Application and Optimization of Deep Learning-Based Modulation Format Recognition in Long-Distance Fiber Optic Communication

With the rapid development of information technology, ultra-wideband signals need to quickly switch modulation formats to ensure high-quality communication when wireless channels are subject to interference. This paper proposes a neural network-based modulation format recognition algorithm, which utilizes the power spectral density in the frequency domain as input and classifies modulated signals through a fully connected neural network. Simulation results demonstrate that the algorithm can accurately identify five modulation formats—BPSK, QPSK, 8PSK, 8QAM, and 16QAM—after 10 km of fiber transmission, achieving a classification accuracy of 100% and a bit error rate below 3.8e-3. Compared to the random forest classifier, the fully connected neural network exhibits superior performance in terms of accuracy. This research provides an efficient and reliable solution for modulation format recognition in optical communication systems. In the future, the network structure can be further optimized, and time-domain and frequency-domain features can be fused to enhance performance.