Edge-Efficient Transformer for End-to-End RF Spectrum Monitoring

Short summary

E-SpecFormer is an edge-efficient transformer for automatic modulation and covert channel recognition in RF spectrum monitoring, featuring LiTAN—a Softmax- and LayerNorm-free attention mechanism that reduces complexity. The Nano variant achieves 86.5% accuracy on RadioML2018 (SNR>0dB) and 94.2% on hardware Trojan covert channel data with fewer than 10k parameters and 92 microsecond per-frame inference on FPGA/CPU. Code is available on GitHub, making it a practical solution for real-time spectrum intelligence on IoT devices.

• •Edge transformer with novel LiTAN attention (no Softmax/LayerNorm) for RF spectrum monitoring
• •Nano variant: <10k params, 86.5% modulation accuracy, 94.2% covert channel accuracy, 92us/frame on FPGA/CPU
• •Four scalable variants (Nano to Large) for diverse hardware constraints; code on GitHub

Generated with AI, which can make mistakes.