Research On The Estimation Of The Number Of Sources Based On Stacking Ensemble Learning

The direction-of-arrival(DOA)positioning technology has been extensively used in the fields of wireless communication,electronics,etc.Right now,most of the classic DOA need to be carried out when the number of sources is firm.The estimation effect of the number of sources directly affects the performance of DOA,so accurate estimation of the number of sources is the premise of DOA estimation.However,most of the current source number estimation algorithms can only be implemented under the condition of a large number of sampling snapshots,and the performance of the algorithm is severely degraded or even invalid under the condition of a small number of snapshots.However,in the real environment,it is generally difficult to ensure that the number of sampling snapshots of the signal is large enough.To realize the estimation of the number of sources in the actual environment,the algorithm must be able to have good estimation performance in an environment with low color noise and low signal-to-noise ratio.In order to improve the estimation of the number of sources in these scenarios,this paper proposes a method for estimating the number of sources based on Stacking ensemble learning,which uses Hilbert-huang Transform(HHT)and the Weighted Geschgorin Disk Estimator Criterion(WGDE)to extract features of signals.The Stacking integrated learning model integrates XGBoost,Light GBM and BP neural network as three parallel primary models,and the Logistic Regression model as the secondary model,and finally gets the source number estimation model XLN-L-Stacking based on Stacking integrated learning.The major research contents of the thesis are as follows:1.To introduce the research background,research purpose and main research significance of this article.And the current research situation of the source number estimation is summarized.2.Briefly introduce the relevant knowledge of the array signal,including the array model and the array signal model.This paper introduces the principle of the feature extraction method Hilbert Huang transform and the data feature enhancement method Weighted Geschgorin Disk Estimator Criterion.Then it focuses on the Stacking ensemble learning algorithm used in this article and the structure and theoretical knowledge of the other two ensemble learning methods.It also introduces the sub-model of XLN-L-Stacking.3.This article focuses on the modeling process,using HHT to extract the phase difference information generated by the different positions of the array elements on the array circular array,and use it as the basic feature.In addition,the augmented Geschgorin circle radius,augmented Geschgorin circle center value,and augmented weighted Geschgorin circle radius value of the signal are extracted by the WGDE as high-dimensional features.Then bring the eigenvalues into training based on the Stacking ensemble learning model,and finally use the trained model to estimate the number of array signal sources.4.It is verified through experiments that the algorithm in this paper is superior to the primary models and other integrated models.Then use theoretical data and RF noise reduction laboratory data to verify that this article has better estimation performance than the other four source number estimation algorithms,and it has better estimation performance for different snapshots,signal-to-noise ratio,and different incident angles.Robustness.Finally,it is verified through experiments that the combination of the features extracted by HHT and the features extracted by the WGDE can better extract the feature information in the signal,thereby improving the estimation accuracy of the algorithm.