Research On Digital Predistortion Combined Polynomial With Neural Network In The RoF System

Radio-over-Fiber(RoF)is a transmission technology that can support large transmission rate and mobility of users.Due to its flexibility,bandwidth efficiency and low cost,RoF technology is considered as one of the major choices for the optical wireless aceess network.In the RoF system,nonlinear distortion components causes the adjacent channel interference due to light sub-carrier modulation,whose main source is the modulator and power amplifier especially.Predistortion technology has the characteristics of low cost,easy implementation which is widely used in the linearization of nonlinear system.This paper mainly studies the digital predistortion technology to deal with the nonlinear distortion in RoF system.The main work of this paper is as follows:This paper studies the structure of RoF system,the nonlinearity,memory effect of its main components and predistortion technology the comprehensively.This paper also analyzed and compared the look-up table method,Volterra model and neural network model.Secondly,based on the theoratical analysis,the experimental platform is built to use radial basis function(RBF)neural network model in the nonlinear characteristics of RoF system modeling and linearization.The neural network training time is greatly reduced compared with the forward neural network model,but its computational complexity is very high while accurately describe the nonlinearity of RoF systemIn order to reach a acompromise of linear effect and computational complexity,this paper proposes a digital predistortion technology which combines memory polynomials and radial basis function(RBF)neural network model.This model uses the memory polynomial model to compensate the RoF system's non-linearity,then the memory effect is compensated by the radial basis function neural network model.In this paper,we use the LTE double carrier signal to linearize RoF system experimently.The experimental results show that the new model have a better performance at suppressing out-of-band regeneration and Third-Order-Intermodulation(IMD3)compared with memory polynomials and radial basis function neural network,the model has an average improvement of 13dB for ACPR of RoF system,the average improvement of IMD3 is 17dB,and effectively increases the ability to linearize RoF systems...