Approximate Kernel Orthogonalization for Antenna Array Processing

We present a method for kernel antenna array processing using Gaussian kernels as basis functions. The method first identifies the data clusters by using a modified sparse greedy matrix approximation. Then, the algorithm performs model reduction in order to try to reduce the final size of the beamformer. The method is tested with simulations that include two arrays made of two and seven printed half wavelength thick dipoles, in scenarios with 4 and 5 users coming from different angles of arrival. The antenna parameters are simulated for all DOAs, and include the dipole radiation pattern and the mutual coupling effects of the array. The method is compared with other state-of-the-art nonlinear processing methods, to show that the presented algorithm has near optimal capabilities together with a low computational burden.