CPNN Algorithm for Adaptive Beamforming

Abstract

This paper presents an effective adaptive beamforming method based on a complex-valued processing neural network (CPNN), named as CPNN algorithm. In the proposed method, the optimal beamforming problem can be formulated as a regression problem of neural networks (NNs). In the CPNN structure, a new real-imaginary merging (RIM) layer and a new imaginary-real merging (IRM) layer are constructed to process complex-valued data with other layers. Via the RIM and IRM layers, the complex-valued data computation in the neurons follows the complex-valued multiplication rule, which makes the mathematical relationship between the input and output of the NN-based beamformer more reasonable. Compared with the previous works in NNs, the proposed CPNN approach provides better beamforming performance, for example, 1) the phase information of the complex-valued data is maintained, which makes the output of the NN-based beamformer more accurate; 2) it does not require prior information of the desired signal, such as the desired direction of arrival, which will avoid errors caused by signal parameters estimation; and 3) it can not only effectively suppress the interference signals but also ensure that the response of the desired signal is distortionless. Simulation results demonstrate the efficiency of the presented approach.