International Conference on Wireless Communications and Signal Processing (WCSP)

Channel Capacity Constrained Secure Semantic Communications Against Eavesdropping

Yuxiang Liu, Xu Wang, Zheng Shi, Jintao Wang, Guanghua Yang, Shaodan Ma

Published Oct 23, 2025• 0 citations

Abstract

With the breakthrough advancements in deep neural networks (DNNs) and natural language processing (NLP), semantic communication systems have emerged as a pivotal direction in the evolution of sixth-generation (6 G) networks. However, the broadcast nature of wireless channels inherently introduces user privacy risks, underscoring the urgency for robust security solutions. Existing secure semantic communication approaches predominantly focus on network-layer encryption, which often entails complex operations and carries inherent risks of key leakage. In contrast, physical layer security (PLS) harnesses the intrinsic time-varying nature and randomness of wireless channels, obviating the need for cumbersome cryptographic mechanisms. To address this critical challenge, this paper proposes a channel capacity-constrained secure semantic communication framework. The core design objective is to maximize channel capacity for legitimate channels while confining the ergodic secrecy capacity within a predefined security threshold. By developing a differentiable mutual information estimator, we establish a dual-channel capacity estimation model-encompassing both legitimate users and eavesdroppers-that is jointly optimized with the semantic communication system. Furthermore, we introduce a secure sentence similarity metric to quantitatively assess the security performance of semantic communication. Simulation results validate that despite a slight degradation in transmission accuracy, the proposed scheme achieves a significant enhancement in communication security.

DOI: 10.1109/wcsp68525.2025.1010343

Related papers

Physical-Layer Secure Transmission for Semantic Communication Systems

Zijian Cao, Hua Zhang, Le Liang +3 more

As a promising paradigm for the sixth-generation (6G) networks, task-oriented semantic communication significantly enhances transmission efficiency. However, it faces complex security challenges, particularly the risk of eavesdropping due to the open nature of wireless channels. To address this issue, we propose a secure semantic communication framework that integrates physical-layer secure beamforming (SBF) to safeguard semantic information from eavesdropping. Specifically, we design an SBF network to generate SBF vectors that focus signal beams on legitimate users to enhance signal power while directing designed artificial noise toward potential eavesdroppers to strengthen jamming. To further improve system adaptability across varying channel conditions, we introduce attention-based channel-aware modules that dynamically optimize the encoding, decoding, and beamforming processes based on perceived channel state information. Furthermore, task-oriented artificial noise is employed to degrade the task performance of eavesdroppers more effectively. Finally, a stepwise training strategy with task-specific loss functions is employed to jointly optimize the SBF and semantic modules, maximizing the performance gap in downstream tasks between legitimate users and eavesdroppers. The simulation results demonstrate that the proposed approach effectively maintains task performance for legitimate users while significantly suppressing eavesdroppers, outperforming conventional methods.

IEEE Transactions on CommunicationsSep 16, 20252 citations

Information-theoretic Secrecy in Multi-user Channels

Ersen Ekrem

Inherent openness of the wireless medium imposes stronger challenges on the security of wireless communications. Information-theoretic security addresses these challenges at the physical layer by using tools from wireless communication theory, signal processing and information theory. In information-theoretic security, physical layer communication is intelligently designed to exploit the characteristics of the wireless medium, such as fading, interference, cooperation, and multi-dimensional signaling, in order to provide or improve security. In this dissertation, we study the security of several fundamental wireless network configurations from an information-theoretic perspective. First, we study the Gaussian multiple-input multiple-output (MIMO) wiretap channel. In this channel, the transmitter sends a common message to both the legitimate user and the eavesdropper. In addition to the common message, a private message is sent only to the legitimate user, which needs to be kept hidden as much as possible from the eavesdropper. We obtain the entire capacity-equivocation region for this channel model. In particular, we show the sufficiency of jointly Gaussian auxiliary random variables and channel input to evaluate the existing single-letter description of the capacity-equivocation region due to Csiszar-Korner. Next, we study the secure broadcasting problem, where a transmitter wants to have secure communication with multiple legitimate users in the presence of an external eavesdropper. We study several special cases of the secure broadcasting problem. First, we consider the degraded multi-receiver wiretap channel, and establish its secrecy capacity region. Second, we consider the parallel less noisy multi-receiver wiretap channel, and obtain its common message secrecy capacity and sum secrecy capacity. Third, we consider the parallel degraded multi-receiver wiretap channel for the two-user and two-sub-channel case, and obtain its entire secrecy capacity region. Finally, we consider a parallel channel model with two sub-channels, where the transmitter can use only one of the subchannels at any time, and characterize its secrecy capacity region. Then, we study the two-user Gaussian MIMO broadcast channel with common and confidential messages. In this channel model, the transmitter sends a common message to both users, and a confidential message to each user which needs to be kept perfectly secret from the other user. We obtain the entire capacity region of this channel. We also explore the connections between this channel model and its non-confidential counterpart, i.e., the Gaussian MIMO broadcast channel with common and private message. Next, we consider the Gaussian MIMO multi-receiver wiretap channel and obtain its secrecy capacity region for the most general case. We first show that even for the single-input single-output (SISO) case, existing converse techniques fall short of proving the secrecy capacity region, to emphasize the need for a new proof technique, which we develop by using the relationships between the Fisher information and the differential entropy. Using this new proof technique, we obtain the secrecy capacity region of the degraded MIMO channel. We then establish the secrecy capacity region of the general MIMO channel by using the channel enhancement technique in conjunction with the capacity result we obtained for the degraded MIMO channel. For the general MIMO channel, we show that dirty-paper coding (DPC) combined with stochastic encoding attains the entire secrecy capacity region. Then, we study the multi-receiver wiretap channel for a more general scenario, where, in addition to confidential messages, the transmitter sends public messages to the legitimate users, on which there are no secrecy constraints. First, we consider the degraded discrete memoryless channel, and obtain inner and outer bounds for the capacity region. These inner and outer bounds match for certain cases, providing the capacity region. Second, we obtain an inner bound for the general discrete memoryless channel by using Marton's inner bound. Third, we consider the degraded Gaussian MIMO channel, and show that jointly Gaussian auxiliary random variables and channel input are sufficient to exhaust the inner and outer bounds. Finally, we provide an inner bound for the capacity region of the general Gaussian MIMO channel. Next, we focus on the multiple access wiretap (MAC-WT) channel whose capacity region is unknown. We consider a special class of MAC-WT channels which we call the weak eavesdropper class, where each user's link to the legitimate receiver is stronger than its link to the eavesdropper. For this class of channels, we develop an outer bound for the secrecy capacity region, which partially matches the achievable region in an n-letter form. We evaluate a looser version of our outer bound for the Gaussian case, and show that our outer bound is within 0.5 bits/channel use of the achievable rates along the individual secrecy rates for all weak eavesdropper Gaussian MAC-WT. Then, we investigate the effects of user cooperation on the secrecy of broadcast channels by considering the cooperative relay broadcast channel (CRBC). We propose an achievable scheme that combines Marton's coding scheme for broadcast channels and Cover and El Gamal's compress-and-forward (CAF) scheme for relay channels. For the Gaussian CRBC, we show that both users can have positive secrecy rates, which is not possible for scalar Gaussian broadcast channels without cooperation. We further investigate the effects of user cooperation on secrecy by considering the multiple access channel with generalized feedback (MAC-GF), which can be viewed as the MAC-dual of the CRBC. We propose a CAF-based achievable secrecy rate region for the MAC-GF. Specializing our results to a Gaussian MAC-GF, we present numerical results which demonstrate that cooperation can improve secrecy for the MAC-GF. Next, we study the two-user one-eavesdropper discrete memoryless compound wiretap channel, and provide the best known lower bound for the secrecy capacity of this compound channel. We evaluate this achievable secrecy rate for the Gaussian MIMO case by using DPC. We show that this achievable secrecy rate achieves at least half of the secrecy capacity of this Gaussian MIMO compound wiretap channel, and also attains the secrecy capacity when the eavesdropper is degraded with respect to one of the two users. Then, we study the degraded compound multi-receiver wiretap channel (DCMRWC), which, in addition to a group of eavesdroppers, has two groups of users, namely the stronger group and the weaker group. We study two different communication scenarios for this channel. In the first scenario, there is only one eavesdropper, and the transmitter sends a confidential message to each group of legitimate users while keeping both messages secret from the eavesdropper. In the second scenario, we study the DCMRWC with layered messages without any restriction on the number of eavesdroppers. For both scenarios, we obtain the secrecy capacity region for the discrete memoryless channel, the parallel channel, and the Gaussian parallel channel. For the Gaussian MIMO channel, we obtain the secrecy capacity region when there is only one user in the second group. Next, we study the two-user fading broadcast channel and obtain its ergodic secrecy capacity region. We show that, thanks to fading, both users can have simultaneous secure communication with the transmitter, although this is not possible in the scalar non-fading Gaussian broadcast channel where only one user can have secure communication. This simultaneous secrecy of both users is achieved by an opportunistic communication scheme, in which, at each time instant, the transmitter communicates with the user having a better channel gain. Then, we study the secure lossy transmission of a vector Gaussian source to a legitimate user in the presence of an eavesdropper, where both the legitimate user and the eavesdropper have vector Gaussian side information. We obtain an outer bound for the rate, equivocation and distortion region. Moreover, we obtain the maximum equivocation at the eavesdropper when there is no constraint on the transmission rate. By using this maximum equivocation result, we show two facts. First, for this problem, in general, Wyner-Ziv scheme is suboptimal, although, it is optimal in the absence of an eavesdropper. And, second, even when there is no transmission rate constraint, an uncoded transmission scheme is suboptimal; the presence of an eavesdropper necessitates the use of a coded scheme to attain the maximum equivocation. Finally, we revisit the secure lossy source coding problem. In all works on this problem, either the equivocation of the source at the eavesdropper or the equivocation of the legitimate user's reconstruction of the source at the eavesdropper is used to measure secrecy. We first propose the relative equivocation of the source at the eavesdropper with respect to the legitimate user as a new secrecy measure. We argue that this new secrecy measure is the one that corresponds to the natural generalization of the equivocation in a wiretap channel to the context of secure lossy source coding. Under this new secrecy measure, we provide a single-letter description of the rate, relative equivocation and distortion region, as well as its specializations to degraded and reversely degraded cases. We investigate the relationships between the optimal scheme that attains this region and the Wyner-Ziv scheme.

Jan 1, 20120 citations

Physical layer security through secure channel estimation

Fawad Ud Din

Wireless communication is highly susceptible to eavesdropping attacks where a malicious user can exploit the broadcast nature of the wireless signals to eavesdrop on confidential information. At the physical layer, there are multiple possible avenues to utilize the inherent randomness of the wireless transmission medium to achieve secure communication. One of the lucrative possibilities is to use the channel state information (CSI) to establish a secure communication link because it is vital in decoding the received signal at the transmitter. This technique is commonly referred to as discriminatory channel estimation (DCE) where the channel estimation performance is intentionally degraded at the eavesdropper as compared to the legitimate nodes to achieve secure communication.In this thesis, we capitalize on full-duplex (FD) transmissions to transmit the pilot signals to obscure the channel estimates from the eavesdropper. FD transmission severely limits the estimation performance at the eavesdropper while the legitimate nodes can acquire robust estimates. The secrecy of channel estimates from the eavesdropper provides a bandwidth-efficient security technique as the channel estimation requires less bandwidth than the data transmission. We have presented the achievable secrecy capacity analysis along with the simulation analysis to illustrate the performance improvements achieved by the proposed novel secure channel estimation. In this thesis, we also study the impact of artificial noise (AN) to improve the secrecy achieved by the use of FD transmission of pilot signals. For a strategically located eavesdropper, we have presented novel DCE techniques where AN signals are transmitted along with pilot signals using FD transmission. The AN signals confuse the nearby eavesdropper to establish a secure and robust communication link. We have also presented novel local adaptive power allocation algorithms, where each legitimate node performs power allocation in the absence of statistical channel characteristics regarding the eavesdropping channel. We have provided an in-depth simulation analysis including a location-based simulation analysis to illustrate that the secrecy performance achieved by the proposed DCE techniques. The simulation results indicate that the proposed DCE techniques achieve robust secure communication against a strategically located eavesdropper: while the other DCE techniques are unable to avoid the leakage of information to the eavesdropper

Jan 1, 20210 citations

Semantic Communication-Assisted Physical Layer Security Over Fading Wiretap Channels

Xidong Mu, Yuanwei Liu

A novel semantic communication (SC)-assisted secrecy transmission framework is proposed. In particular, the legitimate transmitter (Tx) sends the superimposed semantic and bit stream to the legitimate receiver (Rx), where the information may be eavesdropped by the malicious node (EVE). As the EVE merely has the conventional bit-oriented communication structure, the semantic signal acts as the type of beneficial information-bearing artificial noise (AN), which not only keeps strictly confidential to the EVE but also interferes with the EVE. The ergodic (equivalent) secrecy rate over fading wiretap channels is maximized by jointly optimizing the transmit power, semantic-bit power splitting ratio, and the successive interference cancellation decoding order at the Tx, subject to both the instantaneous peak and long-term average power constraints. To address this non-convex problem, both the optimal and suboptimal algorithms are developed by employing the Lagrangian dual method and the successive convex approximation method, respectively. Numerical results show that the proposed SC-assisted secrecy transmission scheme can significantly enhance the physical layer security compared to the baselines using the conventional bit-oriented communication and no-information-bearing AN. It also shows that the proposed suboptimal algorithm can achieve a near-optimal performance.

arXiv:2402.14581International Conference on CommunicationsJun 9, 202416 citations

Semantic-Security Capacity for the Physical Layer via Information Theory

Ziv Goldfeld, Paul Cuff, Haim Henri Permuter

Physical layer security can ensure secure communication over noisy channels in the presence of an eavesdropper with unlimited computational power. We adopt an information theoretic variant of semantic-security (SS) (a cryptographic gold standard), as our secrecy metric and study the open problem of the type II wiretap channel (WTC II) with a noisy main channel is, whose secrecy-capacity is unknown even under looser metrics than SS. Herein the secrecy-capacity is derived and shown to be equal to its SS capacity. In this setting, the legitimate users communicate via a discrete-memory less (DM) channel in the presence of an eavesdropper that has perfect access to a subset of its choosing of the transmitted symbols, constrained to a fixed fraction of the block length. The secrecy criterion is achieved simultaneously for all possible eavesdropper subset choices. On top of that, SS requires negligible mutual information between the message and the eavesdropper's observations even when maximized over all message distributions. A key tool for the achievability proof is a novel and stronger version of Wyner's soft covering lemma. Specifically, the lemma shows that a random codebook achieves the soft-covering phenomenon with high probability. The probability of failure is doubly-exponentially small in the block length. Since the combined number of messages and subsets grows only exponentially with the block length, SS for the WTC II is established by using the union bound and invoking the stronger soft-covering lemma. The direct proof shows that rates up to the weak-secrecy capacity of the classic WTC with a DM erasure channel (EC) to the eavesdropper are achievable. The converse follows by establishing the capacity of this DM wiretap EC as an upper bound for the WTC II. From a broader perspective, the stronger soft-covering lemma constitutes a tool for showing the existence of codebooks that satisfy exponentially many constraints, a beneficial ability for many other applications in information theoretic security.

Software - Science, Technology and EngineeringJun 1, 20165 citations

Wireless Physical Layer Security: Towards Practical Assumptions and Requirements

Biao He

The current research on physical layer security is far from implementations in practical networks, arguably due to impractical assumptions in the literature and the limited applicability of physical layer security. Aiming to reduce the gap between theory and practice, this thesis focuses on wireless physical layer security towards practical assumptions and requirements. In the first half of the thesis, we reduce the dependence of physical layer security on impractical assumptions. The secrecy enhancements and analysis based on impractical assumptions cannot lead to any true guarantee of secrecy in practical networks. The current study of physical layer security was often based on the idealized assumption of perfect channel knowledge on both legitimate users and eavesdroppers. We study the impact of channel estimation errors on secure transmission designs. We investigate the practical scenarios where both the transmitter and the receiver have imperfect channel state information (CSI). Our results show how the optimal transmission design and the achievable throughput vary with the amount of knowledge on the eavesdropper's channel. Apart from the assumption of perfect CSI, the analysis of physical layer security often ideally assumed the number of eavesdropper antennas to be known. We develop an innovative approach to study secure communication systems without knowing the number of eavesdropper antennas by introducing the concept of spatial constraint into physical layer security. That is, the eavesdropper is assumed to have a limited spatial region to place (possibly an infinite number of) antennas. We show that a non-zero secrecy rate is achievable with the help of a friendly jammer, even if the eavesdropper places an infinite number of antennas in its spatial region. In the second half of the thesis, we improve the applicability of physical layer security. The current physical layer security techniques to achieve confidential broadcasting were limited to application in single-cell systems. The primary challenge to achieve confidential broadcasting in the multi-cell network is to deal with not only the inter-cell but also the intra-cell information leakage and interference. To tackle this challenge, we design linear precoders performing confidential broadcasting in multi-cell networks. We optimize the precoder designs to maximize the secrecy sum rate with based on the large-system analysis. Finally, we improve the applicability of physical layer security from a fundamental aspect. The analysis of physical layer security based on the existing secrecy metric was often not applicable in practical networks. We propose new metrics for evaluating the secrecy of transmissions over fading channels to address the practical limitations of using existing secrecy metrics for such evaluations. The first metric establishes a link between the concept of secrecy outage and the eavesdropper's ability to decode confidential messages. The second metric provides an error-probability-based secrecy metric which is often used for the practical implementation of secure wireless systems. The third metric characterizes how much or how fast the confidential information is leaked to the eavesdropper. We show that the proposed secrecy metrics enable one to appropriately design secure communication systems with different views on how secrecy is measured.

Jan 1, 20161 citation