Symbol Detection and Channel Estimation using Neural Networks in Optical Communication Systems

In optical wireless communication (OWC) systems, channel estimation and detection of the transmitted symbols have been conventionally performed using analytical methods assuming that the optical channel follows a certain model, e.g., free-space model, input-dependent noise model, or Poisson model. In practical OWC systems, channels do not necessarily follow a specific model. Hence, it is difficult, if not impossible, to derive analytical models that provide optimal performance in realistic optical channels. Motivated by the success of neural networks in estimation and classification in various fields, we propose a neural network-based methodology for detection and estimation for OWC that does not rely on a channel model. Simulation results show that the proposed learning-based estimation and detection schemes achieve the optimal performance of the maximum likelihood detector under different channel state information assumptions.