Extracting feature signal from the gravity earth tide based on improved ICA

Independent Component Analysis (ICA) is the focal point in blind signal processing (BSP). It aims to separate the relatively independent signals from the mixed signal source. Gravity earth tide signal is a kind of complex mixed signal which is caused by the Moon and the Sun, which means that the ICA can be used to separate gravity earth tide signal. Because of the it is so sensitive to the initial values, that affects the separation effect and even result in in-convergence if the initial values are not chosen appropriately. In order to solve the problem proposed, a new method based on PBIL(Population Based Incremental Learning) to combine ICA is formed. The learning rate and the learning model in PBIL are used to optimize the separation matrix in ICA. The simulation results show that the new method effectively avoids the problems proposed. The new method separates gravity earth tide signal into three parts, which are long-period waves, diurnal wave, and semi diurnal wave. Every part represents the signal which is corresponding to the theory frequencies of the harmonic component in gravity earth tide signal.