Approaching the post non-linearity of blind mixtures by hybrid neural network

It is very difficult to approach the post non-linearity of blind mixtures. The recent neural networks for separating the post non-linear blind mixtures are limited to the diagonal non-linearity. In this paper a hybrid neural network is proposed to separate the post non-linearly mixed blind signals with cross-channel disturbance. This hybrid network consists of a new neural blind de-mixer for approximating the post non-linearity and a common network for separating the predicted linear mixtures. The blind de-mixer is made up of two subnets, which in total produce a `weak' non-linear operator and can approach relatively strong non-linearity by parameter-tuning. A six-step batch learning algorithm based on the fixed-point algorithm and information back-propagation is deduced. Preliminary results on a blind signal separation problem of two sources and four different types of post non-linearity indicate the effectiveness of our model.