Calibrating resolver for LEO satellite ground stations utilizing Deep Neural Network

Abstract

Low Earth Orbit (LEO) ground stations rely on precise rotational position feedback systems, such as resolvers, to accurately track and communicate with satellites. Nevertheless, resolvers often suffer from calibration errors due to mechanical misalignments, temperature variations, and other environmental factors. This study proposes a method for calibrating 16-bit resolvers using Deep Neural Networks (DNNs) to enhance accuracy and reliability in LEO ground station applications. In traditional calibrations, known errors are compensated for by mechanically aligning the resolver. This process often requires manual intervention and periodic recalibration to maintain accuracy. Instead, proposed automated calibration reduces the need for frequent manual calibration, which saves time and resources. Error profiles and characteristics of resolver were easily revealed by DNN's multiple layers of neurons, capable of learning complex input-output mappings. As a result of this software-based error compensation method, target distortion ratio of 16-bit resolver was improved from ~ ±%10 to ~ ±%2. This means the error in the target angle, which can be as high as 1°, can be decreased to a level of 0.2° by using the DNN for calibration. It is also possible to overcome nonideal characteristics of a resolver such as amplitude imbalance, quadrature error, and inductive harmonics with the method presented in this study.