An efficient neural network algorithm for reflector surface error compensation

A neural network algorithm for electromagnetic compensation of reflector surface error effects is formulated. Sets of trained neural networks are used to compute the compensation excitations for array feeds. The networks were trained using data generated with the constrained least squares (CLS) compensation method. Once trained, the calculation of the excitations is accomplished in significantly less time than required by the original constrained least squares algorithm. The surface error profile for a distorted reflector antenna is expanded using bivariate surface basis functions. Each of the trained networks corresponds to one of the expansion functions. Excitations computed using the neural networks are superposed to produce composite compensation excitations for the distorted reflector. The compensation results for a distorted reflector are presented, and the neural network algorithm performance is compared to the original CLS technique.