A Study Of Blind Equalization Algorithms Based On The Multi-Layer Feed-Forward Neural Networks

Blind equalization is an adaptive equalization technique without a training sequence; it can equalize the properties of the channel by using the statistic properties of the signals received and improve communication by overcoming the interferences of the codes. It has been a major focus in digital communication.The artificial neural network is a theoretical mathematical model for the cerebral neural network; it is an information-processing system that is based on stimulating the structure and function of the cerebral neural networks; it can perform complicated operations and establish nonlinear relationships. The use of neural networks for designing equalizer is of important theoretical significance and practical value. The primary work of the study is as follows:1. The drawbacks of blind equalization algorithm based on the three-layer feed-forward neural networks are analyzed; a new one based on four-layer feed-forward neural networks is proposed. Computer simulations show that the equalization effects of the algorithm of the four-layer is improved.2. A new equalization algorithm on five-layer is discussed and its better function than that of the four-layer algorithm is shown by computer simulations.3. Their respective properties of the blind equalization algorithms based on three- , four- and five-layer feed-forward neural networks are studied and compared, the conclusion is drawn that with the increase of the layers of the network, there is less steady residual error and the bit error rate decreases while the convergence speed becomes lower.