Wireless Interference Recognition With Low-Complexity Sparse Multilayer Perceptron

Wireless interference recognition (WIR) refers to the process of identifying and mitigating interference to ensure the reliability and security of communication systems. In this letter, we propose a novel low-complexity WIR network, named layered fusion sparse multilayer perceptron (LFS-MLP). Firstly, the proposed model employs a layered information fusion sparse connection strategy, which combines linear layers and depthwise convolution (DWConv) to process the spatial dimensions and channels of features, effectively integrating global and local information with low complexity. Secondly, we incorporate a bidirectional spatial attention (BSA) module to further enhance the model’s recognition accuracy. Finally, a multi-scale training strategy is employed to improve the training efficiency. Simulation results demonstrate that the proposed methodology achieves higher recognition accuracy while maintaining lower computational complexity compared to its existing WIR counterparts.