Quadrature Sinusoidal Signals Correction of Magnetic Encoders via Radial Basis Function Neural Network and Adaptive Loop Shaping

Magnetic encoders are widely used in industrial motion control, due to their low-cost, simple structures and low environmental requirements. However, the obtained quadrature sinusoidal signals suffer from various disturbances, which affects the accuracy of the magnetic encoders. The current methods which combine the neural network with phase-locked loop (PLL) typically require the knowledge of harmonic orders in advance and use a proportional-integral controller as loop filter of the PLL. In this paper, we propose a new method, in which a radial basis function neural network (RBFNN)-based PLL is combined with adaptive loop shaping. In this method, with the incorporation of RBFNN into PLL, the disturbances could be readily eliminated, thus avoiding additional parameter identification. Furthermore, the adaptive loop shaping served to redesign the PLL's loop filter, aiming to strengthen the high-frequency noise attenuation capability. The method has been validated both theoretically and experimentally, confirming that it is an effective method to improve the accuracy of the magnetic encoders.