Research on the Estimation of the Number of Sources Based on Multiple Features Fusion

张冰玉

王峰

硕士

广东工业大学

信源数估计;;均匀圆阵;;数学模型;;加权盖尔圆准则;;支持向量机

随着阵列信号处理技术的发展,它在野外通信、医学检测、地震勘探和图像处理等信息领域中应用的越来越广泛,其中对信号的波达方向(Direction-Of-Arrial,DOA)进行估计又是这些领域研究的热点。目前,虽然已经有了很多超分辨高精度算法被用于DOA估计中,但是这些超分辨算法性能得到保证的前提是已知正确的信源数,当信源数目错误时,这些算法的估计性能将会受到影响,因此对信号中包含的确切信源数目进行估计是十分有必要的、对阵列信号入射的信源数进行估计已成为阵列信号处理中最基本的研究任务之一。目前,虽然有很多学者提出了各自的信源数目估计算法,但仔细研究后可以发现这些算法为了能估计出尽量多的信源数,都是选用较多的阵列天线数去搭建信号模型,而在实际应用中布置越多的天线数不仅会造成较大的干扰信号,而且也难以搭建阵列模型,所以本文提出了一种基于多重特征融合的信源数估计方法,该方法能在尽量少的天线数下去估计尽量多的信源数。在了解了相关的信源数目估计算法后,本文以加权盖尔圆准则(Weighted Gerschgorin Disk Estimation,WGDE)为理论基础再结合支持向量机(Support Vector Machine,SVM)来设计算法。本文主要探讨了以下几个方面:1、首先了解阵列信号处理的背景知识,以及信源数估计算法的研究现状。2、介绍本文所使用的阵列信号模型,再对几种典型的信源估计算法,包括:假设检验法、基于信息论准则的信源估计算法、经对角加载处理后信源估计算法以及盖尔圆(Gerschgorin Disk Estimation,GDE)算法进行了分析,分别在白噪声和色噪声下对经典的信源估计算法进行了理论数据实验仿真。除此之外,还介绍了一类基于模式分类的信源估计算法。3、因为本文算法是以支持向量机、加权盖尔圆准则这两种算法为理论基础,所以对这两种算法分别做了分析,再介绍了一种基于支持向量机的用于训练多分类器数学模型的软件包Lib SVM。4、重点介绍了本文算法拟使用的多重特征、多分类器数学模型的训练过程以及本文提出的信源估计算法的整个设计流程:首先利用天线数为M的均匀圆阵(Uniform Circular Array,UCA)(以下简称:M-UCA)阵列去接受相互独立入射的远场窄带高斯信号,再对阵列的接收信号作加权盖尔圆准则变换,从中同时获取能用于描述信源个数的增广盖尔圆心值、增广盖尔圆半径值和增广加权盖尔圆半径值,以这三种值作为估计信源的特征参数,将他们融合成高维特征向量,标记后代入Lib SVM中训练多分类器数学模型。最后通过理论数据仿真和射频消音实验室数据仿真验证按照本文方法训练得到的分类器数学模型对信源数的估计性能。由实验结果得到:当阵列信号的入射角、信噪比、快拍数改变时,该数学模型能较好的估计出信源数,而且当信源数比阵列天线数少一个时也能有效地估计地估计出信源数。

With the development of array signal processing technology,it is more and more widely used in information fields such as field communication,medical detection,seismic exploration,and image processing.In these fields the direction of the signal(DOA)estimation is also become a hotspot.Although many super-resolution high-precision algorithms have been used in DOA estimation,the premise of ensuring the performance of these super-resolution algorithms is to know the correct number of sources.When the number of sources is not konw,the estimation performance of these algorithms will be affected.Therefore,it is very necessary to estimate the exact number of sources.So estimating the number of sources in array signal processing has become one of the basic research tasks.At present,although many scholars have proposed their own algorithms for estimating the number of sources.But after careful study,we can find that in order to estimate as many sources as possible,these algorithms use more array antennas to build a signal model.However,in actual applications,the more antennas used the number of interfering signals is larger and the more difficult to build a signal model.Therefore,this article proposes a method for estimating the number of sources based on multi-feature fusion,which can estimate as many sources as possible with as few antennas as possible.After understanding the relevant source number estimation algorithm.This article takes Weighted Gerschgorin Disk Estimation(WGDE)as the theoretical basis and combines Support Vector Machine(SVM)to design the algorithm and mainly discusses the following aspects:1.The first is to understand the background knowledge of array signal processing and the research status of source number estimation algorithms.2.Introduce the array signal model used in this article and several typical source estimation algorithms.Including: hypothesis testing method,source estimation algorithms based on information theory criterion,source estimation algorithm after diagonal loading and Gerschgorin Disk Estimation algorithm,and also introduced a class of source estimation algorithm based on pattern classification.The theoreticaldata experimental simulation of the classic source estimation algorithm is carried out under white noise and colored noise respectively.3.Because this article takes Weighted Gerschgorin Disk Estimation and support vector machine asthe theoretical basis for modeling,so the two algorithms are specifically introduced.Then a software package for training multi-classifier mathematical models based on support vector machine is introduced.4.Introduced the multi-features used in this article,the training process of the multi-classifier mathematical model and the entire design process of the source estimation algorithm proposed in this article.First,the Uniform Circular Array(UCA)is used to receive independent and incident far-field Gaussian signal source and then use the WGDE criterion to transform the receive signal.Obtain the augmented Gerschgorin center value,augmented Gerschgorin radius value and augmented weighted Gerschgorin radius value that can be used to describe the number of sources at the same time.Use these three values as the characteristic parameters of the estimated source,and the features are merged into high-dimensional feature vectors,labeled and substituted these vectors into Lib SVM to train the mathematical model of multi-classifier.Finally,theoretical data simulation and radio frequency silencing laboratory data simulation verify the source estimate performance of the classifier mathematical model.The experimental results show: when the incident angle,signal-to-noise ratio,and number of snapshots of the array signal change,the mathematical model still has a good source estimation performance,and when the number of sources is one less than the number of array antennas,it can also effectively estimate the number of sources.