E-SpecFormer: Edge-Efficient Transformer for RF Spectrum Monitoring

Monitoring the radio frequency (RF) spectrum is crucial for various applications, from telecommunications to security. This paper introduces E-SpecFormer (Edge Spectrum monitoring Transformer), an innovative, edge-efficient Transformer model specifically designed for end-to-end automatic modulation recognition and the detection of covert channels. A key innovation within E-SpecFormer is LiTAN (Linear Tanh Attention Network), a novel attention mechanism that eliminates the need for Softmax and LayerNorm operations. This design choice significantly reduces computational complexity while simultaneously enhancing accuracy in RF-related tasks. E-SpecFormer is offered in four scalable variants (Nano, Small, Medium, Large) to accommodate diverse hardware constraints, making it highly adaptable for various edge deployments. Evaluations on the RadioML2018 dataset for modulation recognition show the Nano variant achieving 86.5% accuracy for SNRs > 0 dB. For hardware Trojan-based covert channel detection, it reaches 94.2% accuracy. Remarkably, these results are achieved with fewer than 10,000 parameters and processing speeds up to 92 microseconds per frame on FPGA/CPU co-execution, outperforming state-of-the-art edge models at a fraction of their cost. This establishes E-SpecFormer as a highly effective solution for real-time spectrum intelligence on Internet of Things (IoT) devices.