Multiscale Parallel Heading Error Processing Model for Polarization Compass

A multiscale parallel processing model of heading error for the polarization compass (PC) is developed in this article. Existing state-of-the-art schemes of raising heading measurement precision are implemented relying on not only ideal assumption in noiseless heading output from the PC but also the polynomial method employed to calculate the heading error caused by the varied attitude angles of PC, which leads to fall sharply in the PC heading accuracy in real applications. Moreover, the aforementioned denoising and heading error compensation are conducted independently. Here we specifically concentrate on a novel and promising parallel processing model for heading error as a way to improve the PC heading accuracy, where a comprehensive analysis of heading error sources is first carried out for the PC. Subsequently, the proposed multiscale parallel error processing model is presented to improve heading accuracy for the PC by repressing noise as well as compensating heading errors that are implemented synchronously. Specifically, variational mode decomposition (VMD) is exploited to decompose the original heading data output from PC into multiple intrinsic mode functions, which characterizes binned modes (BIMFs) with high-to-low frequencies by seeking the ensemble of the decomposed modes and their respective center frequencies. An adaptive multiscale principle component analysis (MS-PCA) scheme denoising approach exploiting the principle contribution rate is performed to attenuate the noise in the high-frequency BIMFs. An effective compensation approach with a long short-term memory (LSTM) modeling is introduced straightforward to compensate for the heading error for the low-frequency BIMF concurrently. The rotational test is carried out to verify that the heading error processed by VMD-PCA denoising and LSTM compensating, which are the most effective among the comparison algorithms and the RMSE of 0.3505 and 0.2668, increased by 80.10% and 82.00%, respectively. The UAV flight test demonstrates that the RMSE of the parallel scheme using VMD-PCA and LSTM performed simultaneously is the minimum values of 0.1282 and 71% higher than the original navigation error. It can be seen from various experimental results that the proposed multiscale parallel processing model of denoising and error compensation procedure performed synchronously shows attractive performance gains with respect to emerging single processing in terms of increasing the accuracy of the PC heading measurement.