Research On Feature Extraction Technology Of Weak Signal Based On Stochastic Resonance

Overcoming the adverse effects of channel noise has always been an important research content in the field of communications.With the increasing complexity of wireless communication environments,how to enhance the weak signals and their characteristics submerged in the background noise has become a hot issue in current research.At present,most weak signal feature extraction techniques are mainly based on the idea of suppressing noise,but the target features will inevitably be suppressed and destroyed to a certain extent.Stochastic resonance can use a nonlinear system to transfer part of the energy of the noise into the signal.Therefore,this subject applies it to the field of non-cooperative communication signal processing,aiming to achieve more effective signal enhancement and feature extraction.The main work and research results are as the follows:1.Aiming at the issue that bistable stochastic resonance cannot effectively deal with many kinds of weak signals and the system parameters are difficult to select,an adaptive parametertuning stochastic resonance(PSR)method based on singular value decomposition(SVD)is proposed.Firstly,starting from the characteristic subspace of the signal,the evaluation function is constructed by using singular value decomposition,and the single-parameter optimization is performed by amplitude normalization.In addition,a moving average filter is added to the stochastic resonance processing module to solve the amplitude drift phenomenon.At last,the artificial fish swarm algorithm(AFSA)for solving the optimal system parameters can converge at a fast iteration speed and achieve the optimal match between the input and the nonlinear system.2.Aiming at the deterioration of the performance of traditional modulation identification technology under low signal-to-noise ratio(SNR),the parameter-tuning stochastic resonance is proposed to extract and enhance modulation identification features.Based on the four kinds of instantaneous characteristics of amplitude,phase,frequency and wavelet transform,the effects of stochastic resonance on a total of seven characteristic parameters are discussed.And it is verified that the phase delay effect of stochastic resonance system will not bring negative impact on feature extraction.Finally,BP neural network is used to classify the six improved features.Simulation results show that PSR can greatly reduce the SNR threshold for successful modulation identification and classification.3.Aiming at extracting the symbol rate of MPSK and MQAM signals at low SNR,a method by combining stochastic resonance and wavelet transform is proposed.Firstly,the adaptive PSR are used to match the optimal system parameters for noisy signals,and then Haar wavelet transform is used to further extract transient information.Finally,a modular method is used to design the overall architecture.This method not only makes up for the disadvantage of poor effect of using stochastic resonance alone and its easy divergence as a non-linear system,but also reduces the influence of the optimal scale of wavelet which is difficult to determine.Simulation results show that the method can improve the output peak to a certain extent and reduce the SNR threshold.4.Aiming at the feature extraction problem of the actual number of subcarriers in OFDM signals,it is proposed to improve the performance of traditional algorithms by using stochastic resonance.Firstly,the adaptive PSR is used to pre-enhance the OFDM signal.Secondly,the cepstrum method and the wavelet improved cepstrum method are used respectively.Finally,the number of subcarriers is determined by performing peak detection on the cepstrum of the signal.The simulation results show that the stochastic resonance can improve the above two methods to a large extent,improving the detection peak,reducing the SNR threshold,and increasing the estimation accuracy.