:rout_resc}), (\ref{eq:rin_resc}), and the definition (\ref{eq:gain}), one obtains that
341: \begin{equation}
342: G[(A,D,\Omega);r]\simeq G\left[ f_r(A,D,\Omega);0\right]
343: \label{eq:gain_resc}
344: \end{equation}
345: for $N\gg 1$, this last equation being just a consequence of the dimensionless
346: character of the SR gain. Equation~(\ref{eq:gain_resc}) is one of the
347: main results of this paper. It shows that the value of the SR gain in
348: the presence of an HF signal is approximately equal to that
349: corresponding to a system in which that HF signal is
350: absent, but for a different set of parameter values. Obviously, the larger the value of $N$, the better the approximation.
351:
352: From eq.~(\ref{eq:gain_resc}) it is easy to prove that, by applying an appropriate strong HF signal to a bistable noisy system driven by a subthreshold sinusoidal signal of interest, one can obtain SR gains greater than unity.
353: Indeed, as shown in~\cite{hanin00}, SR gains greater
354: than unity can be observed in a noisy bistable system driven by a
355: suprathreshold sinusoidal signal. For instance, in fig.~2 of~\cite{hanin00} it is shown that, for a fixed value of the frequency of
356: this signal $\Omega=0.1$, there exists a region of the parameter plane $D-A$ wherein $G>1$. To fix ideas, let us assume that $G[(A_2,D_2,\Omega_2);0]>1$, with $A_2>A_{\mathrm{th}}$, and let us choose a value $r_0$ such that
357: $\sqrt{2\left[1-(A_{\mathrm{th}}/A_2)^{2/3} \right]/3}<r_0<\sqrt{2/3}$. If we define $(A_1,D_1,\Omega_1)=f_{r_0}^{-1}(A_2,D_2,\Omega_2)$, it is easy to check that $A_1<A_{\mathrm{th}}$. Furthemore, according to eq.~(\ref{eq:gain_resc}), $G[(A_1,D_1,\Omega_1);r_0]\simeq G[(A_2,D_2,\Omega_2);0]>1$, for $N\gg 1$. Thus, by applying an HF signal of the type given in eq.~(\ref{eq:cs}) (with $\Omega=\Omega_1$, $r=r_0$, $N\gg 1$, and an arbitrary value of $\varphi$) to a noisy bistable system driven by a subthreshold sinusoidal signal of frequency $\Omega_1$ and amplitude $A_1$, one gets an SR gain greater than unity at a noise strength $D_1$. This is exactly what we wanted to prove.
358:
359: As an example of the above reasoning, let us consider fig.~\ref{fig:scaling}. There, $R_2$ lies within a region of the parameter plane $D-A$ wherein SR gains greater than
360: unity have been observed in the case of a bistable system in the
361: presence of a suprathreshold sinusoidal signal with frequency
362: $\Omega=0.1$ (see fig.~2 in~\cite{hanin00}). Therefore, according to our reasoning, it is expected to observe SR gains greater than unity within the subthreshold region $R_1$ and for the frequency value $\Omega=0.003873$, whenever one applies an HF signal of the type given in eq.~(\ref{eq:cs}), with $r=0.800529$, $N\gg 1$, and an arbitrary value of $\varphi$.
363:
364: In order to illustrate the feasibility of the above argument, in fig.~\ref{fig:sin} we depict the dependence on the noise strength $D$ of the SR gain in a bistable noisy system driven by a sinusoidal subthreshold signal, both in the absence (empty diamonds) and presence (full diamonds) of an HF signal. The SR gain has been evaluated numerically following the method described in~\cite{casdenk03}. The parameter values of the sinusoidal subthreshold signal are $A=0.1$ and $\Omega=0.005$. As expected, in the absence of the HF signal, the SR gain is lower than unity for all the noise strength values. Nevertheless, when an HF signal of the type given in eq.~(\ref{eq:cs}) is applied, with $\Omega=0.005$, $r=0.797$, $N=4000$, and $\varphi=0$, there is a
365: range of noise strength values for which SR gains greater than unity are observed, the maximum value of $G$ being located around $D_{\mathrm{max}}= 0.01$. Since our argument is premised on the equivalence of the exact and the effective dynamics [eqs.~(\ref{eq:langHF}) and (\ref{eq:eff}), respectively] for evaluating the SR quantifiers in the asymptotic limit $N\to \infty$, it is important to check the validity of this approximation for $N=4000$. In fig.~\ref{fig:sin}, we have also plotted with solid line the SR gain obtained from the numerical solution of the effective dynamics in eq.~(\ref{eq:eff}). A glance at that figure reveals that the agreement between both dynamics is excellent for $N=4000$.
366:
367:
368: It is worth noting that the rescaled parameter values of the monochromatic signal in fig.~\ref{fig:sin} [i.e., $A/c(r)^{3/2}\approx 25.347 A_{\mathrm{th}}$ and $\Omega/c(r)\approx 0.106$] are almost equal to the amplitude and frequency of the suprathreshold sinusoidal signal of the dot-dashed line in fig.~5 of~\cite{hanin00}. Thus, according to eq.~(\ref{eq:gain_resc}), the above mentioned solid and dot-dashed lines must be essentially the same, except for a rescaling of the $D$-axis by a factor $1/c(r)^{2}\approx 449.122$. So, for instance, the location of the maximum of the SR gain in the dot-dashed line is approximately equal to $D_{\mathrm{max}}/ c(r)^{2}\approx 4.491$, and the maximum value of the SR gain is approximately $1.2$ in both curves.
369:
370: In conclusion, in this work we have shown that when a noisy input containing a subthreshold
371: monochromatic signal of interest is processed by a bistable system in the
372: presence of an HF signal, SR gains greater than unity can be
373: observed. In order to explain this result, we have proved that the original bistable
374: system in the presence of the HF signal can be approximated by a new
375: bistable system obeying the same stochastic dynamics as the original one but in the absence
376: of the HF signal. In this new system, the amplitude and
377: frequency of the signal of interest, as well as the noise strength, appear
378: rescaled by coefficients depending on the ratio $r$ of the amplitude of the
379: HF signal to its frequency. The main result of the paper
380: establishes that the SR gains of both systems are approximately equal. The
381: importance of this result relies on the fact that, although the
382: original parameter values lie within a subthreshold region, the rescaled parameter values obtained from a convenient choice of $r$ can belong to a suprathreshold region wherein SR gains greater than unity have been observed. As the SR gains in both
383: systems approximately coincide, it is apparent that the addition of the HF signal can give rise to SR gain values greater than unity. Finally, the theoretical results have been illustrated in a
384: particular case by means of numerical simulations. The authors are confident that the results reported in this paper can be useful for the optimal detection of weak monochromatic signals in the great variety of systems modeled by noisy bistable potentials.
385:
386:
387:
388:
389:
390:
391:
392: \acknowledgments
393: The research was supported by the Direcci\'on General de Ense\~nanza
394: Superior of Spain (Grant No. FIS2005-02884) and the Junta de Andaluc\'{\i}a (J. C.-P. and D. C.), the Juan de la Cierva program of the Ministerio de Ciencia y Tecnolog\'{\i}a (D. C.), and the Consejer\'{\i}a de Educaci\'on y Ciencia de la Junta de Andaluc\'{\i}a (FQM-276) (J. P. B).
395:
396: \begin{thebibliography}{0}
397: \bibitem{gamjun98}
398: \Name{Gammaitoni L., H{\"a}nggi P., Jung P. \and Marchesoni F.}
399: \REVIEW{Rev. Mod. Phys.}{70}{1998}{223}.
400:
401: \bibitem{loergin96}
402: \Name{Loerincz K., Gingl Z. \and Kiss L. B.}
403: \REVIEW{Phys. Lett. A}{224}{1996}{63}.
404:
405: \bibitem{ginmak01}
406: \Name{Gingl Z., Makra P. \and Vajtai R.}
407: \REVIEW{Fluct. Noise Lett.}{1}{2001}{L181}.
408:
409: \bibitem{casdenk03}
410: \Name{Casado-Pascual J., Denk C., G\'omez-Ord\'o\~nez J., Morillo M. \and H\"anggi P.}
411: \REVIEW{Phys. Rev. E}{67}{2003}{036109}.
412:
413: \bibitem{casgom_pre_03}
414: \Name{Casado-Pascual J., G\'omez-Ord\'o\~nez J., Morillo M. \and H\"anggi P.}
415: \REVIEW{Phys. Rev. E}{68}{2003}{061104}.
416:
417: \bibitem{casgom03}
418: \Name{Casado-Pascual J., G\'omez-Ord\'o\~nez J., Morillo M. \and H\"anggi P.}
419: \REVIEW{Phys. Rev. Lett.}{91}{2003}{210601}.
420:
421: \bibitem{hanin00}
422: \Name{H\"anggi P., Inchiosa M. E., Fogliatti D. \and Bulsara A.R.}
423: \REVIEW{Phys. Rev. E}{62}{2000}{6155}.
424:
425: \bibitem{casgom06}
426: \Name{Casado J. M., G\'omez Ord\'o\~nez J. \and Morillo M.}
427: \REVIEW{Phys. Rev. E}{73}{2006}{011109}.
428:
429: \bibitem{casbal04}
430: \Name{Casado-Pascual J. \and Baltan\'as J. P.}
431: \REVIEW{Phys. Rev. E}{69}{2004}{046108}.
432:
433: \bibitem{chigia05}
434: \Name{Chizhevsky V. N. \and Giacomelli G.}
435: \REVIEW{Phys. Rev. A}{71}{2005}{011801}.
436:
437: \bibitem{bormar05}
438: \Name{Borromeo M. \and Marchesoni F.}
439: \REVIEW{Europhys. Lett.}{72}{2005}{362}.
440:
441: \bibitem{cubero06}
442: \Name{Cubero D., Baltan\'as J. P. \and Casado-Pascual J.}
443: \REVIEW{Phys. Rev. E}{73}{2006}{061102}.
444:
445: \bibitem{bormar06}
446: \Name{Borromeo M. \and Marchesoni F.}
447: \REVIEW{Phys. Rev. E}{73}{2006}{016142}.
448:
449: \bibitem{jung93}
450: \Name{Jung P.}
451: \REVIEW{Phys. Rep.}{234}{1993}{175}.
452:
453: \bibitem{casgom05}
454: \Name{Casado-Pascual J., G\'omez-Ord\'o\~nez J. \and Morillo M.}
455: \REVIEW{Chaos}{15}{2005}{026115}.
456:
457:
458: \end{thebibliography}
459:
460: \end{document}
461:
462: