2009 International Conference on Test and Measurement

The High-frequency Weak Signal Detection Based on Stochastic Resonance

Abstract-Aiming at the issue of the traditional stochastic resonance only applicable to deal with low-frequency signals, a high-frequency weak signal detection method based on stochastic resonance is proposed. By analyzing the relationship among the traditional stochastic resonance model parameters, the input signal amplitude, the noisy signal and the signal to noise ratio of output, an additional gain is added to the original model which makes the sampling time multiplied reduced in the stochastic resonance system, in order to achieve the mapping from highfrequency signal to low-frequency signal. Studies have shown that the improved stochastic resonance model has accurately detected the high-frequency input signal. Keywords-stochastic resonance; weak signal; high-frequency; SNR of output
I. INTRODUCTION

can be used to find out the failure precursor of some mechanical. In general, the maximum of SNR of output corresponds to the strongest moment of stochastic resonance. In this paper, we will analysis the relationship among the system parameters, the input signal amplitude and the noise signal, when the SNR of output reaching the maximum. By adjusting the various parameters we identify a set of low-frequency band for specific parameters of stochastic resonance, and then refine the model to achieve the mapping from measured high-frequency signal to a low-frequency band, achieving the purpose of the stochastic resonance detecting weak signal in the context of high-frequency.
II. THE PRINCIPLE OF STOCHASTIC RESONANCE

The idea of stochastic resonance (referred to as SR) was proposed by Benzi and others according to the issue of periodic emergence of the ancient glaciers weather in 1981 [1]. In 1983, Fanve and Heslot confirmed the phenomenon of stochastic resonance for the first time by studying the noise dependent on the spectrum in the AC drive Schmitt trigger. In 1988, McNamrar confirmed the existence of stochastic resonance in dual-stable laser ring. The early 90s Collins proposed nonperiodic stochastic resonance theory [3], making stochastic resonance and signal processing combine. Subsequently, the stochastic resonance has been substantially applied in radar, sonar, image processing, speech recognition and other fields. The traditional weak signal detection methods all suppress the noise as much as possible to extract the weak signal, while the characteristic of stochastic resonance is the noise energy transferring to the signal energy when the input signal, noise and non-linear system reach a certain match. When the signal to noise ratio(SNR) of output in stochastic resonance system reaches the maximum, can identify the weak desired signal. Using the characteristic of stochastic resonance, numbers of scholars apply stochastic resonance into different areas of signal processing currently, e.g., chaotic systems [4], mechanical engineering fault [5,6], electronic circuits [7] and soon. At present, the weak signal detection based on the stochastic resonance has made great success, particularly in the field of machinery fault diagnosis. The identified weak signal

A.

The Basic Principle ofStochastic Resonance Stochastic resonance consists of three basic elements: Input

weak signal, noise and bistable nonlinear system. s(t) is set to the useful weak periodic signal, and n(t) is set to the noise signal. When bistable nonlinear system synergies in both
s(t) and n(t) , the output signal x(t) will results in the phenomenon of stochastic resonance.

Bistable nonlinear system can be described by a Langevin equation [8]:

= ax(t) -

bx' (t) + s(t) + n(t)

(1)

Among

them,

s(t) = Asin(2Jrjt+r)

=0

E[n(t)n(t - r)]

=(j28(t)

The corresponding potential function of (1) as follows:

= _ax2 --bx 4 + [s(t) + n(t)]x

1 2

1 4

(2)

The potential function can be described in a double-well potential [9] diagram shown in Fig. 1.

978-1-4244-4700-8/09/$25.00 2009 IEEE 431

ICTM2009

U (x )

A.

f..U

The System Parameters Selection ofStochastic Resonance From the relationship among the signal amplitude, the noise intensity and stochastic resonance we can see, when the system parameters remain the same, the signal amplitude and noise intensity must meet a certain scope of relationship to make system synergy occurs. The smaller the signal amplitude is, the greater the required noise intensity is. When the signal is getting smaller and smaller with a certain proportion of the noise intensity, the system does not generate stochastic resonance, as the signal amplitude is too small to guide particle crossing the barrier.

Figure I.

Potential function chart

Potential function curve is composed by two potential trap points (-xm,xm) and a barrier point (coordinate origin). When both the input signal amplitude A and the noise intensity D are zero, the potential well point is
2

Xm

=~a / b ,

and the

barrier height is t.U = a / 4b . The curve describes an overdamped particle movement : when A is equal to 0, the particle in the noise-driven flips between two potential well, and when A is greater than 0, the cycle change of signal and the trap between the flips of noise-driven are likely to be synchronized, simultaneously to occur stochastic resonance phenomenon.
B. The Simulink Model ofStochastic Resonance The bistable nonlinear differential equation driven by the weak periodic force and white noise described as: x(t) = ax(t) - bx3 (t ) + A sin(2JTft + r) + n(t) (3)

Stochastic resonance theory analysis shows that the barrier is too high will lead to larger damping force of particle movement, causing the energy of input signal and noise is difficult to promote particle crossing the barrier. As it can not produce stochastic resonance effect, the SNR of output will decrease gradually. On the contrary, the barrier is too low will result in bistable system tending to a single steady-state system. Also it can not produce stochastic resonance effect, and thus the SNR of output will decrease . We can see from (2): the barrier height t.U is determine by the bistable system parameters as a and b , that the barrier height is proportional to

a and inversely proportional to b . Therefore , selecting the appropriate system parameters is a key for the system to achieve stochastic resonance.
In general, the biggest the SNR of output, the most obvious effect of stochastic resonance. Adiabatic approximation [10] can clearly give the SNR of output in bistable stochastic resonance system: (4) Deal with a noisy sine signal, among which A = 0.03 , D = 0.024 . Analysis of (3) we take a =1 and b = 1, and draw the relationship between SNR and parameters as a and b , shown in Fig. 3.
'"

The simulink model of bistable stochastic resonance system is shown in Fig. 2.

"

fI

Figure 2. The simulink model of stochastic resonance system

In this model, the Signal Generator module is triggered to detected weak periodic signal. The Band-Limited White Noise module is used to simulate white noise. The Add module sums them into a mixed input signal. The one-time items and threetime items which fed back from output signal mixed with input signal are overlapped through the Integrator module, and the output is displayed from the Scope module. III.
THE DETECTION OF ANY BAND PERIODIC SIGNAL USING STOCHASTIC RESONANCE

10

03

0.'"

...
05

06

01

08

09

Figure 3. The relationship between system parameters and SNR

Can be seen from Fig. 3, when a taking a fixed value, the values of b have little effect on the SNR. When b taking a fixed value and a taking 0.3, the SNR is max and the system achieves optimal stochastic resonance . So generally take b =1.

432

B.

The Mapping From High-frequency Signal to Lowfrequency Signal

From the adiabatic approximation theory and the corresponding linear theory we can see that, only the noise as input of the bistable stochastic resonance system, the output spectrum energy will concentrate in a low frequency range. And when the signal frequency falls in the range, the noise energy will be transferred to the signal energy, making the cycle elements stand out. Therefore, the original stochastic resonance model is only suitable for dealing with lowfrequency signals. In order to achieve high-frequency weak signal detection, the original model has been improved. In the bistable non-linear system, we make t = K * t l It can not only reduce the sampling time multiply but also reduce the signal frequency. The new model as follows:

''''

0 01 0 01

0 03 00

005
~ 11)

O~

0 01

000

AM

01

(b) The frequency-domain spectrum of output signal Figure 4. The time and frequency domain spectrum of output signal

In the new model, the low-frequency signal mapped from the original high-frequency is fo = K *f . The size of K decide the level from original frequency f into a low-frequency fo . An appropriate choice of K can accurately map high-frequency signal to the low-frequency band which can occurs stochastic resonance. As the role of the gain K in the model, the sample value of noise signal will fall. However, the values of K will not affect the noise signal intensity, so that the system will still generate stochastic resonance with unchanged system parameters. IV.
A.
SIMULATION

Remain parameters unchanged and make the input signal frequency gradually increased to 0.05Hz , the system appears the phenomenon of stochastic resonance. Experiments show that when the system parameters unchanged and the input signal frequency at a low band within a limited scope, the system also generates the stochastic resonance clearly.
B. The Improved Stochastic Resonance Model for Highfrequency Signal

In the original model, we add a gain in front of the Integrator module, and obtain the improved simulink model as shown in Fig. 5.

The Original Stochastic Resonance Modelfor Lowfrequency Signal

Select a noisy sine signal with frequency of 0.0 1Hz , the signal amplitude A = 0.3 , noise intensity D = 0.0024 , numerical calculation step !it = 0.06s .According to the selection principle of the system preferences, select the system parameters a = 0.07 and b = 1 . The system will generate stochastic resonance and obtain the output time and frequency domain spectrum shown in Fig. 4 as follows.

Figure 5. The improved model of stochastic resonance

02

In the frequency band ( 10Hz - 30Hz), we arbitrarily select a sine signal whose amplitude A = 0.15 , the noise intensity D = 0.001 . According to parameters selection principle of stochastic resonance, select suitable system parameters a = 0.045 and b = 1 . Remain the parameters unchanged, from experimental research we can see, when the input signal frequency between O.OIHz and 0.05Hz , the system generates stochastic resonance. Select the gain K =1000 , and correspond the input signal frequency to the low-frequency band (0 .0IHz-0.03Hz). The time and frequency domain spectrum of input noisy signal and output signal are shown in Fig. 6.

(a) The time-domain spectrum of output signal

433

identify. After analyzing of its stochastic resonance, the output signal frequency spectrum can be identified, and the useful frequency after mapping is fa = 0.0075Hz . Thus , the original high-frequency is I=K* 10 = 2000*0.0075 = 15Hz . The calculated frequency matching the frequency of the input weak signal, verifies the feasibility of this method. V.
CONCLUSION

1(' )

(a) The time-domain spectrum of input noisy signal

In this paper, we have analyzed the relationship among the system parameters of the stochastic resonance, input signal amplitude, noise signal and the SNR of output to choose the best system parameters. The role of the best parameters, we have found the time for low-frequency signal to produce stochastic resonance. In order to achieve high-frequency weak signal detection, the original stochastic resonance model has been improved. In the new model, select the appropriate gain for mapping the high-frequency signal to low-frequency signal, resulting in the output signal to identify the original highfrequency signal. Simulation results show that the new model has accurately realized the detection of high-frequency signal. Jianhui Wang thanks "985 Project" Process Industry Automation Technology Innovation Project

(b) The frequency-domain spectrum of input noisy signal

REFERENCES [I) RBenzi, A.Sutera, and A.Vulpiani, The Mechanism of stochastic Resonance. 1.Phys.A, vol.I4, 1981, pp. 453-457 . (2) S.Fauve, F.Heslot. Stochastic Resonance in a Bistable System. Phys.Lett, vo1.97 , 1983, pp. 5-7. (3) B.McNamara, RRoy. Observation of stochastic resonance in a ring laser, Phys. Rev. Lett, 1988. (4) G.D.VanWiggeren, RRoy. Communication with Chaotic Lasers. Science 279, 1998, pp. II 98- 1200. (5) CHEN Min, HU Niaoqing, QIN Guojun, and AN Maochun. Application of Parameter-tuning Stochastic Resonance for Detecting Early Mechanical Faults. Journal of Mechanical Engineering, 2009, pp. 133135. (6) WANG Hong-yi. Rub-impact Faults Detection Based on Varied Parameter Stochastic Resonance. Power System Engineering, 2009, pp. 21-26. (7) Hu G., L.Pivka, A.L.Zheleznyak. Synchronization of a One-dimensional Array of Chua's Circuits by Feedback Control and Noise. IEEE Trans. Cir&Sys, vo1.42, 1995, pp. 636-640. (8) CHEN Min, HU Niao-qing, QIN Guo-jun, and ZHANG Yun-an. Best Match Stochastic Resonance in Weak Signal Detection. Journal of Vibration Measurement&Diagnosi s. 2007, pp. 270-272. (9) HE Da-hai, ZHAO Wen-Ii, and MEl Xiao-jun. Mechanical&E1ectrical Engneer Ngmagazne. 2008, pp. 71-74. (10) EBERHART R C, KENNED Y 1. A New Optimization . Proceedings of 6th International Symposium on Micro Machine and Human Science. 1995, pp. 39-43.

1(')

(c) The time-domain spectrum of output signal

IlO.100
III

---,

lO W
0

\AA.
OW 0 03 0(1.1 DOS OlE
'.'Hz)

001

007

0 00

0 00

01

(d) The frequency-domain spectrum of output signal Figure 6. The time and frequency domain spectrum of input noisy signal and output signal

Fig. 6 shows that, in the original signals , the useful signal spectrum which submerged in the noise spectrum is difficult to

434