A Power-Only Fast Calibration Method for Phased Array Using Convolution Neural Network

A power-only fast calibration method for phased array is proposed in this letter. Empowered by the artificial intelligence (AI) techniques, the novel approach significantly reduces the number of required measurements without sacrificing calibration accuracy. To overcome the challenge of acquiring the training data with labels through measurements, we put forward a simple yet effective signal model for phased arrays, facilitating the generation of abundant high-quality training data via numerical calculation alone. A deep neural network model combining real- and complex-valued networks is developed for the implementation of the calibration method. The performance of the proposed method was first examined by numerical simulations and then verified by experiments using a self-developed Ka-band 64-element phased-array. All validation results consistently confirm the efficiency and calibration accuracy of the proposed method.

I. Introduction

Phased arrays find extensive applications in both communication and sensing systems [1], [2], [3], [4], enabling the generation of desired radiation patterns through precise manipulation of the signal amplitude and phase of each antenna element. Unfortunately, unwanted phase and amplitude deviations always occur due to inevitable discrepancies among elements, leading to severe distortion in radiation pattern. To achieve the objective performance, it is crucial to correct these amplitude and phase errors through array calibration. In addition, it is desired to reduce calibration time without degrading accuracy.