Performance estimation of super combined DWDM system employing machine learning

In this work, a 32-channel DWDM system has been evaluated by employing modified duobinary return to zero (MDRZ) and differential phase shift keying (DPSK) modulation techniques at alternate channels. Performance of the system has been evaluated in terms of gain, noise figure, optical signal-to-noise ratio (OSNR), quality- factor (Q-factor) and noise figure (NF). The DPSK modulated channels result in higher values of gain, OSNR and Q-factor with low noise figure values as compared to MDRZ channels. The reduction in FWM components have been reported in case of DPSK modulated channels and by employing erbium-ytterbium doped fiber amplifier (EYDFA). Further, Machine Learning (ML) techniques have been applied for gain spectrum and performance estimation. Levenberg-Marquardt's (LM) provides best fit as compared to other ANN models in case of gain spectrum estimation. The k-nearest neighbors (KNN) and support vector machine (SVM) algorithms have been tested for impairments analysis and performance estimation based on root mean square error (RMSE) and coefficient of determination (R2) as performance indicators. The best-performing model for gain, OSNR, Q-factor and NF are ANN-LM, KNN fine tree, KNN fine tree and SVM cubic, respectively.