An Improved UKF Algorithm for Extracting Weak Signals Based on RBF Neural Network
Guanyu Zhang, Jinxin Luo, Hao Xu, Yitian Wang, Tianhao Wang, Jun Lin, Yang Liu
Abstract
It is difficult to extract effective weak signals from complex noise environment in many engineering applications. To address the problem, an improved unscented Kalman filter (UKF) algorithm based on a radial basis function (RBF)-neural network (NN) is proposed in this paper. This algorithm achieves effective extraction of weak signals under the condition of low signal-to-noise ratio (SNR). It uses RBF-NN to establish a distributed system of state and observation models to replace the traditional UKF modeling method. Then, the RBF-NN is applied to build a constant coefficient model to modify the UKF filtering deviation value, and thus, the accuracy of the algorithm is improved. Finally, simulation experiments are conducted at different SNRs of surface nuclear magnetic resonance signals. Results show that effective weak signals can still be obtained when SNR = -10 dB, verifying the effectiveness of the algorithm.