摘要： 针对传统独立分量分析(ICA)方法无噪假设的局限性，提出基于互累积量的有噪ICA方法。考虑含高斯噪声的瞬时混合模型，以观测信号的互累积量组成一系列对称矩阵，以对称矩阵的联合对角化程度为目标函数，采用粒子群优化算法对混合矩阵进行全局寻优。通过寻优得到混合矩阵，将有噪ICA转化为一维欠定ICA，基于奇异值分解法得到源信号的估计。仿真结果表明，与传统ICA方法相比，该方法对混合矩阵的估计精度较高，可以明显提高分离信号的信噪比。

关键词: 有噪独立分量分析, 欠定独立分量分析, 粒子群优化, 联合对角化, 奇异值分解, 瞬时混合模型

Abstract: In order to solve the limitation of the noise-free assumption in conventional Independent Components Analysis(ICA) methods, a noisy ICA method based on cross-cumulants is proposed. Considering the instantaneous mixture model with Gaussian noise, the cross-cumulants of observed signals are used to make up a set of symmetric matrixes. The joint diagonalization of those matrixes is utilized as the objective function, and the mixing matrix is optimized by particle swarm optimization. Through the estimated mixing matrix, the noisy ICA is converted to one-dimension underdetermined ICA, and the estimations of source signals can be obtained by singular value decomposition. Simulation results illustrate that, in contrast to the conventional ICA methods, the method proposed can get a more accurate estimation of the mixing matrix, and obviously improve the signal-to-noise ratio.

Key words: noisy Independent Component Analysis(ICA), underdetermined Independent Component Analysis(ICA), Particle Swarm Optimization(PSO), joint diagonalization, singular value decomposition, instantaneous mixing model

中图分类号: 

• TP391