Future Internet

Semantic communication systems leverage deep neural networks to extract and transmit essential information, achieving superior performance in bandwidth-constrained wireless environments. However, their vulnerability to backdoor attacks poses critical security threats, where adversaries can inject malicious triggers during training to manipulate system behavior. This paper introduces Selective Communication Unlearning (SCU), a novel defense mechanism based on Variational Information Bottleneck (VIB) principles. SCU employs a two-stage approach: (1) joint unlearning to remove backdoor knowledge from both encoder and decoder while preserving legitimate data representations, and (2) contrastive compensation to maximize feature separation between poisoned and clean samples. Extensive experiments on the RML2016.10a wireless signal dataset demonstrate that SCU achieves 629.5 ± 191.2% backdoor mitigation (5-seed average; 95% CI: [364.1%, 895.0%]), with peak performance of 1486% under optimal conditions, while maintaining only 11.5% clean performance degradation. This represents an order-of-magnitude improvement over detection-based defenses and fundamentally outperforms existing unlearning approaches that achieve near-zero or negative mitigation. We validate SCU across seven signal processing domains, four adaptive backdoor types, and varying SNR conditions, demonstrating unprecedented robustness and generalizability. The framework achieves a 243 s unlearning time, making it practical for resource-constrained edge deployments in 6G networks.

Faced with the access of a large number of devices, and for mobile vehicles with high speeds, some situations may be far from the communication range of the current edge node, resulting in a significant increase in communication latency and energy consumption. To ensure the effectiveness of task execution for mobile vehicles under high-speed conditions, this paper regards intelligent vehicles as edge nodes and establishes a dynamic offloading model in Vehicle-to-Vehicle (V2V) scenarios. A dynamic task offloading strategy based on optimal stopping theory is proposed to minimize the overall latency generated during the offloading process while ensuring the effectiveness of task execution. By analyzing the potential migration paths of tasks in V2V scenarios, we construct a dynamic migration model and design a migration benefit function, transforming the problem into an asset-selling problem in optimal stopping theory (OST). At the same time, it is proven that there exists an optimal stopping rule for the problem. Finally, the optimal migration threshold is determined by solving the optimal stopping rule through dynamic programming, guiding the task vehicle to choose the best target service vehicle. Comparisons between the proposed TMS-OST strategy and three other peer offloading strategies show that TMS-OST can significantly reduce the total offloading latency, select service vehicles with shorter distances using fewer detection attempts, guarantee service quality while lowering detection costs, and achieve high average offloading efficiency and average offloading distance efficiency.

Bioclimatic monitoring at vineyard scale is essential for irrigation management and disease-risk assessment, yet many systems rely on expensive commercial stations or generic IoT nodes with limited validation and little focus on small and medium-sized winegrowers. This application-driven engineering work investigates whether decision-support-grade bioclimatic data for precision viticulture can be obtained from a low-cost station, by proposing a solar-powered proximal node that integrates soil, plant, and atmospheric sensors on a dedicated PCB that communicates via LoRaWAN. The node operates in a 15-min cycle, with sensing parameters selected to provide the minimum information required for key Precision Viticulture applications. It was deployed in a commercial vineyard side by side with a commercial station, quantifying sensor agreement, communication reliability, and energy consumption. The results show low error rates and consistent agronomic interpretation of environmental conditions, disease risk, precipitation events, and soil and water dynamics. The LoRaWAN link reached a 97% packet-delivery ratio with an average consumption of about 2.5 Wh per day. Material cost is approximately 260 €, one order of magnitude lower than a comparable station. These results indicate that, under real vineyard conditions and compared with a commercial reference, the proposed low-cost system provides agronomically useful, reliable bioclimatic monitoring.