基于阵列的神经网络水声通信信号多参数联合估计算法OA北大核心

To address the challenges posed by complex and highly variable underwater acoustic channel(UWAC)with severe attenuation,a novel neural network algorithm was proposed for multi-parameter joint estimation.Multi-channel spectrograms derived from array signals were utilized by the algorithm to improve the detection probability of UWAC signals under non-cooperative conditions and extend the sensing range.A carrier frequency label assignment strategy was designed,in which carrier frequency served as the key physical feature to distinguish different signals,thereby effec-tively mitigating interference from out-of-band signals and noise.End-to-end multi-task learning was adopted to simulta-neously perform signal detection,modulation recognition,and joint estimation of the number of signal sources,carrier frequency,bandwidth,and direction of arrival,eliminating the complex beamforming process typically required in tradi-tional methods before detection and recognition.Simulation results confirm that accurate signal estimation is reliably achieved even in the presence of array element position mismatches and when signals are obscured by noise.Lake ex-periments further demonstrate the practicality and generalization capability of the proposed algorithm.