Tuning of Deep Neural Networks for Over-The-Air Linearization of Highly Nonlinear Wide-Band Active Phased Arrays

Abstract

In this paper, we demonstrate how a deep neural network (DNN) can be used to compensate for nonlinearities and distortion effects introduced by the latest technology of 5G transmitters. A linearization approach based on neural networks can successfully cope with the challenges introduced not only by the high nonlinearity, wide bandwidth, and high frequency but also with challenges due to inter-PA crosstalk and load modulation. The device-under-test used in this experiment, is a state-of-the-art $5\mathrm{G} 4 \times 4$ active phased array (APA) operating in highly nonlinear regions at 28 GHz with a 100 MHz wide 3GPP base station signal and with OTA measured signals used for training. Using the proposed DNN based linearization technique, an improvement of 11 % in error vector magnitude (EVM) and 10 dB suppresion in adjacent channel leakage ratio (ACLR) are achieved which demonstrates the promising potential of this technique for emerging broadband communication systems such as 5G/6G and low earth orbit (LEO) satellite networks.