Generative AI for Secure Physical Layer Communications: A Survey

TL;DR

The paper addresses the challenge of securing wireless communications at the physical layer amid dynamic channels and evolving threats by surveying Generative AI techniques such as GANs, VAEs, AEs, and diffusion models. It maps how these models can enhance confidentiality, authentication, availability, resilience, and integrity, including secure transceiver design, RF fingerprinting, anti-jamming, spoofing defense, anomaly detection, and data reconstruction. Key contributions include a structured synthesis of GAI-enabled methods across these security dimensions, critical assessment of their advantages and limitations, and a forward-looking set of directions (model improvements, multi-scenario deployment, resource efficiency, and secure semantic communication). The work highlights GAI’s potential to learn complex data distributions, simulate challenging channel conditions, and enable adaptive, robust security in next-generation networks, while also noting practical challenges such as training time, real-time adaptation, and deployment at scale.

Abstract

Generative Artificial Intelligence (GAI) stands at the forefront of AI innovation, demonstrating rapid advancement and unparalleled proficiency in generating diverse content. Beyond content creation, GAI has significant analytical abilities to learn complex data distribution, offering numerous opportunities to resolve security issues. In the realm of security from physical layer perspectives, traditional AI approaches frequently struggle, primarily due to their limited capacity to dynamically adjust to the evolving physical attributes of transmission channels and the complexity of contemporary cyber threats. This adaptability and analytical depth are precisely where GAI excels. Therefore, in this paper, we offer an extensive survey on the various applications of GAI in enhancing security within the physical layer of communication networks. We first emphasize the importance of advanced GAI models in this area, including Generative Adversarial Networks (GANs), Autoencoders (AEs), Variational Autoencoders (VAEs), and Diffusion Models (DMs). We delve into the roles of GAI in addressing challenges of physical layer security, focusing on communication confidentiality, authentication, availability, resilience, and integrity. Furthermore, we also present future research directions focusing model improvements, multi-scenario deployment, resource-efficient optimization, and secure semantic communication, highlighting the multifaceted potential of GAI to address emerging challenges in secure physical layer communications and sensing.

Figures (8)

• Figure 1: The structure of the survey paper, where we introduce GAI methods for physical layer security through Communication Confidentiality and Authentication (Section \ref{['CCA']}), Communication Availability and Resilience (Section \ref{['CAR']}), and Communication Integrity (Section \ref{['CI']}).
• Figure 2: The overall architecture of the AE-Based besser2019flexible and VAE-Based secure transceiver lin2020variational. Part A demonstrates a wiretap system model with AWGN. Part B illustrates the whole framework of AE-Based secure transceiver, which is trained by two loss functions: the mean-squared error between transmitter messages and reconstructed messages and the mutual information between the messages and the received symbols by the eavesdropper. In Part C, the VAE-Based secure transceiver adds additional loss function: KL divergence.
• Figure 3: Proposed CGAN training architecture in germain2021mobile. In Part A, the conditional information is the previous magnitudes of the CSI elements associated with time. The output of the discriminator is the probability value, representing the likelihood from zero to one based on its perception of whether the sample is fake or authentic. Part B illustrates the system model structure.
• Figure 4: The overall network structure in cai2020spectrum. Part A illustrates the generator, which is designed as an AE comprising a convolution layer, a fully connected layer, and a de-convolution layer. In Part B, two discriminator modules are crafted using a convolution network and are optimized to focus on local and global details, respectively. Part B describes the system model structure.
• Figure 5: Overall structure of the proposed anti-jamming spectrum access scheme in han2021better. In Part A, a GAN with the Generator (G) and Discriminator (D) is trained to complete missing spectrum data using historical data. In Part B, the generator (G) is implemented as SCN. The CSN utilizes the enriched spectrum data to assist users in selecting optimal communication channels for anti-jamming.