中文摘要：

通过对水声信号进行相空间重构，并对重构后的轨迹矩阵进行奇异值分解，得到表征信号能量和噪声强度的奇异谱。对奇异谱进行分析，得到反映噪声强度的噪声平台。对于那些大于噪声平台的特征值，它们具有较大的方差，对应较大的信噪比。利用具有较大方差特征值对应的特征矢量重构状态空间，也就等效于得到了具有较大信噪比改善的重构。通过对3类多个样本的实际水声信号采用奇异值分解进行降噪处理，得到了较为满意的降噪效果，降噪后的信号波形基本上消除了噪声干扰．为水声信号的进一步处理奠定了基础。

英文摘要：

Noise reduction of underwater acoustic signals radiated by ships and other underwater vehicles plays an important role in both signal detection and feature extraction. Especially for extracting correctly nonlinear characteristic parameters of underwater acoustic signals, which are Lyapunov exponent, fractal dimention and information entrop. The results of noise reduction can severely affect the results of further processing. Singular value decomposition （SVD） which is very popular in matrix theory is used to achieve satisfactory noise reduction of underwater acoustic signals. Firstly , the phase space reconstructions of real signals collected from the sea are established according to Takens theorem. And then the covariance matrix derived from the phase space reconstruction is gotten. After performing SVD for covariance matrix, a singular spectrum which corresponds to the signal energy and noise level is obtained. Using this noise level the distinction between the deterministic signals and random noise can be telled. Twenty samples for each of three different types of real ship signals are chosen to perform noise reduction. The results show that the algorithm is effective and a much more cleaned signal can be obtained from this method.