Artificial neural networks for modeling and simulation of communication systems with nonlinear devices

Nonlinear devices and subsystems present formidable challenges in the analysis of communication systems, and a major motivation for using the simulation approach. Typically nonlinear subsystems are originally described in terms of nonlinear differential equations (NLDE). Directly implementing the numerical solution of the NLDE into simulation block models can be computationally intensive as well as numerically unstable. We present here a simulation methodology that uses artificial neural networks (ANN) to build and efficiently simulate block models of nonlinear devices and subsystems within larger communication system models. We illustrate the usefulness of this approach and the validity of our analysis by showing significant run time savings in the simulation of an optical time-division multiple-access (OTDMA) architecture that involves a two input optoelectronic "AND" device.<>