Internal Calibration System Using Learning Algorithm With Gradient Descent

Abstract

We present a novel approach to internal calibration of a radar system. A Ku-band radar system with internal calibration paths is designed. Thermal drift of a system is mainly caused by active components, which are a high-power amplifier (HPA) and a low-noise amplifier (LNA). We aimed to reduce the drift using a learning algorithm with a gradient-descent method. Hardware offset factors and calibration factors are introduced for the process. In the learning algorithm, a penalty term is formed based on the analysis of local minimum points. The result verifies the proposed internal calibration method. Maximum deviations of gain are 0.0477 dB for the HPA and 0.0132 dB for the LNA. In addition, the maximum deviations of phase are 0.2481 degrees for HPA and 0.0722 degrees for LNA, respectively.