Design of an SCRFNN-based nonlinear channel equaliser

The design of a self-constructing recurrent fuzzy neural network (SCRFNN)-based digital channel equaliser is proposed. The structure and the parameter learning phases are performed concurrently, so the SCRFNN presents quite a high speed for online processing. Specifically, the structure learning is based on the partition of input space, and the parameter learning is based on the supervised gradient descent method using a delta adaptation law. It has been demonstrated that a SCRFNN-based digital channel equaliser possesses the ability to recover channel distortion effectively. The performance of SCRFNN has been compared with the adaptive-based-network fuzzy inference system (ANFIS) and the optimal Bayesian solution. The simulations have been carried out in both real-valued and complex-valued nonlinear channels to ensure the flexibility of the proposed equaliser. The experimental results show that the performance of SCRFNN is close to the Bayesian optimal solution and ANFIS, while the hardware requirement of a trained SCRFNN-based equaliser is much lower.