A RBFNN approach for DoA estimation of ultra wideband antenna array

The problem of direction of arrival (DoA) estimation of ultra wideband (UWB) electromagnetic (EM) waves is addressed and a radial-basis-function neural network (RBFNN) base approach is presented in this paper, in which the DoA estimation is achieved as a mapping which can be modeled using RBFNN trained with input–output pairs. Because RBFNN is characteristic of accurate approximation and good generalization, as well as robustness against interference and scattering from antennas, the proposed method can learn source direction findings of UWB array and work well in existence of manufacturing errors and mutual coupling of UWB array antennas. In order to get rapid training and avoid large network size, a hybrid leaning scheme for RBFNN is adopted. Firstly an unsupervised K-means clustering algorithm is employed to determine the centers of hidden neurons, then a recursive least square (RLS) algorithm is used to obtain the linear weights of the output layer. Moreover, in order to solve the multiple sources tracking of UWB array, we combine eigenvalue decomposition (EVD) with RBFNN to extract the DoAs of multiple sources. The effectiveness of our scheme is demonstrated through several numerical examples. The results show that this method is characteristic of high accuracy and robustness to mutual coupling, when compared to presently available methods in literatures.