DNN‐based interference mitigation beamformer

Abstract

In classical beamforming techniques such as minimum mean squared error and minimum variance distortion-less response, for computing optimum weight vector of an antenna array, direction of arrival (DoA) of desired signal should be known as prior knowledge. In this study, a deep neural-network-based beamformer is proposed for estimating the desired signal in presence of noise and interference without requiring prior knowledge on the DoA of the desired signal. In the proposed beamformer, a bidirectional long short-term memory (bi-LSTM) is employed for estimating samples of interferences. On the other hand, samples of the desired signal are estimated by either another bi-LSTM or a convolutional neural network. The signal to interference and noise ratio (SINR) at the output of the proposed beamformer is 10 dB higher than the SINR at the output of the classical beamformers when the number of available snapshots is as low as 100. The proposed beamformer has promising performance when the input interference to signal ratio is as high as 34 dB and the input signal-to-noise ratio is as low as −10 dB.