Deep-Learning-Based Physical Layer Authentication for Industrial Wireless Sensor Networks.

Area of Science:

• Cybersecurity
• Wireless Sensor Networks
• Machine Learning

Background:

• Industrial wireless sensor networks (IWSNs) face significant security challenges.
• Existing authentication methods often lack efficiency and scalability for IWSNs.
• Physical layer (PHY) authentication offers a robust security enhancement.

Purpose of the Study:

• To propose a deep learning (DL)-based PHY layer authentication framework for IWSNs.
• To evaluate the performance of different DL algorithms for sensor node authentication.
• To identify lightweight and efficient authentication solutions for edge computing in IWSNs.

Main Methods:

• Implementation of three DL algorithms: Deep Neural Network (DNN), Convolutional Neural Network (CNN), and Convolutional Preprocessing Neural Network (CPNN) for PHY layer authentication.
• Utilized Adaptive Moment Estimation (Adam) and minibatch techniques for accelerated neural network training.
• Performance evaluation through simulations and experiments using Universal Software Radio Peripherals (USRPs).

Main Results:

• The improved CPNN-based algorithm demonstrated minimal computational requirements and extremely low latency.
• This CPNN approach enables lightweight, multi-node PHY layer authentication suitable for resource-constrained sensor nodes.
• Edge-side training of neural networks facilitates efficient authentication within edge computing systems.

Conclusions:

• The proposed DL-based PHY layer authentication framework significantly enhances IWSN security.
• The CPNN algorithm presents a promising solution for lightweight and efficient authentication in IWSNs, particularly within edge computing architectures.
• Experimental validation confirms the practical applicability and effectiveness of the proposed methods.