Signal Processing Based on Machine Learning Optical Communication

Abstract

With the rapid development of machine learning, machine learning provides powerful tools to deal with problems in many fields. Its typical characteristic is the ability of self-learning and evolution. Signal equalization and optical spectral measurement are important problems in signal processing in optical communication, but there are few researches on the application of machine learning to solve this problem. This paper mainly focuses on the signal quality improvement and performance parameter monitoring in optical communication and applies machine learning to visible light communication equalization and spectrum analysis. The combination of machine learning algorithm and equalization technology enhances the ability of tracking channel characteristics, achieves intelligent learning and updating of equalizer, and introduces machine learning into spectrum analysis. The analysis of the spectrum was designed according to the intelligent mechanism of different machine learning algorithms. The original material of the analysis was the input data, which was adjusted according to the gap between the labels and output results corresponding to each set of data. All the test time was less than 0.8 s. SVM test time on wavelength, OSNR and bandwidth estimation is minimum (less than 0.34 s). Experiments show that the defined plane generated by support vector machine is more suitable for spectrum analysis, and the ability to summarize features is more suitable for spectrum analysis.