A Statistical Signal Processing and Machine Learning Approach of Dithered Pseudo Random Noise for Evaluating the Performance of Ring Laser Gyro (RLG)

This paper presents a new approach to enhance Ring Laser Gyroscopes (RLGs) performance using Statistical Signal Processing (SSP) and Reinforcement Learning (RL). RLGs are devices that are highly significant for most navigation systems; they are, however, inherently subject to the lock-in effect, static and dynamic, which degrades the accuracy and performance. To minimize these issues, the concept of Dynamic Dither Angle (DDA) and Dither Symmetry Angle (DSA) calculations, which are derived from dithered Pseudo Random Noise (PRN), is injected into the RLG system. The DDA is calculated recursively, providing a method to track the cumulative effect of dither pulses. In contrast, the DSA method takes the form of a balance measure of the errors in the system caused by dither. Deep Deterministic Policy Gradient (DDPG) is used as an optimization method for PRN values to avoid lock-in effects for better measurement accuracy. The RL agent works in interaction with the RLG system and tunes the PRN values such that multiple performance metrics can be fed back, comprising lock-in occurrences and angular measurement accuracies. Experimental results proved that both static and dynamic lock-in effects can be reduced enormously, whereby overall gyroscope performance is enhanced. This SSP-RL integration provides a robust and efficient way to advance RLG technology into more reliable solutions for accurate navigation.