Evaluation of Photonic Reservoir Computing for Use in Short Reach PAM-4 Transmission Systems

Artificial neural networks for use in real-time optical communication systems are difficult to realize in hardware due to their complexity. Here, photonic reservoir computing as an extension of the recurrent neural network offers a good tradeoff between performance and feasibility. As a result, some parts of the digital signal processing may be moved to the optical domain. A single laser with a delayed feedback loop can be used for reservoir computing for which we evaluate the equalization performance for data from a 56 GBd single sideband PAM-4 transmission system. We compare the results with the standard Kramers-Kronig reception.