6G's AI native vision of embedding advance intelligence in the network while bringing it closer to the user requires a systematic evaluation of Generative AI (GenAI) models on edge devices. Rapidly emerging solutions based on Open RAN (ORAN) and Network-in-aBox strongly advocate the use of low-cost, off-the-shelf components for simpler and efficient deployment, for example, in provisioning rural connectivity. In this context, conceptual architecture, hardware testbeds, and precise performance quantification of Large Language Models (LLMs) on off-theshelf edge devices remain largely unexplored. This research investigates computationally demanding LLM inference on a single commodity Raspberry Pi serving as an edge testbed for ORAN. We investigate various LLMs, including small, medium, and large models, on a Raspberry Pi 5 Cluster using a lightweight Kubernetes distribution (K3s) with modular prompting implementation. We study its feasibility and limitations by analyzing throughput, latency, accuracy, and efficiency. Our findings indicate that CPU-only deployment of lightweight models, such as Yi, Phi, and Llama3, can effectively support edge applications, achieving a generation throughput of 5 to 12 tokens per second with less than 50% CPU and RAM usage. We conclude that GenAI on the edge offers localized inference in remote or bandwidthconstrained environments in 6 G networks without reliance on cloud infrastructure.

Generative AI (GenAI) is expected to play a pivotal role in enabling autonomous optimization in future wireless networks. Within the ORAN architecture, Large Language Models (LLMs) can be specialized to generate xApps and rApps by leveraging specifications and API definitions from the RAN Intelligent Controller (RIC) platform. However, fine-tuning base LLMs for telecom-specific tasks remains expensive and resource-intensive. Retrieval-Augmented Generation (RAG) offers a practical alternative through in-context learning, enabling domain adaptation without full retraining. While traditional RAG systems rely on vector-based retrieval, emerging variants such as GraphRAG and Hybrid GraphRAG incorporate knowledge graphs or dual retrieval strategies to support multi-hop reasoning and improve factual grounding. Despite their promise, these methods lack systematic, metric-driven evaluations, particularly in high-stakes domains such as ORAN. In this study, we conduct a comparative evaluation of Vector RAG, GraphRAG, and Hybrid GraphRAG using ORAN specifications. We assess performance across varying question complexities using established generation metrics: faithfulness, answer relevance, context relevance, and factual correctness. Results show that both GraphRAG and Hybrid GraphRAG outperform traditional RAG. Hybrid GraphRAG improves factual correctness by 8%, while GraphRAG improves context relevance by 11%.

The deployment of AI agents within legacy Radio Access Network (RAN) infrastructure poses significant safety and reliability challenges for future 6G networks. This paper presents a novel Edge AI framework for autonomous network optimisation in Open RAN environments, addressing these challenges through three core innovations: (1) a persona-based multi-tools architecture enabling distributed, context-aware decision-making; (2) proactive anomaly detection agent powered by traffic predictive tool; and (3) a safety, aligned reward mechanism that balances performance with operational stability.Integrated into the RAN Intelligent Controller (RIC), our framework leverages multimodal data fusion, including network KPIs, a traffic prediction model, and external information sources, to anticipate and respond to dynamic network conditions. Extensive evaluation using realistic 5G scenarios demonstrates that the edge framework achieves zero network outages under high-stress conditions, compared to 8.4% for traditional fixed-power networks and 3.3% for large language model (LLM) agent-based approaches, while maintaining near real-time responsiveness and consistent QoS. These results establish that, when equipped with the right tools and contextual awareness, AI agents can be safely and effectively deployed in critical network infrastructure, laying the framework for intelligent and autonomous 5G and beyond network operations.

Federated learning (FL) has recently emerged as a tool for efficient distributed learning. In edge-assisted internet of things (IoT) paradigm, FL can achieve a promising speed of convergence in the learning process while preserving privacy. To overcome the inherent wireless network connectivity challenge in FL based IoT networks, unmanned aerial vehicles (UAVs) can play a vital role in offloading the learned models to a central learning server or play as a flying learning server. In this paper, we investigate the physical layer aspects of harnessing UAVs as flying FL server for the distributed model averaging. In particular, we study the effect of the UAV-to-ground communication model in the convergence of the model accuracy and loss curves using well-known benchmark datasets to illustrate the effects. We illustrate the effect of altering the deployment geometry of IoT clients simultaneously with the trajectory design of the UAVs.

Federated Learning (FL) enables distributed model training on edge devices while preserving data privacy. However, FL deployments in wireless networks face significant challenges, including communication overhead, unreliable connectivity, and high energy consumption, particularly in dynamic environments. This paper proposes EcoFL, an integrated FL framework that leverages the Open Radio Access Network (ORAN) architecture with multiple Radio Access Technologies (RATs) to enhance communication efficiency and ensure robust FL operations. EcoFL implements a two-stage optimisation approach: an RLbased rApp for dynamic RAT selection that balances energy efficiency with network performance, and a CNN-based xApp for near real-time resource allocation with adaptive policies. This coordinated approach significantly enhances communication resilience under fluctuating network conditions. Experimental results demonstrate competitive FL model performance with 19 % lower power consumption compared to baseline approaches, highlighting substantial potential for scalable, energy-efficient collaborative learning applications.

Autonomous vehicles (AVs) require real-time, low-latency, and high-throughput communication to ensure safe and efficient operation. Fifth-generation (5G) and beyond communication networks, particularly New Radio (NR) operating at millimeter wave (mmWave) frequencies, offer promising solutions. However, higher frequencies introduce challenges like signal attenuation, blockage, and complexities of beam management. This paper provides a comprehensive survey of 5G and beyond communication networks, emphasizing their role in vehicle-to-everything (V2X) communication. It uniquely addresses the technical challenges of beam training and management, highlighting the application of artificial intelligence (AI) and machine learning (ML) techniques to optimize beam selection and reduce delays. Furthermore, the paper explores future directions, focusing on AI/ML-driven solutions to enhance beamforming and communication reliability in dynamic AV environments.

The ever-increasing penetration of Internet of Things (IoT) applications across all sectors require better information and communication security for IoT devices. Physical Unclonable Function (PUF) circuits are a low-cost method used for generating unique responses, ideal for key generation and device authentication in high-performance microprocessors. PUFs are extracted from manufacturing variations embedded in the hardware of accessible devices, thereby requiring no additional computation or modification. Static Random Access Memory (SRAM) PUFs are widely used with keys generated from power-up values for authentication. This work is based on improving device authentication in the context of LoRa, which is a low-power wide-area network (LPWAN) technology. This paper exploits the integration of Carrier Frequency Offsets (CFOs) and SRAM PUF to create a two-step authentication security solution for LoRaWAN called LoRa-PUF. The power-up state value of three SRAM chips from the same off-the-shelf manufacturer are analyzed for SRAM PUF properties and 36000 packets of CFOs of four LoRa device types have been analyzed for LoRa-PUF. Our results indicate that LoRa SRAMs serve as sources of reliable challenge-response pairs for PUFs and the CFOs of the LoRa device type can be classified during communication with more than 70% accuracy which can be implemented on resource-constrained LoRa microcontrollers.

In this article, we develop a comprehensive framework to characterize the performance of drone assisted Backscatter communication based Internet of things (IoT) sensor network. We consider a scenario such where drone transmits RF carrier which is modulated by IoT sensor node (SN) to transmit its data. The SN implements load modulation which results in amplitude shift keying (ASK) type modulation for the impinging RF carrier. In order to quantify the performance of the considered network, we characterize the coverage probability for the ground based SN node. The statistical framework developed to quantify the coverage probability explicitly accommodates dyadic backscatter channel which experiences deeper fades than that of the one-way Rayleigh channel. Our model also incorporates Line of Sight (LoS) and Non-LoS (NLoS) propagation states for accurately modeling large-scale path-loss between drone and SN. We consider spatially distributed SNs which can be modelled using spatial Binomial Point process (BPP). We practically implement the proposed system using Software Defined Radio (SDR) and a custom designed SN tag. The measurements of parameters such as noise figure, tag reflection coefficient etc., are used to parametrize the developed framework. Lastly, we demonstrate that there exists an optimal set of parameters which maximizes the coverage probability for the SN.

In this paper, we tackle the design issue of optimal deployment of low power wide area network (LPWAN) Internet of Things (IoT) gateways (GWs). We classify GW deployment problem into two different categories, i.e., network-aware and network-agnostic. In network-aware GW deployment, precise location of IoT end devices (EDs) is known and thus the design questions are: 1) where to place GWs, i.e., to maximize received signal strength and 2) given received signal strength which GW should the ED be associated with to balance the network load. For, Network-agnostic GW deployment, same questions are answered in the absence of precise knowledge for the locations of EDs. For the network-aware deployment we borrow tools from machine-learning such as K-means clustering for determination of optimal GW location. Subsequently, the link assignment problem is presented as an integer linear programming optimization. We prove that the network-agnostic GW deployment principle of placement of GWs at highest altitudes, if applied automatically, may lead to very deteriorated network performance increasing the network operational costs. Consequently, we introduce the concept of network-agnostic GW placement algorithm whereby the location of GWs can be estimated without prior knowledge of specific locations of EDs and we use it as a guiding principle to design spatial algorithm for finding GW locations. We show that spatial algorithm can, in principle, provide effective GW placement suggestions compared to a network-aware method such as K-means clustering. We show that using a computational method for GW placement like K-means or spatial algorithm, has a potential of creating competitive network performance using just the same number of GWs, thus cutting down the financial costs of the network and increasing its sustainability.

IEEE 802.11ah has recently emerged as a promising standard for enabling massive machine-to-machine (M2M) communication. In order to support uplink data transmission from dense machine type clients (such as smart meters, IoT end nodes etc.), 802.11ah relies upon the restricted access window (RAW) based Medium Access Control (MAC) protocol. The underlying motivation behind this protocol is to reduce the contention for spectrum access among a large number of devices. The nodes contend with each other in their assigned RAW slot using Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA). In each RAW slot, the throughput depends upon the number of nodes. Current studies have suggested that the duration of each RAW slot should be the same in the entire RAW frame. However in this paper, we argue that the duration of each RAW slot should be chosen according to the size of the group. We present a model where a RAW frame is divided into two sub-frames and the duration of RAW slots in each sub-frame is chosen according to the size of the group. With the help of an analytical framework, we demonstrate that the throughput under our proposed scheme can be significantly enhanced when compared to a conventional implementation.

The Internet of Things (IoT) has been considered as an emerging research area where the IPv6 over low-power wireless personal area network (6LoWPAN) protocol stack is considered as one of the most important protocol suite for the IoT. Recently, the Internet Engineering Task Force has developed a set of IPv6-based protocols to alleviate the challenges of connecting resource limited sensor nodes to the Internet. In 6LoWPAN networks, heavy network traffic causes congestion which significantly degrades network performance and effects the quality of service aspects, e.g., throughput, end-to-end delay and energy consumption. In this paper, we formulate the congestion problem as a noncooperative game framework where the nodes (players) behave uncooperatively and demand high data rate in a selfish way. Then, the existence and uniqueness of Nash equilibrium is proved and the optimal game solution is computed by using Lagrange multipliers and Karush-Kuhn-Tucker conditions. Based on this framework, we propose a novel and simple congestion control mechanism called game theory-based congestion control framework (GTCCF) specially tailored for IEEE 802.15.4, 6LoWPAN networks. GTCCF is aware of node priorities and application priorities to support the IoT application requirements. The proposed framework has been tested and evaluated through two different scenarios by using Contiki OS and compared with comparative algorithms. Simulation results show that GTCCF improves performance in the presence of congestion by an overall average of 30.45%, 39.77%, 26.37%, 91.37%, and 13.42% in terms of throughput, end-to-end delay, energy consumption, number of lost packets, and weighted fairness index (WFI), respectively, as compared to duty cycle-aware congestion control for 6LoWPAN network algorithm.

In this paper, we propose a beamforming-based dynamic spectrum leasing (DSL) technique to improve the spectral utility of bi-directional communication of the legacy/primary spectrum users through the help of colocated secondary users. The secondary users help for a time interval to relay the data between two primary terminals using physical layer network coding and beamforming to attain bi-directional communication with high spectral utility. As a reimbursement, the secondary users, cognitive radios (CRs) in our case, get exclusive access to the primary spectrum for a certain duration. We use Nash bargaining to determine the optimal division of temporal resources between relaying and reimbursement. Moreover, we consider that a fraction of secondary nodes can act selfishly by not helping the primary, yet enjoy the reimbursement time. We measure the utility of the DSL scheme in terms of a metric called time-bandwidth product (TBP) ratio quantifying the number of bits transmitted in direct communication versus DSL. We show that if all secondary nodes act honestly, more than 17-fold increase in the TBP ratio is observed for a sparse CR network. However, in such a network, selfish behavior of CR nodes can reduce the gain by more than a factor of 2.

In this paper, we present a novel licensed-shared access (LSA) spectrum-sharing scheme as an energy efficient way to increase the spectral utility of a network. We consider a two-tier network where a small cell network offers offloading services to a macro network to improve its quality of service. In return, the small cells are rewarded for their cooperation with a number of licenses to operate in the spectrum owned by the macro network. The short link distances and diversity offered by the small cell network to offload the macro user help enhance the total spectral and energy utilization of the network. For effective functioning of the scheme, it is important to determine the division between the fraction of small cells that provide offloading services and those which obtain the licenses. In this paper, we present a comprehensive mathematical model of this LSA-based offloading and evaluate the optimal division between the offloading and licensing small cells. Our analysis shows that such spectrum sharing can lead to commendable gains in the spectral utilization of the network while assuring the desired quality of service for both macro and small network. An intelligent determination of the fraction of offloading small cell network results in efficient energy use for offloading. We test our scheme within the network parameters of the city of Leeds, U.K. Our results indicate that the spectral utility of Leeds can be improved by more than 12×, and the energy efficiency by 16× through our proposed scheme.

In this paper, we propose dynamic spectrum leasing (DSL) for cognitive radio (CR) networks as an efficient alternative to two way direct communication for legacy/primary networks. We show that using physical layer network coding (PNC) and distributed beamforming, the CR nodes can relay the primary network data for some duration while the spectrum is leased to them. We demonstrate that in DSL operational mode, up to 20 times more data can be successfully exchanged between two primary sources as compared to direct two way communication. Also in DSL, as a reimbursement, the otherwise silent CR nodes that help in relaying can get a fraction of time during which they can carry out their own communication thereby increasing the spectral utilization.

In this paper, we aim to present a cooperative relaying based two way wireless communication scheme which can provide both spectral and energy efficiency in future wireless networks. To this end, we propose a novel network coding based Dynamic Spectrum Leasing (DSL) technique in which the cognitive secondary users cooperatively relay the primary data for two-way primary communication. In exchange for the relaying services, the primary grants exclusive access to the secondary users for their own activity. We model the random geometry of the ad hoc secondary users using a Poisson point process. We devise a game theoretic framework for the division of leasing time between the primary cooperation and secondary activity phases. We demonstrate that under these considerations and employing network coding, DSL can improve the number of bits that are successfully transmitted by 54% as compared to un-coded direct two way primary communication. Also the energy costs of the proposed DSL scheme are more than 10 times lower. Employing DSL also enables the cognitive users to get reasonable time for their own transmission after increasing the primary spectral and energy efficiency.

In this paper, we study the impact of spatio-temporal variations of a video sequence on wireless MIMO video transmission. In the light of our findings, we propose a cross-layer MIMO video communication architecture for efficient multimedia communication. Existing methods complement Spatial Multiplexing (SM) and Space Time Block Coding (STBC) with enhanced source and channel coding strategies to improve video PSNR at a MIMO receiver. However, these methods ignore the adverse impacts of spatio-temporal content variations on video quality and decoding complexity. We show that the decision to switch between SM and STBC for effective video communication must be based on both channel conditions and motion intensity of a video. We quantify a video's spatio-temporal characteristics using a motion intensity measure, and propose a simple motion-adaptive MIMO video communication technique which can judiciously switch between SM and STBC configurations. Experimental evaluations show that the proposed technique provides significantly better video quality while having very low decoding complexity at the receiver. Based on this technique, we propose a novel cross-layer motion-adaptive architecture for wireless video communication.

One of the fundamental problems in the wireless sensor networks is the coverage problem. The coverage problem fundamentally address the quality of service (surveillance or monitoring) provided in the desired area. In past, several studies have proposed different formulation and solutions to this problem. Nevertheless none of them has addressed minimum cost solution to the coverage problem. In this paper we consider minimum cost wireless sensor network which provides full coverage to any arbitrary geometric profile under deterministic/random deployment. We then formulate probabilistic coverage model which provides k - coverage probability for minimum cost wireless sensor networks.1