时间频率学报

大型光纤陀螺仪可以实时输出包含地球自转信息的数据，其中有用低频信号被大量噪声淹没，为解算得到高时间分辨率的世界时（UT1），需要分析噪声特性并研究噪声处理方法。通过对原始数据进行功率谱密度分析，结合高频噪声特点，针对性地设置截止频率，并利用巴特沃斯滤波器对数据进行初步处理，再结合经验模态分解(EMD)和累积标准化均值方法筛选信号中的混合项和信息项。利用奇异谱分析对混合项进行分解，依据Hurst指数筛选出其中的有效次分量，将其进行重构以完成对混合项的去噪，将去噪后的混合项、信息项和趋势项重构得到去噪信号。针对去噪信号存在的漂移问题，进一步分析了光纤陀螺仪数据和温度的关系，利用极限学习机模型进行了温度误差补偿。实验结果表明：所提出的方法和经典方法相比去噪效果更好，且去噪后数据的标准差更小，更具优越性。

Large optical fiber gyroscopes can output data in real time that includes information about the Earth's rotation. The useful low-frequency signals in the data are overwhelmed by a large amount of noise. To calculate the universal time(UT1) with high temporal resolution, it is necessary to analyze the noise characteristics and study noise processing methods. By conducting power spectral density analysis on the original data and considering the characteristics of high-frequency noise, the cut-off frequency was set specifically and the data was initially processed using a Butterworth filter. Then, the empirical mode decomposition(EMD) and cumulative standardized mean method were combined to screen the mixed terms and information terms in the signal. The singular spectrum analysis(SSA) was used to decompose the mixed terms, and the effective sub-components were selected based on the Hurst index, which were then reconstructed to complete the denoising of the mixed terms. The denoised mixed terms, information terms, and trend terms were reconstructed to obtain the denoised signal. To address the drift problem in the denoised signal, the relationship between the fiber optic gyroscope data and temperature was further analyzed, and the extreme learning machine(ELM) model was used for temperature error compensation. The experimental results show that the proposed method has a better denoising effect compared with the classical methods, and the standard deviation of the denoised data is smaller. The proposed denoising method exhibits superiority.