An adaptive neuro-fuzzy filter design via periodic fuzzy neural network

This paper presents an adaptive filter which uses periodic fuzzy neural network (PFNN) to treat the equalization of nonlinear time-varying channels. The proposed PFNN is based on a neural network learning ability and fuzzy if-then rules structure. In general, training a fuzzy neural network (FNN, or neuro-fuzzy system) to represent some type of plant and system is relatively straightforward and many methods exist. For a given limited amount of information, the PFNN is applied to solve the estimation of the periodic signals. Several examples are shown to illustrate the effectiveness of the proposed approach. The back-propagation learning algorithm with adaptive (or optimal) learning rate is used to speed up the learning. Furthermore, the PFNN is applied to be a nonlinear time-varying channel equalizer with simple structure and fast inference. Efficiency and advantages of the PFNN are verified by these simulations and comparisons.