Influence of Fiber Nonlinearity on the Capacity of Optical Channel Models

Licentiate thesis, 2017

The majority of today’s global Internet traffic is conveyed through optical fibers. The ever-increasing data demands has pushed the optical systems to evolve from using regenerators and direct-direction receivers to a coherent multi-wavelength network. The future services like cloud computing and virtual reality will demand more bandwidth, so much so that the so called capacity-crunch is anticipated to happen in near future. Therefore, studying the capacity of the optical system is needed to better utilizing the existing fiber network. The capacity of the dispersive and nonlinear optical fiber described by the nonlinear Schrödinger equation is an open problem. There is a number of lower bounds on the capacity which are mainly obtained based on the mismatched decoding principle or by analyzing simplified channels. These lower bounds either fall to zero at high powers or saturate. The question whether the fiber-optical capacity has the same behavior as the lower bounds at high power is still open as the only known upper bound increases with the power unboundedly. In this thesis, we investigate the influence of the simplifying assumption used in some optical channel models on the capacity. To do so, the capacity of three different memoryless simplified models of the fiber-optical channel are studied. The results show that in the high-power regime the capacities of these models grow with different pre-logs, which indicates the profound impact of the simplifying assumptions on the capacity of these channels. Next, we turn our attention to the demodulation process which is usually done by matched filtering and sampling. It is shown that by deploying a proper demodulation scheme the performance of optical systems can be improved substantially. Specifically, a two-user simplified memoryless WDM network is studied, where the effects of nonlinear distortion are considered in the model. It is shown that unlike matched filtering and sampling, with the optimal demodulator, the symbol error rate decreases to zero at high power.