On the Use of SVR based Machine Learning Method for Nonlinearities Mitigation in Short Range Fronthaul Links

Abstract

In this paper machine-learning (ML) dependent digital-predistortion (DPD) methodology is presented that overcomes the impairments of signal and non-linearities in Optical Front-hauls with the use of Radio over Fiber (RoF) systems. Volterra methods have been realized in the past for this purpose, however, they are not straight forward while classical artificial networks necessitate the optimum arrangement. The proffered support vector regression (SVR) method using grid search lessens the issues in Volterra based DPD and traditional artificial neural networks techniques In this work, the experimental evaluation for Long Term Evolution 20 MHz, 256 quadrature amplitude modulated signal using 850 nm Multi Mode Vertical cavity surface emitting lasers and Multi-Mode Fiber is linearized using SVR based DPD. The experimental analysis finds that SVR-DPD gathers the RoF non-linearities hence proving to be extraordinarily robust.