Neural network based adaptive predistortion for radio over fiber links

Nonlinearity in the response of the optical source responsible in the electrical-to-optical conversion process limits the dynamic range performance in transmitting analog and multilevel digital signals over fiber optic link channels. Fiber-based wireless access schemes provide a unique possibility for digital linearization. We describe a technique for predistorting the dynamic response of the optical source using a Radial Basis Functional Neural Network (RBFNN). The input and output data from the link provide the samples used to train the RBFNN. A simple method employing the Least Squares (LS) algorithm is used to provide an optimization to the predistorter NN. Simulation results demonstrate the simplicity of this method compared to other NN architectures and linearization techniques. This result is used in this paper to demonstrate the feasibility of the technique in providing an adaptive predistortion solution for wireless applications.