基于奇异值分解的频率估计新算法

为了提高信号频率估计的精确度，提出了一种新的自适应滑动窗奇异值算法（sliding window adaptive SVD, SWASVD）.该算法基于奇异值算法将包含信号信息的矩阵分解到一系列奇异值和奇异值矢量对应的时频子空间中，从而分离信号信息与其他噪声信息的特点，推导了连续奇异值算法，产生两个辅助矩阵，在行列式处理中，采用减少秩的方法消除噪声，推导出的近似矩阵减少了复杂计算，使用matlab进行仿真，与多重信号分类谱估计法（MUSIC）进行了比较.结果表明，该新算法使用了滑动窗的概念，对陡峭信号变化有很好的鲁棒性，应用该方法可以在频率估计方面获得更准确的结果.

New approach based on singular value decomposition for frequency estimation

A new method based on SVD was proposed to deal effectively with frequency estimation.The method based on SVD's high quality subspace estimation,designed a new tracking technique for sliding window data matrices,then derived from the classical bi-orthogonal iteration,produced two assistant matrices.In array processing,the approach reduced rank to have a noise-cleaning effect,introduced a low-rank approximation of the data matrix,and offered faster tracking response to sudden changes.Results showed that compared to MUSIC method,the method proposed produced synthetic signals in the frequency estimation context and showed excellent performance in frequency estimation and robustness to abrupt variations.