Lightweight Channel Estimation Method for Dynamic OFDM System Based on Deep Learning

This paper presents a lightweight channel estimation technique for Orthogonal Frequency Division Multiplexing (OFDM) systems based on deep learning. At the heart of the method is a Convolutional Neural Network (CNN) that integrates a Multi-Scale Large Kernel Convolution (MLKC) module and a Convolutional Gating (CG) unit. The MLKC module, utilizing diverse large kernel sizes and a residual architecture, adeptly captures multi-scale features, enhancing the model's responsiveness to dynamic channel conditions. The CG unit optimizes the feature aggregation process through a spatial attention mechanism, improving the model's discernment of salient features. Experiments demonstrate that the proposed method outperforms existing approaches across various channel environments, while maintaining a lower count of learnable parameters, showcasing a dual advantage of high efficiency and superior performance.