基于阵列的神经网络水声通信信号多参数联合估计算法  

Array-based neural network algorithm for multi-parameter joint estimation of underwater acoustic communication signals

作　　者：成乐 刘悦 胡正良[2] 朱宏娜[1] 罗斌[1] CHENG Le;LIU Yue;HU Zhengliang;ZHU Hongna;LUO Bin(School of Information Science and Technology,Southwest Jiaotong University,Chengdu 610031,China;College of Meteorology and Oceanology,National University of Defense Technology,Changsha 410073,China)

机构地区：[1]西南交通大学信息科学与技术学院,四川成都610031 [2]国防科技大学气象海洋学院,湖南长沙410073

出　　处：《通信学报》2025年第1期67-78,共12页Journal on Communications

基　　金：国家重点研发计划基金资助项目(No.2021YFC3101402,No.SQ2021YFF0500035)。

摘　　要：针对水声信道复杂多变且衰减严重等问题,为提升非合作条件下水声通信信号的检测概率并扩大感知范围,设计了一种新型基于阵列多通道时频谱输入的神经网络多参数联合估计算法。该算法通过引入载波频率标签分配策略,将载波频率作为区分不同信号的关键物理特征,有效避免了频带外信号和噪声的干扰;利用端到端的多任务学习,能够同时完成信号检测、调制模式识别,以及对信号个数、载波频率、带宽和波达方向的联合估计,从而避免了传统算法中需要先进行波束成形再进行检测识别的复杂流程。仿真实验结果表明,在阵列阵元位置失配和信号被噪声掩蔽的情况下,所提算法仍能实现准确的信号估计。进一步的湖上实验验证了所提算法的实用性和泛化能力。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 effectively mitigating interference from out-of-band signals and noise.End-to-end multi-task learning was adopted to simultaneously 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 traditional 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 experiments further demonstrate the practicality and generalization capability of the proposed algorithm.

关 键 词：多参数联合估计 波达方向估计 调制模式识别 阵列信号处理 神经网络 

分 类 号：TN911.7[电子电信—通信与信息系统]