论文部分内容阅读
针对现有独立分量分析算法的分离效果依赖于非线性对比函数的选择,并且无法有效地分离超高斯和亚高斯混合信号这一现象,提出了一种基于遗传算法的独立分量分析算法,该算法采用直方图法根据信号的样本序列来估计信号的概率分布,解决了信号间互信息的计算问题,然后通过遗传算法最小化信号间的互信息,实现了对线性混叠信号的分离;同时,针对标准遗传算法存在的一些缺点如局部搜索能力差、容易出现早熟收敛等,提出了一种改进遗传算法,提高了遗传算法的寻优能力。对模拟信号的分离结果表明,基于改进遗传算法的独立分量分析算法的性能优于FastICA算法,对亚高斯和超高斯信号的混合信号具有优异的分离能力。模拟仿真实验结果同时也证实了改进遗传算法的寻优能力。
The separation effect of the existing independent component analysis algorithm relies on the selection of non-linear contrast function and can not effectively separate the super-Gaussian and the sub-Gaussian mixed signal. An independent component analysis algorithm based on genetic algorithm is proposed. Histogram method is used to estimate the probability distribution of the signal according to the sample sequence of the signal, which solves the problem of calculating the mutual information between signals, and then separates the linear aliasing signal by minimizing the mutual information between signals through genetic algorithm. At the same time, Aiming at some shortcomings of the standard genetic algorithm, such as poor local search ability, easy premature convergence and so on, an improved genetic algorithm is proposed to improve the searching ability of genetic algorithm. The results of the separation of the analog signals show that the performance of the independent component analysis algorithm based on the improved genetic algorithm is better than that of the FastICA algorithm, and it has excellent separation ability for mixed signals of the sub-Gaussian and super-Gaussian signals. Simulation results also confirm the optimization ability of improved genetic algorithm.