Embedding-Assisted Attentional Deep Learning for Real-World RF Fingerprinting of Bluetooth

A scalable and computationally efficient framework is designed to fingerprint real-world Bluetooth devices. We propose an embedding-assisted attentional (Mbed-ATN) suitable for fingerprinting actual Its generalization capability analyzed in different settings the effect of sample length anti-aliasing decimation demonstrated. The embedding module serves as a dimensionality reduction unit that maps high dimensional 3D input tensor 1D feature vector further processing by ATN module. Furthermore, unlike prior research this field, we closely evaluate complexity model test its with dataset collected under time frame experimental setting while being trained on another. Our study reveals 9.17× 65.2× lesser memory usage at 100 kS when compared benchmark -GRU Oracle models respectively. Further, proposed Mbed-ATN showcases 16.9× fewer FLOPs 7.5× trainable parameters Oracle. Finally, show subject greater lengths 1 MS, results 5.32× higher TPR, 37.9% false alarms, 6.74× accuracy challenging setting.