Modulation format and number of users classification in multipoint free-space optical communication using convolutional neural network

A methodology based on convolutional neural network (CNN) is proposed for joint classification of transmitting user number and modulation format in a multiuser free-space optical communication (FSOC) link. The proposed methodology relies on amplitude information of received mixed signal. In-phase and quadrature components of users that are sharing time and bandwidth resources transmitting into the same optical wireless access point and interfering within each other are analyzed. The proposed approach utilizes the constellation diagrams of the received mixed symbols to generate image data sets that are fed into CNN input. The designed CNN model with three convolutional layers was tested for: varying image resolutions, image-data set size, varying number of received symbols, and atmospheric turbulence to identify optimal parameters and processing time for system design and implementation. The results indicate that the CNN model can blindly and accurately identify the communicating device number and their optical modulation format with classification accuracy up to 100% for various SNRs. Moreover, the CNN demonstrated robustness against atmospheric turbulence and suggested immunity to additive noise. Therefore, the proposed methodology proved to be a promising and feasible solution for practical implementation of an intelligent optical wireless receiver for aerial and terrestrial FSOC links.