**Overview**: Fiber optic gyro (FOG) is an inertial sensor based on the Sagnac effect. It has the advantages of high reliability, high measurement accuracy, and ease of integration. It has become an ideal device for inertial navigation systems. The collected FOG drift data is affected by many factors such as the light source, fiber bending, and ambient temperature, making it often submerged in the noise and leading to difficulties in direct modeling compensation. In order to establish an accurate error compensation model, data preprocessing is demanded to output data on the gyro.

In this paper, a hybrid EMD-LWT filtering algorithm based on empirical mode decomposition (EMD) and lifting wavelet transform (LWT) threshold denoising is proposed to preprocess gyro signals. Firstly, the steps of empirical mode decomposition are introduced. After the signal is decomposed by EMD, a finite number of high-to-low frequency intrinsic mode functions (IMFs) are obtained. The low order part represents the high frequency part of the signal, which usually contains a sharp part or noise; An IMF with a large order corresponds to the low-frequency part of the signal, and it is generally considered that the noise in the low-frequency component has little effect. It is decomposed into noise-dominated IMF sets, where noise and effective information coexist and a signal low-frequency trend. The threshold filtering method based on EMD is a process to select and threshold three types of IMF sets. The information entropy and the energy of the signal serve as a measurement of the complexity of the signal and determine the boundaries of the noise component and the mixed modal component.

Considering that the traditional EMD time-scale filtering algorithm simply removes one or more IMF components to achieve filtering, resulting in the useful signals along the corresponding components being deleted together. It will lead to severe signal distortion. The lifting wavelet analysis is introduced into the EMD method, and the high-frequency IMF component is subjected to the narrowband re-decomposition of the lifting wavelet to improve the resolution of the high-frequency component; considering the noise decomposition after being distributed on each IMF component, combined with the characteristics of wavelet threshold denoising. All IMF components are subjected to wavelet threshold denoising.

A static FOG data was collected as a test signal for verifying the effectiveness of the algorithm. The hybrid EMD-LWT was compared with the wavelet transform (WT) and the lifting wavelet transform (LWT) threshold filtering methods. The simulation results show that the root mean squared error (RMSE) of the signal is reduced by 63% through the EMD-LWT filtering algorithm and the denoising effect is obvious.