Semantic-Aware Edge Intelligence for UAV Handover in 6G Networks

Aubida A. Al-Hameed 2 , Mohammed M. H. Qazzaz 1 2 , Maryam Hafeez 1 , and Syed A. Zaidi 1 · 2025

Read Semantic-Aware Edge Intelligence for UAV Handover in 6G Networks by Aubida A. Al-Hameed 2 , Mohammed M. H. Qazzaz 1 2 , Maryam Hafeez 1 , and Syed A. Zaidi 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-22668.txt.

Introduction

Aubida A. Al-Hameed and 3 co-authors hand drone handovers in future 6G networks to a small language model on board. Semantic communication sends what matters for the goal, which can cut energy and latency. Their framework deploys a lightweight MobileBERT, fine-tuned with low-rank adapters, on the drone itself. The model reads flight and radio measurements with many attributes and classifies the right handover action.

Alongside the decision, the model outputs reason tags that explain its rationale in context. On a rule-based synthetic data set of handover scenarios, it learned the rules with high accuracy. Its reason tags also scored strongly. The design aims to keep drone links reliable while making each switch auditable.

Catalog record

CREATORAubida A. Al-Hameed 2 , Mohammed M. H. Qazzaz 1 2 , Maryam Hafeez 1 , and Syed A. Zaidi 1
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-22668

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Semantic-Aware Edge Intelligence for UAV Handover in 6G Networks Aubida A. Al-Hameed 2 , Mohammed M. H. Qazzaz 1 2 , Maryam Hafeez 1 , and Syed A. Zaidi 1 Abstract 6G wireless networks aim to exploit semantic awareness to optimize radio resources. By optimizing the transmission through the lens of the desired goal, the energy consumption of transmissions can also be reduced, and the latency can be improved. To that end, this paper investigates a paradigm in which the capabilities of generative AI (GenAI) on the edge are harnessed for network optimization. In particular, we investigate an Unmanned Aerial Vehicle (UAV) handover framework that takes advantage of GenAI and semantic communication to maintain reliable connectivity. To that end, we propose a framework in which a lightweight MobileBERT language model, fine-tuned using Low-Rank Adaptation (LoRA), is deployed on the UAV. This model processes multi-attribute flight and radio measurements and performs multi-label classification to determine appropriate handover action. Concurrently, the model identifies an appropriate set of contextual ”Reason Tags” that elucidate the decision’s rationale. Our model, evaluated on a rule-based synthetic dataset of UAV handover scenarios, demonstrates the model’s high efficacy in learning these rules, achieving high accuracy in predicting the primary handover decision. The model also shows strong performance in identifying supporting reasons, with an F1 micro-score of approximately 0.9 for reason tags. I Introduction Recently, the expansion of Unmanned Aerial Vehicles (UAVs) in various applications has seen an unprecedented rise from surveillance to delivery, communication relaying, and intelligent network connectivity [ 1 ] . During a UAV mission, maintaining seamless connectivity is essential through handover mechanisms. Traditional handover decisions depend on the exchange of data measurements (such as the Received Power Reference Signal (RSRP), the Channel Quality Indicator (CQI)) between the UAV and the base station (BS). This results in considerable signaling overhead and potential delays [ 2 ] . The advent of 6G networks heralds a shift towards semantic communication [ 3 , 4 , 5 ] . In contrast to traditional communication systems that prioritize bit-level precision, semantic communication focuses on conveying the intended message (goal-related information), leading to more efficient and effective data transfer [ 6 ] . This approach is especially beneficial for resource-limited devices, such as UAVs, and in dynamic settings where understanding context is crucial. Edge intelligence facilitates processing near the data source, allowing localized decision making and decreasing dependence on centralized cloud processing [ 7 , 8 ] . Figure 1 illustrates the proposed conceptual process of a semantic communication system for UAV handovers. Here, the UAV generates a succinct semantic message from the local data it processes. In this paper, we propose an edge-intelligent UAV handover system that embodies principles of semantic communication within a controlled environment. A lightweight language model, MobileBERT, fine-tuned using Low-Rank Adaptation (LoRA), is deployed on the UAV [ 9 ] [ 10 ] . This model processes a textual representation of the complex radio and flight measurements to predict a primary handover decision from four distinct actions. Then identify a set of contextual ”Reason Tags” that justify the decision based on predefined rules. After generating a concise semantic message for communication with the BS in service, the assessment and key rationale are provided. Our contributions include: (i) a multi-label classification framework for joint handover decision and reason tag prediction at the UAV edge (evaluated on a synthetic dataset); (ii) a demonstration of MobileBERT with LoRA for efficient fine-tuning on this task using synthetic data; and (iii) a system that generates semantic messages, discussing their potential for reducing communication overhead. Experimental results on a generated dataset of 5000 scenarios show near-perfect accuracy for the primary handover decision and strong performance in identifying relevant contextual reasons, highlighting the model’s ability to learn the underlying data generation rules. This acts as an introductory validation, emphasizing the importance of further studies on applying this to real-world, non-rule-based contexts. The paper is structured as follows: Section II reviews related research. Section III details the design and approach of the system. Section IV describes the experimental setup, with results and analysis in Section V . Section VI concludes and suggests future work. Appendix A offers examples of generated semantic messages. Fig. 1: Conceptual overview of the proposed semantic communication framework for UAV handover. II Related Work The semantic communication paradigm in the context of Shannon’s information theory. This paradigm has gained attention with the emergence of 6G networks [ 3 ] . This paradigm focuses on the transmission of meaningful information rather than the precise delivery of bits. Recent research explores the use of deep learning techniques for encoding and decoding semantic messages, highlighting significant compression efficiencies and robustness against channel noise by transmitting only the data relevant to the task [ 6 , 11 ] . Using Large Language Models (LLM) in semantic communication, especially within edge-based IoT networks, is an emerging investigative strand [ 5 ] . Our research aligns with this by concentrating on extracting and communicating a predefined ”semantic essence” (decisions and justifications) from UAV handover assessments in a synthetic context. The management of UAV handover involves unique challenges due to aspects such as high mobility, three-dimensional motion, and strict Quality of Service (QoS) needs [ 1 ] . Traditional handover techniques based on fixed signal strength levels often fail. Hence, intelligent handover strategies powered by machine learning (ML) and artificial intelligence (AI) have been proposed [ 12 ] . These include methods such as Q-learning, fuzzy logic, and neural network-based techniques to forecast handover requirements and select target cells [ 13 ] [ 14 ] . However, many current intelligent solutions still depend on sending vast amounts of raw data to centralized controllers or BSs for decision making. Enhancing network intelligence, particularly on the UAV itself, is vital for efficient and autonomous UAV functions [ 7 , 15 ] . Edge AI can reduce latency, preserve backhaul capacity, and increase privacy. Onboard processing for UAV tasks such as navigation, object recognition, and network management is becoming more viable as lightweight deep learning models advance [ 16 ] . The convergence of large AI models with aerial edge devices, often involving edge-cloud collaboration, is a growing research domain that addresses the computational demands of these models [ 8 ] . Our research leverages the deployment of a fine-tuned MobileBERT model on UAVs to facilitate local handover assessments. Large Language Models (LLMs) and their scaled versions possess notable capabilities in comprehending contextual nuances and executing complex reasoning, even when confronted with text-based information [ 17 ] . The integration of LLMs with UAVs represents a rapidly evolving area of study, aimed at exploring their ability to improve autonomy, perception, planning, and interaction between humans and UAVs [ 18 , 19 ] . Fine-tuning models like BERT and MobileBERT for specific tasks is common [ 9 ] . Effortless fine-tuning of LoRA parameters (PEFT) reduces trainable parameters, improving LLM for edge deployment [ 10 ] . Our method uses MobileBERT with LoRA to decode UAV scenario texts for multi-label classification, aligning sensor data with semantic labels through learned patterns. III Proposed System Model and Methodology Our proposed system empowers the UAV with edge intelligence to assess handover scenarios, make a

Read the complete text