Over-the-air Behavioral Modeling of Millimeter Wave Beamforming Transmitters with Concurrent Dynamic Configurations Utilizing Heterogenous Neural Network

Abstract

In this paper, a novel heterogenous neural network for over-the-air (OTA) behavioral modeling of millimeter-wave (mmW) beamforming transmitters with concurrent dynamic configurations is proposed. Different from conventional behavioral modeling methodology, the signal for model construction is obtained from the OTA measurement. By integrating concurrent dynamic configurations with transmitter input/output, the behaviors of enormous transmitter states can be accurately characterized by only one model, which is validated by experiments on a dual-channel mmW beamforming transmitter with 125 state combinations of concurrent dynamic configurations of input power, operating frequency and beam direction. Cross validation results indicate that proposed model can achieve great performance on both training and validation sets, which is very promising to be employed in 5G and future intelligent communication systems.