采用混合核支持向量机的DOA估计

摘要

A large number of samples and complex structure are needed when modeling with artificial neu-ral network(ANN). And each ordinary kernel function of support vector machine(SVM) has advantages and disadvantages. Moreover,the theory for selecting the parameters of SVM model is still incomplete. In order to solve these problems,SVM based on hybrid kernel function is proposed to improve the performance of the direction of arrival( DOA) estimation,in which the parameters of SVM and weight coefficient of the hybrid kernel are optimized by binary particle swarm optimization ( PSO ) algorithm. The hybrid kernel function obtains high generalization ability and learning ability. Firstly,the polynomial functions are used to test the SVM based on hybrid kernel function and the method is verified to be effective. Finally,the method is used for modeling DOA estimation,and the model is validated by comparing its results with those of RBF neural network( NN) and SVM based on a single kernel function and MUSIC algorithm. Experiments show that the method improves the DOA estimation accuracy and can achieve stable results.%人工神经网络( ANN)进行建模时通常需要准备大量的数据样本，同时网络结构一般都比较复杂；而采用支持向量机( SVM)进行建模时，不同核函数有不同的效果，各有利弊，且选取SVM模型参数的理论支撑尚不完整。为了解决这些问题，提出了一种基于混合核函数的支持向量机来改善来波到达角( DOA)的估计性能，并结合二进制粒子群算法( PSO)来对混合核函数进行参数寻优。该混合核函数由全局核函数和局部核函数构成，提高了SVM的泛化能力和学习能力。首先通过拟合多项式函数，验证了该混合核SVM的有效性。将该方法用于DOA估计建模，在不同信噪比和快拍数下，通过与径向基函数( RBF)神经网络、基于各单一核函数的SVM和MUSIC算法预测结果对比，混合核SVM均方差有所降低，提高了DOA估计的精度且有更好的稳定性。