Kezhi Wang

Accurate and interpretable ship-to-shore end-to-end (E2E) communication latency prediction is important for safe and efficient inland waterway transportation. However, existing methods suffer from data sparsity, heterogeneity, and limited interpretability. Large language models (LLMs) show strong reasoning capabilities, but they usually perform poorly in numerical latency prediction. In this work, we propose a unified LLM-based framework that integrates high-dimensional environmental data representation, physically consistent data augmentation via a conditional diffusion model (CDM), and a tabular to question-answer transformation for low-rank adaptation (LoRA)-based LLM training. This unified framework achieves enhanced latency prediction performance while providing explicit environmental attribution and faithful natural-language explanations. Experimental results demonstrate that the proposed framework consistently outperforms representative baselines across multiple metrics, reducing the mean absolute error by approximately 42% while providing reliable and interpretable predictions.

Six-dimensional movable antenna (6DMA) is a promising solution to improve the performance of future wireless networks by flexibly adjusting three-dimensional (3D) positions and rotations of antenna surfaces. Previous studies have verified the potential of 6DMA in communication/sensing, failing to explore its integration into integrated sensing and communication (ISAC) systems. In this letter, we propose a fairness-aware 6DMA-enabled ISAC system for non-uniform user distribution. Specifically, we introduce a functionally reconfigurable 6DMA-based base station (BS) architecture, in which the mode selection, 3D positions, and rotations of all 6DMA surfaces, as well as the receive beamforming matrices at the BS are jointly optimized under a max–min fairness principle to prioritize the performance of the most vulnerable user. We aim to maximize the minimum signal-to-interference-plus-noise ratio (SINR) across all ISAC users. To tackle the non-convex optimization problem, we propose an alternating optimization (AO) framework that decomposes the non-convex optimization problem into three subproblems and alternately solves them in an iterative manner. Finally, simulation results demonstrate the superiority of the proposed scheme, which achieved the improvement of system performance while ensuring the fairness of communication and sensing.

This study proposes a novel Agentic Retrieval-augmented generation with Mamba-Attention Integrated Transformer (ARMAIT) framework for multi-Unmanned Aerial Vehicle (UAV) trajectory optimization. The framework is built upon Large Language Models (LLMs), incorporating Retrieval-Augmented Generation (RAG) empowered by Agentic AI and integrated with a UAV-specific knowledge base. Through the Agentic RAG, the LLM autonomously interprets high-level task requirements and identifies the key components necessary for trajectory optimization, including model inputs and outputs, network architecture, reward functions, and task constraints. To support efficient modeling across different system scales, we introduce the Mamba-Attention Integrated Transformer (MAIT), a hybrid neural architecture that combines the long-range dependency modeling capability of attention mechanisms with the efficient temporal dynamic representation of Mamba. Furthermore, a Trajectory-Group Relative Policy Optimization (T-GRPO) method is proposed to achieve unified policy gradient optimization in both discrete and continuous trajectory spaces for MAIT training. Extensive experimental results validate the feasibility and effectiveness of the proposed ARMAIT framework.

In this paper, we propose an optimisation framework for deployment of multiple reconfigurable intelligent surfaces (RISs) to meet the wireless coverage demands for smart warehouses. Specifically, we are the first to formulate a unified network optimisation task that jointly considers RIS placement and beamforming to maximize overall network coverage with a deterministic channel model to accurately describe the multipath effect for the warehouse. To address this problem, we design a hybrid optimisation framework composed of three synergistic modules. (1) A Large Language Model (LLM) acts as a semantic planner that generates physically feasible multi-RIS configurations, jointly determining the placement and beamforming directions guided by structured prompts and environment-aware embeddings. (2) A Genetic Algorithm (GA) module performs local numerical refinements to enhance the precision of LLMgenerated solutions under physical constraints. (3) A Diversity Reflection and Correction (DiRect) module evaluates structural similarity among candidate configurations and triggers additional semantic regeneration to maintain exploration diversity. These three modules form an alternating iterative process in which LLM reasoning, GA-based evolution, and DiRect-driven regeneration collectively guide the optimisation toward high-coverage configurations. Extensive simulations validate the effectiveness and robustness of the proposed framework. Compared with traditional heuristics, reinforcement learning methods, and LLMguided baselines, our hybrid framework achieves 10%-15% higher coverage within 10-20 iterations. The performance consistently scales with the number of RISs and element sizes, and remains stable under varying transmitter positions, demonstrating strong adaptability to complex smart warehouse layouts. Overall, the proposed hybrid optimisation framework provides a scalable and physically grounded solution for RIS-assisted network deployment optimisation in realistic industrial environments.

In this paper, we address the problem of fair access and Age of Information (AoI) optimization in 5G New Radio (NR) Vehicle to Everything (V2X) Mode 2. Specifically, vehicles need to exchange information with the road side unit (RSU). However, due to the varying vehicle speeds leading to different communication durations, the amount of data exchanged between different vehicles and the RSU may vary. This may poses significant safety risks in high-speed environments. To address this, we define a fairness index through tuning the selection window of different vehicles and consider the image semantic communication system to reduce latency. However, adjusting the selection window may affect the communication time, thereby impacting the AoI. Moreover, considering the re-evaluation mechanism in 5G NR, which helps reduce resource collisions, it may lead to an increase in AoI. We analyze the AoI using Stochastic Hybrid System (SHS) and construct a multi-objective optimization problem to achieve fair access and AoI optimization. Sequential Convex Approximation (SCA) is employed to transform the non-convex problem into a convex one, and solve it using convex optimization. We also provide a large language model (LLM) based algorithm. The scheme's effectiveness is validated through numerical simulations.

In current molecular communication (MC) systems, performing computational operations at the nanoscale remains challenging, restricting their applicability in complex scenarios such as adaptive biochemical control and advanced nanoscale sensing. To overcome this challenge, this paper proposes a novel framework that seamlessly integrates computation into the molecular communication process. The system enables arithmetic operations, namely addition, subtraction, multiplication, and division, by encoding numerical values into two types of molecules emitted by each transmitter to represent positive and negative values, respectively. Specifically, addition is achieved by transmitting non-reactive molecules, while subtraction employs reactive molecules that interact during propagation. The receiver demodulates molecular counts to directly compute the desired results. Theoretical analysis for an upper bound on the bit error rate (BER), and computational simulations confirm the system’s robustness in performing complex arithmetic tasks. Compared to conventional MC methods, the proposed approach not only enables fundamental computational operations at the nanoscale but also lays the groundwork for intelligent, autonomous molecular networks.