A Study On The Estimation Of Far-field Narrow-band Sources Under Small And Fast Beats

In array signal processing,direction-of-arrial(DOA)of signal source,as a popular research direction,has wide applications in sonar,radar,communications,medical detection,electronic countermeasures and other fields.Traditional DOA estimation algorithm is based on the number of signal sources has been determined,if the estimated number of signal sources and the actual number of signal sources are not consistent,it will lead to an error in DOA estimation.Therefore,in order to achieve the correct DOA estimation,it is necessary to realize the accurate estimation of the number of signal sources.Most of the existing signal source number estimation algorithms are implemented when the number of sampling quick beats is large,but in the real environment,it is often unable to satisfy the large number of sampling quick beats,if the number of sampling quick beats is small,it will affect the performance of the algorithm,or even fail.Based on this,in order to make under the small snapshot number is good for source number estimation,this paper puts forward a kind of based on BP(Back Propagation)neural network source number estimation method,the method of the number of different source array signal feature extraction,using artificial neural network as a classifier to do the characteristics of the training,the trained neural network model is used to estimate the number of signal source.The main contents of the paper are summarized as follows:1.The research background and significance of this topic are introduced in detail,and the current research status at home and abroad is reviewed and summarized.2.Two commonly used array models,uniform circular array and uniform linear array,are studied,and the influence of error estimation of source number on DOA estimation is analyzed,and the main factors affecting the estimation of signal source number are analyzed.3.The classical signal source number estimation algorithms in white noise environment and color noise environment are introduced in detail respectively.Through Matlab simulation experiment,the performance of each algorithm under different fast beat number,different SNR and different signal source number is compared and analyzed.4.The BP neural network and hilbert-huang transform(HHT),two theoretical pillars of the source number estimation algorithm in this paper,are introduced.Firstly,the instantaneous phase can be obtained from the original array signal through HHT,and then the training set of BP neural network can be obtained after feature extraction of the instantaneous phase.At the same time,in order to achieve higher accuracy of BP neural network,a double hidden layer neural network structure is put forward,in which the number of nodes in the double hidden layer can be determined by trial and error method,and some improved methods are proposed for the inherent defects of BP neural network.5.This paper mainly introduces the realization of the algorithm of source number estimation,as well as the experimental simulation and analysis of the method.In order to solve the problem that the detection probability of most source number estimation algorithms decreases or even fails when the sampling speed is small,this paper proposes a source number estimation algorithm based on BP neural network.The instantaneous phase of the original array signal can be used as the theoretical basis for the source number estimation because of the phase difference of each array element in UCA.Firstly,the instantaneous phase of the original array signal can be obtained by HHT,and then the data set needed for BP neural network training can be obtained by feature extraction of the instantaneous phase.Finally,the trained BP neural network can be used to estimate the array signal with different number of signal sources.In order to validate this paper estimates the performance of the algorithm,this paper adopted the theory of data and the measured frequency sound lab data,through the algorithm in this paper and the other four algorithms in different number of fast,do simulation under different signal-to-noise ratio factors such as the contrast experiment,so as to verify the effectiveness of the algorithm in this paper a good estimation performance and practicability.