One of the key requirements for future 5th Generation (5G) Industrial Internet of Things (IIoT) networks will be to deliver low latency to support different production processes. To this end, 5G New Radio (NR) provides Configured Grant (CG) scheduling for periodic traffic, additionally to conventional Grant-Based Scheduling (GBS). However, in view of the complexities introduced by spatio-temporal traffic correlations in IIoT, a fixed scheduler configuration may be suboptimal: GBS introduces excessive signaling overhead, while CG leads to inefficient resource utilization and latency degradation when traffic is not perfectly periodic. To solve these critical issues, we propose Hybrid Centralized Uplink Scheduler (HCUS), a new scheduling framework that dynamically learns the type of traffic generated by User Equipments (UEs), and adapts resource allocation accordingly. HCUS operates per-UE, and dynamically switches between GBS and Configured Grant (CG), optimizing resource allocation while preserving low End-to-End (E2E) latency. We consider both mixed periodic and aperiodic uplink traffic to model different network load conditions and IIoT applications. Extensive simulations show that HCUS achieves up to four times lower latency than GBS and CG while maintaining high reliability, even considering traffic correlations or periodicity changes, making it a robust and scalable solution for next-generation IIoT scenarios.

Industry 5.0 envisions the seamless integration of automation, human-machine collaboration, and intelligent systems to enable highly flexible and adaptive manufacturing. However, this puts stringent requirements on the communication systems extending beyond the capabilities of current 5G technologies. This gap motivates the adoption of the digital twin (DT) paradigm, where a digital replica of physical assets and networks enables continuous monitoring, simulation, and optimization of industrial processes. This article investigates the optimization of network access protocols, which are critical to maintaining uninterrupted industrial production. Specifically, we develop a reinforcement learning (RL) algorithm embedded within the DT framework to ensure reliable communication and production flow continuity. In a congested factory environment, an autonomous guided vehicle (AGV) must transmit data to a base station (BS) using an ALOHA-like protocol at terahertz (THz) frequencies. The DT-enabled RL model dynamically learns traffic patterns and adaptively selects backoff (BO) times for network access, maximizing reliability without requiring prior knowledge of system topology. Extensive evaluations across diverse operating scenarios-ranging from static to mobile settings, varying traffic loads, and different data and action space sizes-confirm the robustness, scalability, and adaptability of the proposed solution in highly dynamic industrial environments against conventional and advanced baselines.

The challenge of ultra-low latency is boosting the interest of Mobile Network Operators (MNOs) in advanced mitigation strategies relying on latency decomposition to trace root causes. This work proposes the first hybrid framework combining analytical and data-driven modeling to estimate Latency-Reliability (LR), defined as the percentage of packets received within a target latency threshold, for different service classes in cellular networks. A preliminary full-stack analysis of Next Generation NodeB operation, coupled with a queuing-theoretic latency model, identifies critical delay stages and maps them to relevant Key Performance Indicators (KPIs) from the wide range of network measurements collected in the Operational Support System. These KPIs feed a module that integrates Supervised Learning, expert knowledge, and statistical analysis to predict LR per service class at cell level, empowering MNOs to take proactive actions in latency-sensitive scenarios. Assessment conducted over real network data shows that the proposed approach achieves latency estimation errors below 10% across service classes, allowing for accurate LR predictions for voice traffic in the absence of ground-truth data. Additionally, dimensionality reduction improves both computational efficiency and model interpretability, supporting targeted latency mitigation.

Using cellular networks outdoors generally leads to terminals experiencing elevated power levels and increased interference from neighboring cells. These conditions are significantly different from those encountered by the same terminals when they are indoors, where absorption caused by building walls reduces the available serving cell received power and mitigates interference from other cells. Conditions where mobile radio services are predominantly used by terminals located outdoors or, conversely, predominantly used within buildings, typically mix depending on the city area, the specific time, or peculiar circumstances. Therefore, the ability to track the evolution of the dominant indoor or outdoor context within a network over time is crucial for automatically adjusting, optimizing, and enhancing radiomobile service quality. Although numerous methods exist for estimating a device's indoor or outdoor status, based on specific positioning strategies, our approach stands apart due to its real-time applicability to a large-scale set of smartphones across any current 4G/5G 3GPP radiomobile network worldwide—something not achieved by other solutions in the literature. The proposed model is validated using an extensive dataset sourced from a live 5G network access.

Industrial Internet of Things (IIoT) networks will provide Ultra-Reliable Low-Latency Communication (URLLC) to support critical processes underlying the production chains. However, standard protocols for allocating wireless resources may not optimize the latency-reliability trade-off, especially for uplink communication. For example, centralized grant-based scheduling can ensure almost zero collisions, but introduces delays in the way resources are requested by the User Equipments (UEs) and granted by the gNB. In turn, distributed scheduling (e.g., based on random access), in which UEs autonomously choose the resources for transmission, may lead to potentially many collisions especially when the traffic increases. In this work we propose DIStributed combinatorial NEural linear Thompson Sampling (DISNETS), a novel scheduling framework that combines the best of the two worlds. By leveraging a feedback signal from the gNB and reinforcement learning, the UEs are trained to autonomously optimize their uplink transmissions by selecting the available resources to minimize the number of collisions, without additional message exchange to/from the gNB. DISNETS is a distributed, multi-agent adaptation of the Neural Linear Thompson Sampling (NLTS) algorithm, which has been further extended to admit multiple parallel actions. We demonstrate the superior performance of DISNETS in addressing URLLC in IIoT scenarios compared to other baselines.

We investigate cross-layer performance diagnostics for an O-RAN instance by jointly analyzing application-level latency and radio-layer behavior from a real measurement campaign. Measurements were conducted at multiple link distances (2, 6 and 11 meters) using two representative UE configurations (a commercial smartphone and a modem-based device), under both static conditions and a controlled dynamic obstruction scenario. Rather than relying on averages, the study adopts tail-focused latency characterization (e.g., 95th percentile and exceedance probabilities) and connects it to scheduler- and link-adaptation indicators (e.g., block error behavior, modulation/coding selection and signal quality). The results reveal (i) UE-dependent differences that primarily manifest in the latency tail, (ii) systematic scaling of tail latency with distance and payload and (iii) cases where radio-layer dynamics are detectable even when end-to-end latency appears stable, motivating the need for cross-layer evidence. Distinct from much of the existing literature (often centered on throughput, simulated setups, or single-layer KPIs) this work contributes a measurement-driven methodology for interpretable O-RAN diagnostics and proposes lightweight, window-based "degradation flags" that combine tail latency and radio indicators to support practical monitoring and troubleshooting.

In large-scale Internet of things networks, efficient medium access control (MAC) is critical due to the growing number of devices competing for limited communication resources. In this work, we consider a new challenge in which a set of nodes must transmit a set of shared messages to a central controller, without inter-node communication or retransmissions. Messages are distributed among random subsets of nodes, which must implicitly coordinate their transmissions over shared communication opportunities. The objective is to guarantee the delivery of all shared messages, regardless of which nodes transmit them. We first prove the optimality of deterministic strategies, and characterize the success rate degradation of a deterministic strategy under dynamic message-transmission patterns. To solve this problem, we propose a decentralized learning-based framework that enables nodes to autonomously synthesize deterministic transmission strategies aiming to maximize message delivery success, together with an online adaptation mechanism that maintains stable performance in dynamic scenarios. Extensive simulations validate the framework's effectiveness, scalability, and adaptability, demonstrating its robustness to varying network sizes and fast adaptation to dynamic changes in transmission patterns, outperforming existing multi-armed bandit approaches.

Sixth-generation (6G) wireless networks evolve from connecting devices to connecting intelligence. The focus turns to Goal-Oriented Communications, where the effectiveness of communication is assessed through task-level objectives over traditional throughput-centric metrics. As communication intertwines with learning at the edge, distributed inference over wireless networks faces a critical trade-off between task accuracy and efficient radio resource use. Traditional communication schemes (e.g., OFDMA) are not designed for this trade-off, often facing challenges related to scalability and latency. Therefore, we propose a novel goal-oriented framework that integrates over-the-air computation with spatio-temporal graph learning. Leveraging the wireless channel as an analog aggregation layer, the proposed framework enables low-latency message passing while efficiently aggregating semantically relevant features from distributed nodes. Theoretical analysis confirms that our analog architecture converges to the expressive power of digital message passing, while offering decisive scalability advantages. We assess the framework in proactive line-of-sight blockage prediction for millimeter-wave networks. Through high-fidelity ray-tracing simulations, the framework exhibits strong inductive generalization to unseen networks and adapts to domain shifts via lightweight transfer learning, matching or even outperforming digital baselines with significantly reduced communication overhead.

Digital Twin (DT) technology has emerged as a transformative paradigm across various domains, offering powerful capabilities to monitor and optimize processes prior to real-world deployment. DTs are well-suited for next-generation deployment, as well as industrial applications, in which the dynamicity and complexity of processes pose significant challenges. This study proposes a novel DT-based framework that integrates network infrastructure with shop-floor industrial operations, where robotic arms execute pick-and-place tasks and communicate with a base station (BS) using a contention-based medium access control (MAC) protocol (i.e., ALOHA and carrier sense multiple access (CSMA)) at THz frequencies. The framework aims to optimize closed-loop production systems by enabling continuous and bidirectional data exchange between the robotic arms and the BS. A centralized reinforcement learning (RL) model enables joint optimization of backoff (BO) selection and task allocation, enhancing production efficiency while preserving workflow continuity. The results prove that the proposed framework significantly outperforms state-of-the-art (SoTA) solutions for network and industrial operation performance, achieving a 55.2% reduction in average latency, a 9.8% improvement in success probability, and a 27.96% increase in processed product rate. These findings highlight the robustness and effectiveness of the proposed framework across varying operational scenarios and MAC protocols.

Industrial environments feature heterogeneous data sources with evolving and often conflicting communication requirements. This dynamic, multi-goal setting poses significant challenges, particularly due to the strict and variable demands on key performance metrics, such as latency and reliability, crucial for wireless networks enabling Industrial Internet of Things (IIoT) applications. Indeed, some sensors require low latency and high success probability, while others can tolerate delays, and these priorities can change over time as industrial processes evolve. Optimizing Medium Access Control (MAC) protocols in such conditions requires an accurate understanding of the temporal and spatial variability of sensor needs, something that traditional MAC designs are not well equipped to handle. To address this, we propose a Reinforcement Learning (RL)-based approach for MAC optimization in industrial networks. The proposed model builds upon a Carrier Sense Multiple Access (CSMA) MAC protocol, adaptively assigning Contention Window (CW) values to individual sensors to reflect the evolving communication demands of industrial processes. Its performance is evaluated against traditional benchmark approaches that rely on fixed CW configurations. Results demonstrate that the proposed solution dynamically adjusts network behavior to meet multi-goal requirements, satisfying both latency and reliability constraints by aligning CW assignments with sensor-specific needs in real time.

Efficient multiple access remains a key challenge for emerging Internet of Things (IoT) networks comprising a large set of devices with sporadic activation, thus motivating significant research in the last few years. In this paper, we consider a network wherein IoT sensors capable of energy harvesting (EH) send updates to a central server to monitor the status of the environment or machinery in which they are located. We develop energy-aware ALOHA-like multiple access schemes for such a scenario using the Age of Information (AoI) metric to quantify the freshness of an information packet. The goal is to minimize the average AoI across the entire system while adhering to energy constraints imposed by the EH process. Simulation results show that applying the designed multiple access scheme improves performance from 24% up to 90% compared to previously proposed age-dependent protocols by ensuring low average AoI and achieving scalability while simultaneously complying with the energy constraints considered.

The emergence of Terahertz (THz) frequency wireless networks holds great potential for enabling various high-demand services, including Industrial Internet of Things (IIoT) applications. These applications benefit significantly from the ultra-high data rates, low latency, and high spatial resolution offered by THz frequencies. However, a primary well-known challenge of THz networks is their limited coverage range due to high path loss and vulnerability to obstructions. This paper addresses this limitation by proposing two novel multi-hop protocols, Table-Less (TL) and Table-Based (TB), respectively, both avoiding centralized control and/or control plane transmissions. Indeed, both solutions are distributed, simple, and rapidly adaptable to network changes. Simulation results demonstrate the effectiveness of our approaches, as well as revealing interesting trade-offs between TL and TB routing protocols, both in a real IIoT THz network and under static and dynamic conditions.

The efficient operation of cellular networks requires precise tuning of configuration parameters such as the antenna downtilt or the transmit power. Data-driven methods, which can integrate feedback from monitoring data, are promising but face challenges related to sample efficiency, scalability, and safety, limiting their real-world application. In this work, we introduce an innovative online coverage and capacity optimization framework that combines model-free exploration using Monte Carlo tree search with a safe, model-based baseline derived from a probabilistic differentiable network twin. We formulate the optimization task as a sequential tree search problem, developing specialized policies to guide the exploration of the configuration space. The differentiable network twin aids both the selection policy, by pruning the search space with a prior action distribution, and the rollout policy, enabling domain knowledge-based selection of remaining actions. Our results demonstrate that the proposed approach effectively guides the optimization process, outperforming both purely model-free and model-based methods. Our solution improves safety by reducing the risk of testing poorly performing configurations, enhances the model-based solution, and can also compensate for severe model mismatches in the digital twin. The framework thus addresses a common obstacle in applying data-driven optimization to real-world network deployments.

The traditional black-box and monolithic approach to Radio Access Networks (RANs) has heavily limited flexibility and innovation. The Open RAN paradigm, and the architecture proposed by the O-RAN ALLIANCE, aim to address these limitations via openness, virtualization and network intelligence. In this work, first we propose a novel, programmable scheduler design for Open RAN Distributed Units (DUs) that can guarantee minimum throughput levels to User Equipments (UEs) via configurable weights. Then, we propose an O-RAN xApp that reconfigures the scheduler's weights dynamically based on the joint Complementary Cumulative Distribution Function (CCDF) of reported throughput values. We demonstrate the effectiveness of our approach by considering the problem of asset tracking in 5G-powered Industrial Internet of Things (IIoT) where uplink video transmissions from a set of cameras are used to detect and track assets via computer vision algorithms. We implement our programmable scheduler on the OpenAirInterface (OAI) 5G protocol stack, and test the effectiveness of our xApp control by deploying it on the O-RAN Software Community (OSC) near-RT RAN Intelligent Controller (RIC) and controlling a 5G RAN instantiated on the Colosseum Open RAN digital twin. Our experimental results demonstrate that our approach enhances the success percentage of meeting throughput requirements by 33

In modern industrial internet of things (IIoT) networks, efficient management of communications resources is crucial to match stringent application requirements. Differently from traditional resource allocation policies, goal-oriented communications is an emerging paradigm that aims at better optimizing network resource usage by prioritizing the transmission of information that is most relevant to a given task. Moreover, recent trends show increasing interest in distributed machine learning-based optimization to enhance network performance while reducing the reliance on centralized control. In this context, graph neural networks (GNNs) have emerged as a powerful tool for learning distributed policies among nodes facilitating their cooperation. In this paper, we introduce a goal-oriented, distributed framework to optimize uplink scheduling requests for coordinated message transmission to a remote server, while minimizing communication overhead. Our approach employs GNN-based distributed unsupervised learning framework that does not require a centralized controller. Extensive simulations in a 3GPP-compliant industrial scenario demonstrate that the proposed solution effectively reduces redundant uplink scheduling requests while achieving efficient and scalable coordination across multiple networks. Our findings highlight the potential of GNNs for learning distributed policies that enhance communication efficiency in wireless industrial IoT systems.

In the ever-evolving landscape of industrial connectivity, significant strides have been made in the integration of 5th generation (5G) cellular technology with industrial internet of things (IIoT) systems. At the same time, data-driven analytics has become an effective tool for leveraging information from interconnected industrial devices, enabling organizations to gain valuable insights and make informed decisions. However, among these advancements, the holistic perspective of end-to-end analysis related to their integration remains a critical aspect that has yet to be comprehensively addressed. To this end, we investigate 5G IIoT network architectures that support automated guided vehicles (AGVs) on a factory floor as an illustrative example. In particular, we leverage real sensor data collected by AGVs to estimate their remaining useful life (RUL) using a deep learning (DL)-based pipeline. We conduct an in-depth analysis to assess the compatibility of 5G New Radio infrastructures with the aforementioned case study, focusing on round trip time (RTT) requirements and emphasizing the inter-dependencies between communication network and data-driven application.

Beamforming is a crucial component in modern large-scale multiple-input single-output (MISO) wireless networks. However, its practical deployment is often constrained by the incomplete channel state information (CSI) and the overhead associated with centralized optimization methods, such as the weighted minimum mean-square error (WMMSE) algorithm. In this work, we propose the problem of distributed beamforming with varying degrees of CSI incompleteness and develop a graph neural network (GNN)-based approach, which enables transmitters to collaboratively optimize their beamforming strategies using only local and partial CSI. Our approach combines imitation learning with unsupervised learning for model training, where the former provides a warm start and the latter allows for superior performance beyond the "expert" strategy, while leveraging the GNN for a distributed implementation. We evaluate the proposed approach using ray tracing-based simulations in both indoor and outdoor scenarios, demonstrating that it outperforms WMMSE, particularly in environments with noisy or incomplete CSI. These results highlight the potential of our GNN-based solution for scalable, robust, and distributed beamforming strategies in future wireless networks.

The Industrial Internet of Things (IIoT) paradigm has emerged as a transformative force, revolutionizing industrial processes by integrating advanced wireless technologies into traditional procedures to enhance their efficiency. The importance of this paradigm shift has produced a massive, yet heterogeneous, proliferation of scientific contributions. However, these works lack a standardized and cohesive characterization of the IIoT framework coming from different entities, like the 3rd Generation Partnership Project (3GPP) or the 5G Alliance for Connected Industries and Automation (5G-ACIA), resulting in divergent perspectives and potentially hindering interoperability. To bridge this gap, this article offers a characterization of (i) the main IIoT application domains, (ii) their respective requirements, (iii) the principal technological gaps existing in the current literature, and, most importantly, (iv) we propose a systematic approach for assessing and addressing the identified research challenges. Therefore, this article serves as a roadmap for future research endeavors, promoting a unified vision of the IIoT paradigm and fostering collaborative efforts to advance the field.

The growing complexity and heterogeneity of industrial internet of things (IIoT) networks have driven research towards scalable and decentralized resource allocation strategies. Power control in such networks, particularly in interference-limited environments, is a key challenge due to varying network conditions and device capabilities. In this paper, we introduce a novel framework leveraging graph neural networks (GNNs) for decentralized power allocation in large-scale IIoT networks. By utilizing a centralized training / decentralized execution (CTDE) approach, our method applies deep deterministic policy gradient (DDPG) based unsupervised learning to optimize power allocation, with GNNs enabling localized decision-making through message passing among neighboring transmitters. We show the effectiveness of the proposed framework in both ad-hoc one-to-one and one-to-many network scenarios using 3rd generation partnership project (3GPP)-compliant simulations. Moreover, the generalization tests address the adaptability of the proposed model to unseen conditions, including wireless networks with varying densities and optimization constraints.

This paper assesses the channel characteristics in an industrial setting for Industrial IoT (IIoT) applications, comparing elementary reference scenarios with real plants, like those of Robopac-Aetna Group and the Bi-Rex pilot line in Italy. Signal propagation inside the facilities is simulated using 3DScat, a ray tracing (RT) tool. The work focuses on the extrapolation of the major large-scale channel parameters, such as the path loss exponent (PLE) and the standard deviation of shadowing, for base station (BS)-to-node and node-to-node communications at sub-6 GHz, mmWaves, and THz frequencies. Results prove that, despite their differing characteristics, the investigated scenarios exhibit similar trends in terms of PLE and shadowing standard deviation as the operating frequency and transmitter height are varied.