Machine Learning for Performance Enhancement in Fronthaul Links for IOT Applications

We present A-typical machine learning (ML) based digital predistortion (DPD) solution for performance enhancement in analog optical front-hauls (OFH) for internet of things (IoT) based applications. Volterra based DPD has been realized in the past which becomes quite cumbersome due to complexity and choice of coefficients. Whereas the traditional Artificial Neural Networks techniques require time and optimization to determine the best model configuration. The proposed support vector regression (SVR) method is used that alleviates the nonlinearities and uplifts the OFH performance optimally. In this work, the experimental evaluation is made for 5G new radio (NR) signal having 256 quadrature amplitude modulation using 1550 nm Mach Zehnder Modulator and dispersion compensation fiber having 1 km link length. The experimental results suggest that SVR-DPD results in performance enhancement as compared to traditional volterra methods such as generalized memory polynomial, hence proving to be exceptionally operational.