SVM-Based DOA Estimation with Classification Optimization

Direction of arrival (DOA) estimation is an important issue in many applications such as radars, wireless communications, objects detection and imaging. The knowledge of DOAs allows the use of control algorithms to enhance systems performances namely capacity growing in wireless communication. Improving DOA estimation methods and reducing computational time are thus of most interest. We present an optimization of a Support Vector Machine (SVM) based approach for DOA estimation. Correlation matrix analysis leads to predictors selection, training dataset reduction, and improvement of generalization capability up to 80% of the SVM system. The approach is successfully tested on a two-dimensional Direction of Arrivals (DOA) estimation.