Abstract: This paper investigated Artificial Intelligence (AI)-aided communication signal processing approach to address the model mismatch and performance degradation of traditional Orthogonal Frequency Division Multiplexing (OFDM) receivers caused by rapidly time-varying channels, doppler spectrum spreading, and multi-domain compound interference in high-speed railway environments. Based on the equivalent baseband model of OFDM systems, this paper constructed three lightweight AI-aided processing models, including a Deep Neural Network (DNN) model for channel estimation and equalization, a two-dimensional Convolutional Neural Network (2D-CNN) model for modulation recognition based on time-frequency feature extraction, and a 1D-CNN model for time-domain interference suppression. Comparative simulations were carried out against classic algorithms including the Least Square (LS) and Minimum Mean Square Error (MMSE) channel estimation methods, the traditional statistical feature-based modulation classifier, and the notch filtering combined with clipping interference suppression algorithm. The experimental results show that the constructed DNN model can effectively compensate the residual error of time-varying channel estimation under low Signal-to-Noise Ratio (SNR), outperforming the traditional LS algorithm and approximating the optimal performance of the MMSE algorithm. In small-sample scenarios, the overall performance of the 2D-CNN model is roughly equivalent to that of conventional algorithms, with performance gains under extreme SNR conditions. The 1D-CNN model exhibits significantly stronger anti-interference capability and signal recovery performance than the traditional notch filtering and clipping algorithm under moderate-to-severe composite interference.