Dongfeng Yuan

The integration of integrated sensing and communication (ISAC) with backscatter communication (B-ISAC) presents promising applications in the field of internet of things (IoT), enabling low-power communication and improved system sensing performance. This paper proposes a novel reconfigurable intelligent surface (RIS) enabled B-ISAC system, wherein the RIS facilitates the backscattering of its own information to the base station (BS) and dynamically adjusts signal reflection paths to optimize communication and sensing performance at a low cost. An optimization problem is formulated to maximize the system communication mutual information (CMI) and sensing mutual information (SMI), and a deep unfolding algorithm based on the projected gradient ascent (DUPGA) is proposed. Specifically, inspired by the iterative process of the projected gradient ascent (PGA) algorithm, we design an interpretable neural network to achieve the BS hybrid beamforming and the RIS phase-shift design. Simulation results verify that the proposed DUPGA algorithm offers significant reductions in computational complexity compared to traditional alternating algorithms, while achieving similar CMI and SMI performance.

Semantic communication emerges as a pivotal technology for realizing the 6G vision. However, existing semantic-aware reconstruction systems rely on syntactic-level loss functions for optimization, failing to focus on the precise recovery of semantic information and resulting in suboptimal semantic fidelity. Meanwhile, the autoregressive decoding architecture adopted in text semantic communication introduces prohibitive high latency. To address these two issues, we propose a robust information bottleneck (RIB) guided non-autoregressive semantic communication (NASC) scheme, named RIB-NASC. First, we pioneer the integration of the RIB criterion into semanticaware reconstruction systems, formulating a direct optimization objective tailored for underlying semantic recovery and establishing an informativeness-robustness trade-off. Second, we derive a tractable variational lower bound for the RIB objective via variational approximation and a novel synonymous mapping-based semantic posterior estimation strategy. Third, we design a lightweight non-autoregressive semantic decoder architecture based on Transformer encoder, enabling high-speed parallel semantic decoding during inference. Extensive simulation results demonstrate that the RIB-NASC scheme significantly outperforms baseline schemes in terms of semantic recovery performance (BLEU score and sentence similarity) and achieves a decoding delay reduction of nearly 96.7% compared to traditional autoregressive decoding.

To simultaneously address the traffic congestion and computational load imbalance among roadside units (RSUs) in vehicular networks caused by uneven vehicles distribution, this paper proposes a Cluster-based Collaborative Route Guidance Strategy (CCRGS) by jointly taking into account the road congestion and RSU load. The overall framework comprises a personalized initial route planning method and a dynamic rerouting strategy. Specifically, a multi-factor road weight model is established by integrating travel time, RSU computational resource utilization, and road popularity. Based on this model, a bi-objective A* algorithm is developed to generate personalized routes that align with user preferences while balancing network load. By integrating a dynamic road weight update mechanism, the algorithm enables implicit inter-vehicle coordination. The density-based spatial clustering of applications with noise (DBSCAN) algorithm is adopted to group vehicles and construct cluster-specific subgraphs, enabling rational, localized decision-making. In addition, a periodic rerouting strategy is designed, combining status monitoring, adaptive vehicle selection, and route updates to timely cope with possible road congestion and RSU overload. Extensive simulations on a real-world urban road network demonstrate that, compared to the shortest path algorithm, CCRGS reduces the average travel time by 14.76%–46.17% and decreases the number of vehicles with unmet computational demand by 67.23%–95.04%, while significantly improves the load balance across both roads and RSUs. These results validate the effectiveness of CCRGS in enhancing traffic efficiency, service continuity, and the overall system stability in vehicular networks.

Asynchronous federated edge learning (Asy-FEEL) has drawn intensive attention due to its ability to effectively address the straggler issue caused by the heterogeneity of the participated mobile devices (MDs). The quality of Asy-FEEL depends highly on the number of participating MDs. Since the local training of federated learning consumes computation resources of MDs, it inevitably reduces the resources that can be devoted to their own tasks (OTs). Therefore, there is always a decreased incentive of MDs to participate in Asy-FEEL, subsequently reducing the amount of Asy-FEEL tasks executed by MDs, thereby failing in achieving satisfied Asy-FEEL performance. How to effectively utilize the limited resources and schedule Asy-FEEL tasks and OTs to satisfy the quality of service requirements on the OTs and at the same time encourage MDs to participate to execute Asy-FEEL tasks is of vital importance. To this end, a joint tasks scheduling and resource allocation problem is formulated and investigated within a dynamic multitasks mobile edge computing (MEC) network, where Asy-FEEL tasks and OTs coexist. Since the problem is a dynamic stochastic optimization problem, a Lyapunov-based dynamic joint tasks scheduling and resources allocation (Lya-DJTR) algorithm is proposed to determine tasks scheduling, computational resource and bandwidth allocation, and transmit power control at MDs simultaneously. Simulation results demonstrate the superiority of the proposed algorithm in improving the efficiency of Asy-FEEL while ensuring the real-time processing of OTs when compared to baseline algorithms.

Due to the reliance of task processing on specific service models, service caching has emerged as a promising technology for enhancing low-latency performance in vehicular edge computing (VEC) networks. When vehicular long-term service caching mismatches real-time task requirements, the mobility and autonomy of service vehicles enable them to fetch service models from providers actively. However, highly dynamic topology and resource constraints of VEC networks call for a more adaptable approach for jointly designing vehicular service fetching and task offloading. To this end, this paper aims at jointly optimizing active service fetching, task offloading, and communication and computing resource allocation to boost low-latency performance in two-tier caching-assisted VEC networks. In doing so, a total task completion delay minimization problem is formulated, where the social-mobility-aware network topology, vehicle classification, delay tolerance requirements, and resource constraints are taken into account. Since it is a mixed-integer non-linear programming (MINLP) problem, it is typically NP-hard. To solve it effectively, we decompose it into three subproblems, which can be efficiently solved by an iterative algorithm based on block coordinate descent (BCD) with low complexity. Extensive simulation results demonstrate that our proposed algorithm achieves fast convergence and reduces the task completion delay by 9.27%-59.77% compared to five representative baselines.

Collaborative machine learning has been a key technology to enhance the development of intelligent transportation systems with the advantage of privacy protection and communication efficiency. However, model training on vehicles requires a large amount of computation resource especially that the machine learning (ML) models are massive nowadays. In this paper, we shed light on the situation that a part of the vehicles may have low computational capability and not able to participate in the training process. To address this problem, we propose a Vehicular Edge Hybrid Federated Split Learning (VEHFSL) paradigm considering the data privacy protection and resource constraints in Internet of Vehicles (IoV). Firstly, we select the vehicles to participate in the training process by sorting the weighted sum of the expected sojourn time and training data volume of each vehicle. Then, we formulate a joint optimization problem to minimize the training latency of one round by jointly optimizing the training mode selection, computation frequency and transmit power allocation. Given that the formulated optimization problem is mixed-integer nonlinear programming (MINLP), which is typically NP-hard, we design a joint mode selection and resource allocation algorithm (JMSRA) by performing block coordinate descent (BCD) technique to decompose it into three sub-problems and iteratively solving them to achieve a near-optimal solution efficiently with the balance of the computational complexity and optimality of the solution. Finally, extensive simulations are performed to validate the superiority of our proposed VEHFSL paradigm in resource-constrained vehicular networks.

In this paper, we study a cooperative sensing network with integrated sensing and communication (ISAC) unmanned aerial vehicles (UAVs) for handling diverse sensing tasks with different types, accuracy requirements, and sensing intervals. Unlike existing works that overlook task differences in real-world applications, we aim to enable UAVs to adjust their service time slots for each task and the number of UAVs assigned to the task, preventing excessive resources to tasks with low demand. To this end, we maximize the task completion rate while minimizing cost by optimizing UAV trajectories, task associations, and power control. Given the problem’s mixed-integer nature and the partial observability of UAVs, we design a multi-agent proximal policy optimization with action branching and masking (MAPPO-ABAM) algorithm to tackle it. Simulation results show that our algorithm achieves the highest completion rate and the lowest sensing cost compared to four benchmark methods.