Supervised regression modelling for mitigation of four-wave mixing in dense wavelength-division multiplexing systems

Abstract

A recent global crisis associated with COVID-19 has encouraged millions of people to work from home, thus causing a drastic increase in overall network traffic, data-rate requirements and end network capabilities. This has also produced more noise, cross-talk and undesirable optical-fibre nonlinearities, especially a four-wave mixing (FWM) effect that deteriorates performance of dense wavelength-division multiplexing (DWDM) systems. A presence of FWM in the DWDM systems imposes increasing complexity and latency of networks, and decreases their spectral efficiency. In its turn, this degrades efficient utilization of optical bandwidth. To mitigate the above problems, we suggest a supervised regression modelling (SRM). A relevant SRM-DWDM approach performs self-parametric optimization of the DWDM systems with machine-learning techniques and finds real trade-offs among various factors that affect the FWM. Our model reduces complexity of modelling and computational time, resulting in accurate and reliable prediction of parameter values. We also evaluate the performance of our SRM-DWDM technique by comparing its data with the iterative results obtained for different parameters (e.g., output signal-to-noise ratio, Q-factor, signal power and noise power). Finally, we specify the procedures necessary for global optimization of DWDM systems.