Optical power monitoring based on back propagation neural network algorithm

In Dense Wavelength Division Multiplexing (DWDM) optical communication networks, the Optical Performance Monitor (OPM) can monitor key indicators for instance optical power, central wavelength, bandwidth, and optical signal-to-noise ratio in real-time. However, traditional OPM modules face the challenges like low accuracy, high cost, and unavoidable physical impacts. This paper proposes using the Back Propagation Neural Network (BPNN) algorithm to improve optical performance monitoring accuracy by reconstructing spectrum graphs to calculate optical power. Using simulated spectrum graphs as the dataset, the model is trained and validated with 0 dBm signals and evaluated with 5, 10, 15, and 20 dBm signals. Results show that the BPNN algorithm outperforms traditional OPM algorithm in reconstructing spectrum graph, improving optical power calculation accuracy and simplifying computational steps. With BPNN algorithm, the Mean Square Error (MSE) of the reconstructed spectrum is approximately 0.0011. This method not only enhances monitoring accuracy but also reduces errors in the OPM module. Although the training complexity increases, the prediction complexity has been decreased. Future research could optimize the structure and parameters of the BPNN model to improve its performance and explore its applicability in other optical communication systems.