【摘要】 为解决时域和频域盲解卷积算法在时延长的情况下计算量大、不适于水声信道的实时计算问题,本文研究了一种基于二阶统计量的时域与频域相结合的快速盲解卷积算法。该算法首先对接收数据进行分段,段内数据进行分块处理,并给出一个新的在宽带情况下的代价函数表达范式;然后采用自然梯度法调整自适应权矩阵,进行块内数据离线、段间数据在线更新的自适应滤波,以获得源信号的估计。仿真结果表明,该方法结合了宽带和窄带算法的优点,对于线性卷积混合的接收数据取得了很好的分离效果,且计算量小,更适于在实时信号处理中应用。更多还原

【Abstract】 To solve the problem of huge computation of blind deconvolution in time-domain and frequency-domain algorithms in fast processing real underwater acoustic environment,a new algorithm based on the two-order cumulant is proposed.Firstly,the received data are segmented and each of the segment data is blocked.Then,a general broadband formulation of a new cost function is introduced.And the natural gradient descent method is used to optimize the weighted matrix of the adaptive filter.Finally,each block is processed off-line and each segment on-line to obtain the source evaluation.Experimental results show that the algorithm combines the wide-band with the narrow-band algorithms and gives good separation for the received data of the linear convolution mixture.Thus,it is more suitable for the real-time signal processing because of its lower computation.更多还原