Multi-scale Principle Manifold Learning Noise Reduction Method for Telemetry Vibration Signal

Telemetry vibration signal contains a great deal of information that reflects the status of the aircraft during the test, and has characteristics of strong noise, non-linear, non-stationary, transient impact and so on. How to effectively denoise the signal without changing the dynamic characteristics of the system directly related to the accuracy of flight state analysis. To solve this problem, a phase space multi-scale principle manifold learning noise reduction method based on double tree complex wavelets was proposed. Firstly, the noisy signal was orthogonally decomposed into each scale band by the double-tree complex wavelet transform. Then according to the spatial distribution of signal and noise in each scale band, the wavelet coefficients in each scale were respectively identified by the phase space principle manifold learning algorithm, which could remove noise by projecting the scale coefficients from the high dimensional phase space to the intrinsic dimensional space projection. Finally, the denoised wavelet coefficients were reconstructed to obtain the vibration signal. Simulation and experimental results showed the effectiveness of the method.