Accurate carrier frequency synchronization is essential for reliable OFDM semantic communications, since carrier frequency offset (CFO) not only impairs OFDM reception but also degrades semantic reconstruction. However, existing CFO estimation methods mainly rely on signal-domain observations and may become limited in joint integer and fractional CFO estimation under low-SNR and complex-channel conditions. To address this issue, this paper proposes a semantic-assisted CFO estimation framework for OFDM semantic communications, termed SeSy-IF. A semantic synchronization header is constructed from a reference image, and CFO is estimated from the distortion patterns observed in the reconstructed semantic output. A key insight is that integer and fractional CFO leave distinct semantic-domain signatures in the reconstructed output. Integer CFO mainly causes cyclic subcarrier displacement and global structural misalignment, whereas fractional CFO mainly induces phase perturbation and inter-carrier interference, resulting in finer-grained reconstruction degradation. Based on these properties, SeSy-IF performs staged joint estimation through integer candidate generation, fractional refinement, and semantic-domain screening. Experiments over AWGN, Rayleigh fading, and realistic channel conditions show that SeSy-IF achieves lower MSE and MDR than representative baselines in both fractional CFO estimation and joint integer and fractional CFO estimation.
Wireless simultaneous localization and mapping (SLAM) is a key enabler for integrated sensing and communication (ISAC) systems, yet its reliability is fundamentally challenged by rich multipath propagation, which gives rise to virtual targets and ambiguous environmental interpretations. This paper proposes a multipath-semantic, belief-propagation-based SLAM framework that explicitly embeds multipath propagation topology into the SLAM posterior inference. By introducing a physics–semantic dual-layer state representation, multipath interactions are modeled as latent semantic states, enabling joint localization, mapping, and multi-bounce (MB) multipath discrimination within a unified Bayesian formulation. The proposed framework supports robust SLAM inference and real-time discrimination of multipath-induced virtual targets of arbitrary order under general, non-linear agent motion. Simulation results and comparison with existing approaches based on Monte Carlo trials, together with real measurement data, validate the effectiveness of the proposed approach and illustrate its robustness under challenging multipath conditions.
Due to the rapid evolution of communication technologies, low-Earth orbit (LEO)-assisted networks have become a promising solution to provide seamless, high-throughput, and wide-coverage communication services. Meanwhile, multi-functional reconfigurable intelligent surfaces (MF-RISs) have gained significant attention for their self-sustainability and versatility. This paper investigates a multi-LEO downlink self-sustainable network enhanced by both satellite-mounted (S-MF-RISs) and ground-deployed (G-MF-RISs) MF-RISs. By jointly optimizing active beamforming at LEO satellites and configurations on S-/G-MF-RISs, we aim to maximize the achievable sum rate while satisfying the user service requirements and ensuring the self-sustainability restrictions on both LEO satellites and G-MF-RISs. To solve this problem, we propose an Adaptive-weighted fuSion decision TRAnsformer-based reinforcement Learning (ASTRAL) algorithm. Compared to the conventional deep reinforcement learning (DRL) method, the proposed ASTRAL approach further analyzes the long-term energy utilization strategy leveraging a decision transformer architecture. Furthermore, an adaptively-weighted fusion mechanism is designed, justifying the importance of the balance between cumulative planning and immediate adaptability. Simulation results validate the effectiveness of the proposed ASTRAL algorithm, with a performance improvement of approximately 22.4% in sum rate compared to the conventional deep deterministic policy gradient method. Moreover, integrating both S-MF-RISs and G-MF-RISs yields further enhancement of about 7% and 8.6% in rate performance relative to the scenarios lacking S-MF-RISs and G-MF-RISs, respectively.
In many critical application scenarios of modern wireless communications, such as wireless sensor networks and drone swarms, the information generated by high-density source nodes are subject to considerable correlation. This paper aims to investigate the problem of how the source correlation can be exploited to enhance the system load and error rate performance of state-of-the-art multiple access (MA). From a theoretical perspective, we first present the achievable description rate-region of distributed source compression and the achievable data-transmission rate-region of MA channel. We reveal that exploiting the source correlation can enlarge the admissible rate-region, and the two regions have to be matched so as to maximize the overall admissible rate of the system. Yet, a simple combination of existing distributed source coding (DSC) and MA schemes is insufficient to achieve this matching. It also cannot handle the interaction effects between them in practical scenarios. From a practical perspective, we put forth a new interleave-division MA with correlated sources (IDMA-CS) scheme. Given the statistics of the source correlation, the transmitters employ a low-density parity-check (LDPC) based DSC which computes the compressed information in the form of syndrome. Then, the distributively compressed messages are channel-encoded, interleaved, and transmitted simultaneously to the base station (BS). The BS receiver performs multiuser decoding (MUD) and DSC decoding. We propose a new “turbo-like” joint MUD and DSC decoding method that exploits the source correlation to facilitate MUD. Further, we optimize the rate profile based on variance evolution, where the constraints of both DSC and MA are considered. Numerical results show that the proposed IDMA-CS scheme can increase the system load to more than 1.3-3 times that of conventional IDMA when the source exhibits medium to strong correlation.
Traditional pilot spoofing attacks (PSAs) pose a threat to wireless time-division duplex (TDD) communications systems, but typically require prior knowledge of pilot sequences and network synchronization, and are vulnerable to randomized pilot designs. To overcome these limitations, we propose a novel nonreciprocal reconfigurable intelligent surface (NR-RIS)-enabled channel spoofing attack (CSA) that passively manipulates the wireless environment to assist the eavesdropper. By passively intervening in the uplink training phase and the downlink channel transmissions, the NR-RIS misleads the transmitter into employing downlink beamforming that is misaligned with the intended receiver and that can enhance the eavesdropping threat. We present several heuristic NR-RIS strategies for CSA using only partial channel state information (CSI), which can effectively deteriorate the secrecy of the legitimate system. Moreover, we demonstrate that CSA can be successfully implemented even without CSI by leveraging deep reinforcement learning techniques that configure the NR-RIS to enhance its eavesdropping capability. When rate feedback from the receiver is available to the adversary, the NR-RIS can induce the transmitter to enhance the data transmission rate, causing more information leakage. Simulation results verify the effectiveness of the proposed NR-RIS CSA approaches, revealing a critical vulnerability in TDD-based communication systems.
