为克服传统盲源分离算法分离效果差、计算量大且输出信号尺度模糊的缺点,提出了一种新型频域快速盲源分离算法.该算法在分析时域水声信号混合模型的基础上,构建新型频域混合模型,采用混合神经网络计算某一频率的分离矩阵,以此来估计全局分离矩阵.新算法较好地克服了尺度模糊问题,极大地减小了计算量,增加了分离算法的灵活性,分离性能较好.水声信号仿真实验和湖试实验均验证了算法的有效性.%A novel fast blind source separation algorithm in frequency domain is proposed to overcome the shortcomings of traditional blind source separation algorithms,including poor performance,high computational complexity,and ambiguity of output signal.In the proposed algorithm,a new mix model in frequency domain is accepted based on the mixmodel of underwater in time domain,then the global separated matrix is achieved by the separated matrix at some frequency computed by the multineural network.The novel algorithm has qualities of true scales,less computational complexity and flexible separation.The validity of the proposed algorithm is proved by simulations in underwater acoustic channel and lake experiments.
展开▼