Proposal and FPGA implementation of DBSCAN clustering nonlinear detector for MMW RoF system

Abstract

We here propose and experimentally validate an improved density-based spatial of applications with noise clustering approach for mitigating the nonlinear distortions of analog millimeter-wave (MMW) radio over fiber (RoF) systems, to offer a self-adaptivity to various modulation formats as no training process and initialization parameters (e.g., signal constellation size) are required. In addition, fueled by the Manhattan distance clustering criteria, the FPGA implementation of such a machine-learning algorithm is achieved for verifying its practical feasibility. Validated by experiments, our proposal can effectively improve the nonlinearity tolerance of a 60-GHz MMW-RoF system transmitting single-carrier 64-QAM, 128-QAM and 256-QAM signals. Specifically, it helps to obtain a 1.25-dB improvement in the receiving sensitivity for the 64-QAM transmission in a fiber-wireless MMW channel consisting of 5-km fiber and 1-m wireless links.