Reinforcement Learning Based Information Fusion for Harmonic Estimation

The far-field detection of the helicopter has been a hot issue with the development of aeroacoustic theory. The aerodynamic noise generated by helicopters is the main detection carrier for helicopter far-field detection. The rotor noise is a major source of aerodynamic noise of the helicopter, which mainly comes from harmonic noise. Therefore, estimating the fundamental frequency of harmonic signals accurately is a critical issue. The detection range and robustness can be improved by using distributed microphone arrays instead of a single array. In this paper, the reinforcement learning based information fusion for harmonic estimation algorithm is proposed to solve the inconsistency of parameter estimation. The fast Non-linear Least Squares (NLS) algorithm is used to estimate the fundamental frequency in a single array. The more accurate estimation frequency can be obtained using the Q-learning algorithm to fuse the estimation results from distributed microphone arrays. The Fisher information is designed as the reward function in reinforcement learning. The proposed algorithm is verified by numerical simulations. The simulation results show that the proposed algorithm is feasible and effective for the distributed microphone arrays information fusion.