基于变分贝叶斯算法和MLP网络的后非线性混合盲源分离方法研究

传统的后非线性模型往往要求其后非线性函数是可逆的,否则无法进行源信号的分离。然而在实际中,这一要求并不完全满足。针对此不足,结合变分贝叶斯推论和多层感知器网络,提出一种改进的多层感知器后非线性模型,它通过多层感知器来模拟后非线性函数,实现对不可逆后非线性函数混合的盲分离。仿真和实验结果表明该方法是有效的。

Post-nonlinear blind separation of the source signals based on variational bayesian theory and MLP

In the traditional Post-nonlinear blind separation model of the sources, the post-nonlinearities is always required to be invertible functions. However, in practical not all models can suffice to this condition. In order to overcome this deficiency, combining of Bayesian inferring and MLP networks. a new improved method of MLP Post-nonlinear model is proposed. In the proposed method, the post-nonlinearities is modeled with multi-layer perception (MLP) networks, which also works for non-invertible post-nonlinearities. The simulation and experiment results show that the proposed method is very effective and has good robustness.