Performance Enhancement of Optical Comb Based Microwave Photonic Filter by Machine Learning Technique

Abstract

Machine learning technique is employed to design a microwave photonic filter (MPF) consisting of an optical comb generator and a phase modulator (PM), to realize centralized management of radio signal delivery in optical networks. The proposed comb generator successfully generates up to 103 optical carriers with adjustable wavelength spacing and number of optical carriers. It exhibits large tunability, flat band response and high tone-to-noise ratio. Thus, the proposed optical comb-based filter is capable to provide flexible tunability with respect to its center frequency and 3-dB bandwidth. In our experimental setup, different wavelength spacing settings and number of optical carriers are fed into the PM for modulation and their corresponding frequency responses are measured by a network analyzer. The experimental data have been analyzed to correlate with the simulation results and theoretical predictions. Besides, a subcarrier multiplexing (SCM) technique can be applied to a multi-user optical system incorporating the proposed filter since its frequency response is varied by the accumulated dispersion. After measuring the characteristic of the proposed filter, a set of frequency responses is collected and fed into the convolutional neural network (CNN) model to obtain the inverse mapping between frequency response to the wavelength spacing and fiber length. As a result, the well-trained model can successfully predict the wavelength spacing and fiber length with high accuracy.