Low-Complexity and Robust Modulation Format Recognition Method Based on Multi-Stage Feature Extraction

A low-complexity, robust modulation format recognition (MFR) method based on multi-stage feature extraction is proposed for adaptive processing of varying modulation formats in future optical networks. First, signal distribution features are obtained using histogram statistics. These features are then reduced in dimensionality by extracting harmonic components through Fourier series decomposition. Finally, a lightweight neural network recognizes modulation format labels from the distribution features, ensuring efficient and accurate MFR. Simulations and experiments validate the performance of the proposed method. Simulation results show a 100% recognition rate for 4QAM, 8QAM, 16QAM, 32QAM, 64QAM, 128QAM, and 256QAM at Signal-to-Noise Ratios (SNRs) of 0 dB, 4.5 dB, 8 dB, 10.5 dB, 13 dB, 16 dB, 20 dB, achieving MFR at a bit error rate threshold of 0.04 with some SNR margin in an additive white Gaussian noise channel. Compared to the amplitude distribution method and other 9 MFR methods, the proposed method has the widest SNR identification range, covering all seven QAM formats from 4QAM to 256QAM. Additionally, it demonstrates the lowest computational complexity of O(N) among all methods. Experimental validation in WDM long-distance transmission systems under varying optical SNRs and channel counts further confirms the method's effectiveness and robustness.