Blind Equalization with a Linear Feedforward Neural Network

摘要

In this paper, we introduce a linear feedforward neural network for blind equalization in digital communications. The approach is based on a fundamental theorem, which makes the training procedure of the neural network very simple. The training is equivalent to a stochastic approximation algorithm and can be implemented recursively every time a sample data is received. As usual, the received signal is oversampled so that the channel can be described by a full-column rank matrix; the neural network searches for the inverse of the matrix. A simulation example is given to illustrate the performance of the neural network.