An Intelligent and Lightweight Physical Layer Authentication Scheme for MEC-enabled IoT Networks

This paper proposes a lightweight Three-Layer Convolutional Neural Network(3L-CNN) based physical layer authentication (PLA) method for mobile edge computing-enabled IoT (MEC-IoT) networks. A novel channel state information (CSI) processing architecture is established where real and imaginary components are transformed into two-channel images with 64x64 resolution for network inputs. Two data augmentation techniques, Average Data Augmentation (ADA) and Exponentially Weighted Average(EWA) are developed, to enhance temporal correlation preservation and mobility pattern extraction in mobile scenarios, effectively mitigating training data scarcity for mobile devices. The core 3L-CNN architecture remains streamlined, employing progressive feature extraction through three convolutional layers with 64x8, 32x16, and 6x32 configurations, optimized by a hybrid loss function combining 50% Negative Log-Likelihood and 50% Cross-Entropy to refine classification boundaries. The architecture demonstrates 99% authentication accuracy for 10 devices configuration and maintains 96.6% accuracy for 30 devices respectively. It also exhibits superior robustness with 95% accuracy at 0 dB SNR. This lightweight solution achieves comparable performance to complex models while reducing training time by 66%, making it suitable for resource-constrained mobile edge computing-enabled IoT applications.