Implementation and Evaluation of LS-SVM Optimization Methods for Estimating DoAs

Important technological advancement in designing smart array antennas has been encouraged many researchers to concentrate their work on the two main concepts of the direction of arrival (DoA) and beamforming techniques. The preliminary objective of beamforming techniques includes, electronically, the mainbeam in the direction of interest at a certain time and measuring the output power. In this scenario, the main practical challenge resides in achieving maximum output power in which the direction of steered mainbeam coincides with the direction of arrivals. Since the involved problems in most DoA estimation optimizations consist of a lot of unknown parameters including direction of arrivals, SNRs, signal waveforms and samples of noises in the array output, it may become impossible to build a large enough training dataset for covering the distributions for all the aforementioned test data. An alternative way to overcome this constraint which we aim at stressing in this work involves employing support vector machine algorithms for separating unknown components of the actual input in the higher dimensional feature space. In this work, we have implemented the decision directed acyclic graph (DDAG) and Vapnik-Chervonenkis (VC) methods for the least squares support vector machine (LS-SVM) algorithms for estimating DoAs. We have rigorously verified that DoAs are very much affected the antenna array geometries. In addition, we have investigated the quality of the communication channel by the concept of bit error rate (BER).