论文部分内容阅读
与自然梯度盲源分离算法相比,不完整自然梯度算法避免因源信号非平稳或幅值快速变化而引起的数值不稳定.在深入分析和推导该算法的基础上,针对其中非线性激活函数难以确定的困难,提出一种利用峰度对激活函数进行自适应选择的改进算法.该算法无需已知源信号的先验信息,既保留了不完整自然梯度算法恢复非平稳源信号的优势,又可使其适用于服从任意分布的源信号.仿真比较结果表明,该方法性能优于选择正切函数作为激活函数的不完整自然梯度算法,分离效果较好.
Compared with the natural gradient blind source separation algorithm, the incomplete natural gradient algorithm avoids the instability of the numerical value caused by the non-stationary source signal or rapid amplitude changes.Based on the in-depth analysis and derivation of the algorithm, the nonlinear activation function An improved algorithm is proposed to utilize kurtosis to adaptively select the activation function.This algorithm does not need to know the prior information of the source signal and retains the advantages of the incomplete natural gradient algorithm to recover the non-stationary source signal, But also make it suitable for subjecting to any distributed source signal.The simulation results show that the proposed method is superior to the incomplete natural gradient algorithm with the choice of tangent function as the activation function and the separation effect is better.