Multi-Carrier Filtering Method for Satellite Mobile Communication System Based on CNN+LSTM

To address the issues of adjacent channel interference, nonlinearTo address the issues of adjacent channel interference, nonlinear hardware resource limitations faced by multi and hardware resource limitations faced by multi-carrier signal demodulation in satellite mobile communication systems, this paper proposes a joint filtering method based on Convolutional Neural Networks (CNN) and Long Short-Term Memory networks (LSTM). By designing multi-scale convolutional kernels to hierarchically extract time-frequency features, combined with bidirectional LSTM to model temporal correlations, and introducing an attention mechanism to dynamically focus on key frequency bands, it replaces the explicit extraction and downconversion process of traditional Kaiser window half-band filters. Experiments show that under the condition of 15dB Signal-to-Noise Ratio (SNR), this scheme reduces the Symbol Error Rate (SER) by $40 \%$ compared to traditional methods ($0.7 \%$ vs $1.8 \%$), improves the Adjacent Channel Rejection Ratio by 4 dB, and reduces hardware resource usage by $50 \%$.