This paper evaluates Wi-Fi Fine Timing Measure ment (FTM) for 3D indoor positioning and compares it with Ultra-Wideband (UWB). We follow a two-phase experimental methodology. First, we quantify ranging precision in a 45 m corridor under Line-of-Sight (LoS), Non-Line-of-Sight (NLoS), and mixed conditions. Second, we assess 3D positioning accuracy in a laboratory using nonlinear multilateration with four anchors, under both LoS and strong NLoS conditions. The evaluation uses commercial off-the-shelf FTM devices (Google Nest Wi-Fi and Pixel 6 Pro) and a Qorvo MDEK1001 UWB development kit. Results show that Wi-Fi FTM provides meter-level 3D positioning with minimal infrastructure, while UWB achieves substantially higher ranging precision and lower 3D positioning error under LoS. Under strong NLoS, both technologies suffer from reduced range availability, preventing 3D estimation when fewer than four ranges are returned. These results clarify the accuracy versus deployability trade-offs when selecting FTM or UWB for practical 3D indoor positioning.

Real-time Internet-of-Things (IoT) analytics in smart cities demand millisecond-scale responsiveness and energy efficiency requirements that challenge cloud-only deployments, which struggle with latency and bandwidth constraints. In brief, this paper presents a unified three-layer edge-fog decision stack comprising (i) queueing-theoretic performance modeling, (ii) NSGA-II Pareto-based offline trade-off mapping, (iii) a training-free Genetic-Algorithm Offloader for online adaptation, (iv) formal tail-latency guarantees, and (v) comprehensive validation through both simulation and a Raspberry Pi/Jetson prototype. We introduce a novel three-layer decision framework that couples a queueing-theoretic edge-fog pipeline with offline Pareto optimization and online adaptation. Our key contributions are: (1) an analytical model, validated through simulation, demonstrating edge latency of 10 ± 2 ms and fog latency of 40 ± 10 ms (processing time only, excluding network RTT); (2) NSGA-II-based offline optimization yielding Pareto-optimal configurations that reduce energy and cost by 32.6% while adding only 0.7 ms of latency; (3) A novel training-free Genetic-Algorithm Offloader (GAO) for real-time adaptation, eliminating the costly training phase required by DRL; (4) Formal tail-latency guarantees via Kingman’s bounds that ensure provable SLA compliance; and (5) a comprehensive evaluation against 15 baselines, including state-of-the-art DRL and hybrid approaches. In simulation, GAO achieves 2.03 ms 99th-percentile latency (4.7× better than the strongest DRL baseline, GNN-SAC, and 8× better than vanilla DQN), 100% SLA compliance, and 25% energy reduction. On a Raspberry Pi 5 and NVIDIA Jetson Orin Nano hardware prototype, GAO achieves 18.2 ms p99 latency and outperforms Azure IoT Edge, KubeEdge, and Eclipse ioFog by over 30% on latency and 22% on energy.

This paper addresses the challenge of achieving precise 3D localization of multiple objects in indoor environments using millimeter-wave (mmWave) sensing. mmWave positioning systems have recently emerged as a promising technology offering cm-level accuracy and robustness; however, the radar-like nature of mmWave technology presents challenges in multi-target positioning, particularly in complex environments where distinguishing between multiple objects becomes difficult. To address this, we explore clustering as a solution to analyze data from mmWave sensors and group similar data points, facilitating the identification of distinct targets. This paper aims to leverage the potential of mmWave radar technology to achieve precise ranging and angling measurements in multi-target environments, presenting a comprehensive methodology for evaluating the performance of mmWave sensors for achieving 3D positioning accuracy using four clustering approaches: K-Means, DBSCAN, Affinity Propagation, and BIRCH. The experimental results highlight the potential and challenges of each approach in terms of accuracy, robustness and execution time.

ABSTRACT The increasing demand for ultra‐reliable, low‐latency, and high‐throughput connectivity in dense urban environments presents significant challenges for next‐generation 6G networks. Traditional cellular networks, with their fixed cell boundaries and centralized base station control, are inadequate to meet the dynamic needs of such environments. A promising solution is the cell‐free network architecture, where a distributed set of access points (APs) jointly serve users without fixed cell boundaries. However, efficient access point selection and accurate user localization are crucial to achieving high performance in such networks. This paper presents a decentralized approach using Belief‐Desire‐Intention eXtended (BDIx) agents for dynamic AP selection and localization within a cluster‐based cell‐free 6G communications network. Various clustering algorithms (K‐means, DBSCAN, self‐organizing maps, MeanShift, ClusterGAN, and Autoencoders) are evaluated for their ability to optimize network throughput, energy efficiency, and spectral utilization. A hybrid localization framework, such as centroid‐based, differential circles, and multilateration methods, is employed to achieve accurate user positioning. The results demonstrate that machine learning‐based clustering methods, notably Gaussian mixture model (GMM), self‐organizing map (SOM), and ClusterGAN, offer significant improvements in throughput (up to 46.3%) and power reduction (up to 32.8%) over traditional methods. Regarding localization, deep learning models such as MLP, CNN, and TCN outperform deterministic methods, achieving sub‐meter accuracy with minimal errors (MeanDist < 1 m, > 0.999). Overall, the proposed solution enhances system scalability, energy efficiency, and positioning accuracy, establishing a promising foundation for future 6G networks. In our reference implementation, we instantiate the pipeline with a GMM for AP/UE clustering and a multilayer perceptron (MLP) regressor for localization.

Data Processing Units (DPUs) are becoming increasingly popular, especially for use in conjunction with Warehouse-Scale Computers (WSCs) due to their ability to handle networking functions and data-centric workloads. Cost-performance, energy efficiency, network 1/0, and batch processing workloads are important design factors for WSCs. Recent developments in AI and the never-ending increase in demand for data processing, cloud computing, and HPC set the optimisation of all those design factors as a high priority. DPUs can be utilised to achieve significant improvements in all those areas. This includes in-line network processing and upcoming enhanced security paradigms such as post-quantum cryptography (PQC) for quantum re-silient communications or software-defined perimeters (SDP) for confidential computing implementations. Being P4-enabled and dRMT-based, DPUs allow for the reconfigurability of the network traffic without the need to change the hardware. However, the network performance on such devices is only sometimes deter-ministic since the actual traffic and the rules, both of which have to do with packet processing, are not known during compile time. In this paper, we envision how the network performance on DPUs can be accelerated. We describe the challenges that negatively impact the bandwidth and latency: the complex steering pipeline and the massive runtime needed to optimise. These challenges arise from the lack of information during compile time that is only known during runtime. Thus, we envision optimising during runtime by leveraging DPUs' reconfigurability on the network 110. For this, we discuss the significant factors that must be considered to accelerate the network performance on such devices and we propose a solution for them.