CIF: Small: Statistical Approach to Signal Processing for Long-Haul Fiber-Optic Communication Sytems

University Of Virginia Main Campus, Charlottesville VA

Investigators

Abstract

Society depends on fiber-optic networks to carry its huge data demands. Recent literature on optical communications has predicted a plateau in the throughput increase of long-haul fiber-optic communication systems, estimating that the demand for throughput will exceed supply within the next ten years or so. The long term solution to this problem is the systematic replacement of all long-haul fiber (millions of miles just in the US) to more effective fibers. This research project explores a less drastic shorter-term solution. It seeks to answer the following question: if limiting assumptions made in deriving this throughput plateau are relaxed, by using more sophisticated channel modeling and signal processing, can the system capacity be increased sufficiently to delay or even obviate the costly fiber upgrade? In this project a full probabilistic description of the signal and noise co-propagation through optical fibers is developed so that statistical signal processing approaches can be effectively used on this difficult channel. A probabilistic representation of the fiber-optic system including its correlated, nonstationary, and nonGaussian qualities for wavelength division multiplexed systems and emerging elastic optical networks is first derived. The research includes the derivation of the statistics of each mode of a Gaussian mixture model for the joint signal and noise processes as they propagate through a long-haul fiber. The models developed are used to derive accurate closed-form predictions of the system performance. Signal processing algorithms and receiver designs that optimize the above performance metrics are then developed. The program leverages current models developed for deterministic signals. The focus is in generating accurate yet simple models to make them more generally applicable to both legacy and new optical networks.