Enhancing the Initial Position Estimation of a Resolver in a Servo- Motor-Controlled Satellite Ground Station With Regression Techniques

Abstract

Resolvers are very accurate rotating position feedback mechanisms utilized by Low Earth Orbit (LEO) ground stations to track a satellite and communicate with it. Mechanical misalignments and temperature fluctuations are some of the environmental reasons that commonly cause resolvers to suffer from problems in calibration. The contribution of this work is to present a methodology for the calibration of 16-bit resolvers using 14 different Machine Learning (ML) techniques that improve the accuracy and reliability of LEO ground stations. Conventional calibration techniques include mechanical adjustment of the resolver for known inaccuracies. This often involves much manual refinement and recalibration to achieve any reasonable degree of accuracy. On the other hand, the proposed automatic calibration would reduce the need to routinely perform human calibration, thereby reducing wastage of time and other resources. Machine Learning statistical algorithms can learn from data and generalize to new data, including complex input-output mappings, and have made such error profiles and resolution features visible. This software-based error-compensation technique improved the target distortion ratio of a 16-bit resolver from approximately ±10% to approximately ±2%. In cases where ML is used for calibration, it is possible to reduce the goal angle error—which can reach up to 1°—to a level of 0.2°.

Cite this article as: Y. Sancar and R. Görkem Birdal, "Enhancing the initial position estimation of a resolver in a servo-motor-controlled satellite ground station with regression techniques," Electrica, 25, 0167, 2025. doi:10.5152/electrica.2025.24167.