Research On Weak Signal Detection And DOA Estimation Based On Nonlinear Method

Weak signal detection aims to detect and identify the relevant parameters of the target weak signal under strong background noise.With the rapid development of intelligent technology,the weak signal in space grows explosively.As the cornerstone of the smooth development of intelligent technology in communication,the accurate acquisition,transmission,reception,and classification of weak signals has important research value and significance.In the complex and changeable background environment in the communication network,higher requirements are put forward to transmit and detect weak signals.It requires the high detection accuracy and efficiency of weak target signals,also estimating the location of transmitting devices of those.Compared with linear detection methods,nonlinear detection methods have incomparable advantages in weak signal detection.Chaos theory can effectively detect weak signals under a low signal-to-noise ratio(SNR).Through training,neural network technology can classify and learn the characteristic parameters of weak signals.Therefore,it realizes the construction of an intelligent network and provides an implementable solution for parameter detection of targeting weak signals in communication.However,in the communication environment,various weak target signals have their characteristic parameters,which brings many challenges to high-precision weak signal detection.There are various characteristics of weak signals widely existing in the communication network.This dissertation mainly studies two characteristic parameters of the target weak signal—the frequency detection of the weak signal and the estimation of the direction of arrival(DOA)of the weak signal.For the frequency detection of weak signal,based on the immunity to noise and sensitivity to initial conditions of the chaos theory,the frequency of the weak target signal can be effectively detected under the background of strong noise by discriminating the output state of the chaotic system.Thus,the detection performance is boosted.The challenge is how to expand the application of chaos theory from the frequency detection of a single-period signal to the other communication scenario,such as the frequency detection of multiple non-variable-frequency signals,single variable-frequency signal,and multiple variable-frequency signals.In addition,for the angle estimation of weak signals,in practical applications,due to factors such as the imperfection of receiving array,the unknown number of signals,and low SNR,the robust design of the DOA estimation is required.Based on the above concerns,this dissertation mainly focuses on the frequency detection of weak signals with different characteristics and the angle estimation in imperfection array with unknown signal numbers under low SNR.The research contents are as follows:Firstly,given the problem of non-variable-frequency weak signal transmission,the traditional chaos theory can effectively detect a single non-variable-frequency weak signal under Additive White Gaussian Noise(AWGN).Still,in practical applications,due to the bandwidth limitation of the receiver,the background noise is no longer AWGN.In addition,when the number of weak signals to be detected increases and their frequencies are very dense,the traditional chaotic detection system is no longer applicable.Therefore,it is modeled as frequency detection of target signals in multiple non-variable-frequency signals with adjacent frequency(MNVFS-A)under narrow-band noise.This dissertation first studied the influence of narrow-band noise on the critical state,then established a criterion for discriminating the output state of the system based on the Melnikov method,and proved that the increase of the oscillator’s amplitude could reduce the influence of noise effectively.Since the target signals are MNVFS-A,to reduce the influence of the detection blind zone,a differential oscillator array is proposed to discriminate the output state of the system in this dissertation.Moreover,the concept of generalized output state is proposed to re-divided the system output state.At the same time,a detection algorithm based on the period duration ratio(PDR)is proposed,which discriminates the system’s output state from a new perspective and can obtain the frequency values of multiple weak signals with adjacent frequencies,respectively.Finally,the algorithm’s performance is verified in system complexity,detection speed,and detection blind zone.Secondly,the problem of variable-frequency weak signal transmission,which widely exists in the communication network,is modeled as detecting the frequency modulation rate of target signals under low SNR.The research in this dissertation is divided into two parts: the frequency detection of single variable-frequency signal(SVFS)and multiple variable-frequency signals(MVFS).Considering the frequency changes with time,we establish a chaotic oscillator array by taking the linear frequency modulation signal as the research object.Then the frequency range of the chirp signal can be determined quickly by the characteristic of the critical state of the system output state.Meanwhile,to analyze the output state of the system qualitatively,this dissertation establishes a chaos-based system consisting of several oscillators to determine the output state by the Lyapunov exponent simultaneously.In addition,a detection algorithm based on the Lyapunov exponent is proposed,which can obtain the output period state of each oscillator and the corresponding time point.Then we can get the relevant parameters(frequency modulation rate,frequency range)of a single variable-frequency signal.Furthermore,for the frequency detection problem of multiple variable-frequency signals with overlapping frequencies(MVFS-O),this dissertation introduces the concept of overlapping frequency ratio to classify and analyze the frequency overlap of various signs.Finally,the simulation results show that the detection scheme proposed in this paper can effectively detect the relevant parameters of the variable frequency weak signal.Finally,given the problem of the location estimation of weak signals in the communication network,it is modeled as the DOA estimation of the weak target signals.Given the high loss and high dynamics of the communication environment,the perfect receiving array cannot be guaranteed.Therefore,this dissertation proposes a scheme for highprecision estimation of DOA under the conditions of imperfection array,the unknown number of signals,and low SNR.The scheme includes two parts: denoising autoencoder and deep neural network.Denoising autoencoder can reconstruct the original data from the corrupted data.Then,the influence of noise on the weak signal is reduced effectively,and the robustness of the whole system is improved.Deep neural networks have a powerful feature representation ability.It can map between the DOA of the weak target signal and the divided angle grid.The weight coefficient of DAE-DNN can be adjusted through training data.The strong learning ability and generalization can be used to overcome the impact of the imperfection of the array and the unknown number of signals.Simulation results show that the estimation performance of the proposed scheme can be significantly improved under the condition of imperfection array,the unknown number of signals in low SNR while guaranteeing the accuracy of DOA estimation.The research provides a reference to overcome the difficulties existing in the application process of weak signal detection and location in the communications network from the perspective of the specific implementation.