Integrated sensing, communication, and computation (ISCC) provides a critical enabling platform in supporting the diverse services in the Internet of Vehicles (IoV). However, effective heterogeneous IoV service provisioning relies on both communication-centric and beyond-communication performance metrics, making unified resource allocation challenging. Moreover, competition from concurrent services for limited multi-dimensional resources is intensified in dynamic vehicular environments. In this paper, we investigate the resource allocation problem for concurrent communication and target classification services in an ISCC-enabled IoV system. To solve the problem, we first introduce the value of service (VoS) to unify communication rate and classification accuracy into a common measure that captures the degree of heterogeneous service fulfillment. To reduce the complexity of dynamic problem optimization, we propose a digital twin-assisted proximal policy optimization (DTPPO) algorithm, in which the digital twin exploits both current and historical information to generate predictive information, thereby enhancing policy learning in dynamic environments. Furthermore, we develop a large language model (LLM)-enhanced DTPPO (LLM-DTPPO) algorithm, which leverages the contextual understanding and domain knowledge of LLMs to reshape the reward function and improve resource allocation performance under multi-dimensional resource competition. Simulation results based on real-world vehicle mobility traces demonstrate that the proposed algorithms outperform existing benchmark schemes.
Bangzhen Huang, Zhang Liu, Lianfen Huang et al.· IEEE Transactions on Network...· 0 citations
The emergence of sixth-generation (6G) communication systems promotes an integrated sensing and communications (ISAC) framework to address growing demands for seamless connectivity. This paper proposes a cooperative resource allocation method (CRAM) that enables uninterrupted service for mobile terminals (MTs) moving across heterogeneous base stations (BSs) without resource handovers—inspired by a spotlight tracking a moving actor. Specifically, we tackle the challenge of dynamically managing resources in macro–micro heterogeneous ISAC networks by exploiting real-time MT speed and location information. CRAM achieves a spotlight effect for MTs through two core strategies: a resource pre-allocation mechanism driven by network topology sensing, and a resource sharing strategy that mitigates ping-pong effects in micro-BS overlapping regions. For high-velocity MTs, spectrum resources are assigned at macro-BSs to enhance service quality and minimize handovers. Furthermore, we incorporate the Cramer–Rao lower bound to derive optimized resource allocation policies adapted to diverse MT speeds and locations. Simulation results demonstrate that CRAM tailors resource distribution to MT characteristics, guaranteeing zero-interruption connectivity while maximizing system performance. In comparison with existing benchmarks, CRAM improves system throughput by 24.5% and 20.9%, reduces the total number of handovers by 29.1% and 16.5%, and increases the average signal-to-interference-plus-noise ratio (SINR) for all MTs by 41.3% and 30.3%. These outcomes highlight CRAM’s potential to redefine 6G network management by substantially boosting network efficiency and user experience.
Sai Zou, Meifang Wang, Minghui Liwang et al.· IEEE Transactions on Cogniti...· 0 citations
Future 6G networks are envisaged to tightly integrate communication, sensing, and computing, demanding real-time, intent-driven intelligence at the edge. While large language models (LLMs) excel in intent recognition and semantic reasoning, their application to real-time network lifecycle management at the edge is limited by heterogeneous application intents (APPIs), dynamic network conditions, and severe resource constraints. This paper proposes a novel lightweight LLM architecture, KGLlama-KD, that synergizes knowledge graphs (KGs) with knowledge distillation (KD) to enable intent-driven networking and enhance 6G edge intelligence. Specifically, a KG is constructed to formally describe the relationships among application scenarios, functional primitives, performance requirements within APPIs, and the correspondences between APPIs and network service requests (NSRs), thereby producing a structured intent training dataset. Building upon the Llama 3 foundation model, a two-phase optimization framework is designed to support lightweight edge deployment while preserving translation fidelity. The LLM is first fine-tuned with KG guidance and compressed via KD in the cloud, and then deployed on resource-constrained edge nodes to perform real-time, accurate, and efficient APPIs interpretation. Experiments validate that KGLlama-KD achieves 95% accuracy for APPI understanding, surpassing DeepSeek and Qwen by an average of 8%. The distilled model reduces inference latency by 60% compared to full-scale LLMs, fulfilling the sub-100 ms requirement for 6G latency-sensitive services.
Bing Wu, Sai Zou, Minghui Liwang et al.· IEEE Transactions on Mobile...· 3 citations