This paper addresses the joint design of Unmanned Aerial Vehicles (UAVs) trajectory and radio resource management (RRM) in dynamic wireless environments by leveraging a multi-agent deep reinforcement learning (MADRL) framework. In contrast to prior works that either assume constant synchronization between agents and the controller or overlook the communication cost, we explicitly model the interaction between UAVs and the central controller. We propose an adaptive synchronization strategy that selectively transmits model parameters and experience data based on their relevance, enabling a resource-aware RRM algorithm that optimally balances learning performance and communication overhead. The MADRL agents optimize their trajectories based on rewards that incorporate priorities derived from the RRM layer, which jointly manages both uplink and downlink communications. Simulation results demonstrate that our event-driven synchronization strategy outperforms periodic baselines in both convergence speed and communication overhead, towards scalable deployment in realistic urban environments.

In upcoming 6G networks, unmanned aerial vehicles (UAVs) are expected to play a fundamental role by acting as mobile base stations, particularly for demanding vehicle-to-everything (V2X) applications. In this scenario, one of the most challenging problems is the design of trajectories for multiple UAVs, cooperatively serving the same area. Such joint trajectory design can be performed using multi-agent deep reinforcement learning (MADRL) algorithms, but ensuring collision-free paths among UAVs becomes a critical challenge. Traditional methods involve imposing high penalties during training to discourage unsafe conditions, but these can be proven to be ineffective, whereas binary masks can be used to restrict unsafe actions, but naively applying them to all agents can lead to suboptimal solutions and inefficiencies. To address these issues, we propose a rank-based binary masking approach. Higher-ranked UAVs move optimally, while lower-ranked UAVs use this information to define improved binary masks, reducing the number of unsafe actions. This approach allows to obtain a good trade-off between exploration and exploitation, resulting in enhanced training performance, while maintaining safety constraints.

The integration of 5G New Radio (NR) into the manufacturing industry stands as a pivotal aspect of the Industrial Internet of Things (IIoT) paradigm, mitigating constraints posed by traditional wired communication technologies in terms of flexibility, mobility, and adaptability. However, ensuring comparable performance levels presents challenges in the scheduling design of 5G NR networks, particularly due to the emphasis on uplink transmissions within industrial applications. Indeed, current uplink schedulers proposed by the 3rd Generation Partnership Project (3GPP) exhibit performance trade-offs, forcing the network to acquire knowledge of uplink traffic and faltering in the presence of heterogeneous traffic sources, commonplace in industrial environments. In this paper, we thus propose a Reinforcement Learning (RL)-based approach that dynamically assigns users to the most suitable uplink scheduler without prior or acquired knowledge of the different traffic patterns. Through 3GPP-compliant network simulations, we demonstrate the effectiveness of our approach in optimizing network performance while addressing the complexities of heterogeneous IIoT environments and maintaining standard compliance.

Unmanned Aerial Vehicles (UAVs), functioning as Unmanned Aerial Base Stations (UABSs), hold considerable potential for augmenting vehicular network performance through on-demand enhanced radio coverage. A pivotal challenge lies in devising algorithms that efficiently optimize UABS trajectories under strict Radio Resource Management (RRM) and coverage gap discovery. This can be tackled using Deep Reinforcement Learning (DRL) models. However, their effectiveness relies on the relevance of acquired knowledge to the current scenario, posing a challenge when the underlying dynamics or governing rules undergo modifications. To address this issue, we propose a framework integrating a deep meta-learning algorithm to enhance the adaptability of our DRL-based trajectory design to newly encountered scenarios. Scenarios may vary in mobile users’ movement profiles, UABS take-off zones, or new service maps. Our numerical results demonstrate that an agent that leverages information from prior tasks achieves target performance in fewer episodes compared to a conventional DRL agent, while also ensuring superior long-term training proficiency.

It is expected that future 6G vehicular networks will rely on Unmanned Aerial Vehicles (UAVs) acting as flying Base Stations, namely Unmanned Aerial Base Stations (UABSs), in order to provide a wide range of services that current terrestrial networks cannot manage. Vehicles may exploit strong links with the UABS, enabling applications such as advanced driving and extended sensing. In this context, if vehicular users are satisfied with an appropriate Quality of Experience (QoE), they are able to upload a given amount of data for a given time window, continuously. To allow this, an efficient path planning is fundamental. This paper presents a Deep Reinforcement Learning (DRL)-based solution, where a novel reward function is proposed with the aim of offering a continuous service to vehicles. Results are presented in terms of the percentage of satisfied users and both, a continuous action space as well as a discrete action space, are considered exploiting two different DRL algorithms (i.e., Double Dueling Deep Q Network (3DQN) and Deep Deterministic Policy Gradient (DDPG)) in order to compare the two and select the best one according to the described scenario of interest.

Unmanned aerial base stations (UABSs) can be deployed in vehicular wireless networks to support applications such as extended sensing via vehicle-to-everything (V2X) services. A key problem in such systems is designing algorithms that can efficiently optimize the trajectory of the UABS in order to maximize coverage. In existing solutions, such optimization is carried out from scratch for any new traffic configuration, often by means of conventional reinforcement learning (RL). In this paper, we propose the use of continual meta-RL as a means to transfer information from previously experienced traffic configurations to new conditions, with the goal of reducing the time needed to optimize the UABS's policy. Adopting the Continual Meta Policy Search (CoMPS) strategy, we demonstrate significant efficiency gains as compared to conventional RL, as well as to naive transfer learning methods.

Low-power wide-area networks (LPWANs) have become an important enabler for the Internet of Things (IoT) connectivity. Application domains, such as smart cities, smart agriculture, intelligent logistics, and transportation, require communication technologies that combine long transmission ranges and energy efficiency. Recent and future trends make the long-range wide-area network (LoRaWAN) and narrowband-IoT (NB-IoT) the most prospective drivers of the IoT business. In this article, after discussing the main features of the two technologies, we carry out a fair quantitative comparison between the two, investigating different performance indicators, in order to guide designers in the selection of the most appropriate technology, depending on the application requirements.

Low-power wide-area network technologies are used to interconnect a number of devices in a simple and efficient way. One of these technologies, LoRaWAN, is deemed as one of the most promising due to its capability to allow long-range communications with very small energy consumption. LoRaWAN networks are managed by a network server implementing an adaptive data rate (ADR) algorithm to allocate proper data rates to end devices (EDs). However, the standard ADR solution focuses only on the link-level performance and assigns transmission parameters to EDs one-by-one in an independent way. In this article, we propose a novel and more efficient ADR algorithm, denoted as collision-aware ADR (CA-ADR), which tries to minimize the collision probability when assigning data rates by considering the entire set of EDs in the network and keeping the link-level performance under control. The performance of CA-ADR is characterized and benchmarked against the standard solution as well as another proposal presented in the literature. An integrated simulation-experimental approach is used to assess results for large-scale networks and to compare two architectures based on cloud and fog computing. Results show that CA-ADR outperforms standard solutions when connectivity is good, whereas it behaves similarly in large areas. It is also shown that the improvement with respect to the benchmark solutions does not depend on the channel model considered (no shadowing, uncorrelated, and correlated shadowing). Finally, a fog-based architecture is proved to be feasible, with the advantage of reducing the end-to-end latency.