*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Transmission Performance Optimization in Fiber-wireless Access Networks using Machine Learning Techniques
Committee:
Dr. Matthieu Bloch, ECE, Chair, Advisor
Dr. Xiaoli Ma, ECE
Dr. David Anderson, ECE
Dr. John Barry, ECE
Dr. Shiwen Mao, Auburn Univ.
Abstract: The objective of this dissertation is to enhance the transmission performance in the fiber-wireless access network through mitigating the vital system limitations of both analog radio over fiber (A-RoF) and digital radio over fiber (D-RoF), with machine learning techniques being systematically implemented. The first thrust is improving the spectral efficiency for the optical transmission in the D-RoF to support the delivery of the massive number of bits from digitized radio signals. Advanced digital modulation schemes like PAM8, discrete multi-tone (DMT), and probabilistic shaping are investigated and implemented, while they may introduce severe nonlinear impairments on the low-cost optical intensity-modulation-direct-detection (IMDD) based D-RoF link with a limited dynamic range. An efficient deep neural network (DNN) equalizer/decoder to mitigate the nonlinear degradation is therefore designed and experimentally verified. Besides, we design a neural network based digital predistortion (DPD) to mitigate the nonlinear impairments from the whole link, which can be integrated into a transmitter with more processing resources and power than a receiver in an access network. Another thrust is to proactively mitigate the complex interferences in radio access networks (RANs). The composition of signals from different licensed systems and unlicensed transmitters creates an unprecedently complex interference environment that cannot be solved by conventional pre-defined network planning. In response to the challenges, a proactive interference avoidance scheme using reinforcement learning is proposed and experimentally verified in a mmWave-over-fiber platform. Except for the external sources, the interference may arise internally from a local transmitter as the self-interference (SI) that occupies the same time and frequency block as the signal of interest (SOI). Different from the conventional subtraction-based SI cancellation scheme, we design an efficient dual-inputs DNN (DI-DNN) based canceller which simultaneously cancels the SI and recovers the SOI.
