DSLN: Securing Internet of Things Through RF Fingerprint Recognition in Low-SNR Settings

Abstract

The explosive growth of Internet of things (IoT) has mandated the security of data access. Although authentication methods can enhance network security, their vulnerability to malicious attacks may be a barrier for the wide deployments in IoT scenarios. To address the security issue, we advocate the use of physical layer security through radio-frequency (RF) fingerprint recognition. Observing that most RF fingerprint recognition methods show a degradation of performance under low signal-to-noise ratio (SNR) environments, we present a dynamic shrinkage learning network (DSLN) to enhance security for IoT applications, particularly in the setting of low SNR. We design a novel dynamic shrinkage threshold for improving the accuracy of recognition under low-SNR environments. Additionally, we design an identity shortcut for reducing the running time of RF fingerprint recognition. In comparison with convolutional neural network (CNN), recurrent neural network (RNN) and a hybrid CNN+RNN network (CRNN), our proposed DSLN yields accuracy improvements of up to 20%. Moreover, DSLN can reduce running time by up to 60%, indicating its great potential to a real-time IoT system, e.g., an intelligent automotive system.

Fine-Grained Radio Frequency Fingerprint Recognition Network Based on Attention Mechanism

“…Previous works [38][39][40] have typically exploited element-wise summation or concatenation to fuse multi-scale features, which means that semantically strong parts are weighted the same as semantically weak parts. Inspired by [41], an ASFFM is proposed to adaptively learn the spatial weight of fusion for feature maps at each scale.…”

Radio Frequency Fingerprinting Identification of Few-Shot Wireless Signals Based on Deep Metric Learning

“…They can realize automatic feature extraction and recognition of the input data through training. Wu et al [11] proposed the dynamic shrinkage learning network (DSLN) for IoT identification; the method includes a dynamic shrinkage threshold which improves the low signal-to-noise ratio (SNR) recognition performance, and an identity shortcut which increases the running speed. Zong et al [12] used an improved version of the VGG-16 model to classify signals and trained 1,000 signals from five transmitters.…”