Blind equalization in underwater acoustic communication by recurrent neural network with bias unit

Recurrent neural network structure is formed by adding bias unit to feedforward neural network (FNN), which was applied in underwater acoustic communication. The neural network by adding bias unit can take full advantage of statistical information of received signals; consequently, it raises convergence speed effectively and enhances the tracing ability of neural network blind equalization in time-varying channels, thus, equalization performance can be improved. Results of the simulation by computer and experimentation in a channel pool show that neural network with bias unit obtain better performance than traditional FNN in blind equalization of underwater acoustic channel.