Research Topics

Fundamentals of Integrated Sensing and Communication (ISAC) over UAV-Enabled Networks

Abstract: Integrated Sensing and Communication (ISAC) has emerged as a key enabler for next-generation wireless systems, particularly in unmanned aerial vehicle (UAV)-enabled networks where mobility, flexibility, and line-of-sight dominance create unique opportunities for joint functionality. This paper investigates an ISAC framework for UAV-enabled networks in which aerial platforms simultaneously provide wireless connectivity and high-resolution sensing services. We develop a unified system model that captures the coupling between communication throughput, sensing accuracy, UAV trajectory design, and power allocation. To efficiently balance these competing objectives, we formulate a joint optimization problem that maximizes aggregate spectral efficiency while guaranteeing sensing performance and mobility constraints. A distributed learning-based approach is proposed to adaptively optimize beamforming, resource allocation, and UAV trajectories under dynamic channel and target conditions. Simulation results over realistic air-to-ground channels demonstrate that the proposed ISAC-enabled UAV network significantly improves communication capacity while achieving high-precision target localization, highlighting the strong synergy between sensing and communication integration. The findings provide practical design insights for future 6G aerial networks supporting intelligent transportation, emergency response, and smart city applications.

Deep Reinforcement Learning for Wireless Networking and Security

Abstract: The increasing complexity, heterogeneity, and dynamic nature of modern wireless networks pose significant challenges to conventional rule-based optimization and security mechanisms. Deep Reinforcement Learning (DRL) has emerged as a powerful data-driven framework capable of enabling autonomous, adaptive, and scalable network control. This paper investigates the application of DRL to wireless networking and security, focusing on joint resource management, interference mitigation, and proactive threat defense in dynamic environments. We develop a unified learning framework in which network agents interact with time-varying channels and adversarial entities to optimize long-term system performance while maintaining robust security guarantees. The proposed approach integrates spectrum allocation, power control, and user scheduling with intrusion detection and anti-jamming strategies under partial observability and stochastic conditions. By leveraging deep neural function approximation and policy optimization techniques, the framework enables distributed decision-making with limited signaling overhead. Simulation results demonstrate that the DRL-based scheme significantly improves spectral efficiency, latency, and resilience against attacks compared to conventional optimization and static security baselines. The results highlight the potential of DRL to support intelligent, self-healing, and secure wireless networks for beyond-5G and 6G systems.

Learning What Matters: Semantic Communication via

Deep Contextual Multi-Agent Bandits

Abstract: Conventional wireless systems are designed to reliably transmit bits, regardless of their contextual importance. However, emerging intelligent applications require the efficient delivery of task-relevant information rather than raw data. This paper proposes a semantic communication framework that learns what matters by leveraging deep contextual multi-agent bandits. In the proposed architecture, distributed agents observe heterogeneous environments and adaptively select semantic representations that maximize task performance under communication constraints. We model semantic encoding and transmission as a contextual bandit problem, where each agent learns to prioritize informative features based on environmental context, channel conditions, and application objectives. A deep neural representation module captures high-dimensional semantics, while a multi-agent coordination mechanism balances exploration and exploitation across distributed nodes with limited feedback. The resulting framework jointly optimizes semantic compression, transmission scheduling, and resource allocation to enhance task-oriented efficiency. Simulation results demonstrate substantial reductions in communication overhead while preserving or improving task accuracy compared to conventional data-centric and fixed-compression schemes. The findings highlight the potential of contextual bandit learning to enable scalable, adaptive, and goal-driven semantic communication for next-generation wireless networks.

Page updated

Google Sites

Report abuse