KAN Enhanced CNN-BiLSTM for Accurate Modeling of Optical Fiber Channels

This paper proposes a hybrid model combining convolutional neural network (CNN), bidirectional long short-term memory (Bi-LSTM) and Kolmogorov-Arnold Network (KAN) for single mode fiber channel modeling. Simulation results demonstrate that the proposed model achieves noticeable improvement in modeling accuracy compared to conventional Bi-LSTM under long-haul transmission scenarios.