Inertially Controlled Two-Dimensional Phased Arrays by Exploiting Artificial Neural Networks and Ultra-Low-Power AI-Based Microcontrollers

The use of Artificial Intelligence (AI) in electronics and electromagnetics is opening many attractive research opportunities related to the smart control of phased arrays. This is particularly challenging especially in some high-mobility contexts, such as drones, 5G, automotive, where the response time is crucial. In this paper a novel method combining AI with mathematical models and firmware for orientation estimation is proposed. The goal is to control two-dimensional phased arrays using an Inertial Measurement Unit (IMU) by exploiting a feed-forward neural network. The neural network takes the IMU-based beam direction as input and returns the related phase shift matrix. To make the method computationally efficient, the network structure is carefully chosen. Specific and discretized cross-section regions of the array factor (AF) main lobe are considered to compute the phase shift matrices, used in turn to train the neural network. This approach achieves a balance between the number of phase-shifting processes and spatial resolution. Without loss of generality, the proposed method has been tested and verified on
$4\times 4$
and
$6\times 6$
arrays of 2.4 GHz antennas. The obtained results demonstrate that reconfigurability time, easiness of use, and scalability are suitable for a wide range of high-mobility applications.