Application of the block recursive least squares algorithm to adaptive neural beamforming

Spatial beamforming using a known training sequence is a well-understood technique for canceling uncorrelated interferences from telecommunication signals. Most of online adaptive beamforming algorithms are based on linear algebra and linear signal models. Both in the transmitter amplifier and in the array receiver nonlinearities may arise, producing distorted waveforms and reducing the performance of the demodulation process. A nonlinear spatial beamformer with sensor arrays may use a neural network to cope with communication system nonlinearities. In this work we show that a feedforward neural network trained with a LS-based algorithm may converge in a time suitable to most applications.