Tandem structure neural network-based channel power optimization in wavelength-division multiplexing systems

For C-band wavelength-division multiplexing (WDM) transmission systems, achieving balanced output channel power is a practical goal for performance maintenance, given the moderate fiber nonlinearity and sometimes outdated control and monitoring devices in these deployed systems. To accomplish this, we propose a digital twin (DT)-based tandem neural network (NN) structure for optimizing launch power profile. The proposed structure consists of two components: an optimization NN, which generates the optimized launched channel power based on a target output power profile, and a forward DT NN, pre-trained to predict output power profile after multi-span transmission. The proposed method enables fast and efficient launch power optimization on simulated links with arbitrary channel loadings and has been validated using open-source experimental data from links with heterogeneous spans and naked amplifiers. The power profile ripple can be reduced from more than 20 dB to 1.2 dB, providing an effective solution for practical C-band WDM systems.