International audience

Accurate and reliable localization is a key requirement for 6G network operations, but it can be particularly challenging in outdoor environments. In this paper, we propose a machine learning (ML)–based localization framework that leverages angle of arrival (AoA) as a feature extracted from channel state information (CSI). The proposed approach employs high-resolution AoA estimation algorithms, including multiple signal classification (MUSIC) and estimation of signal parameters via rotational invariance techniques (ESPRIT), which feed a hierarchical, two-stage classifier to identify specific trajectories (hereafter referred to as tracks) in a given outdoor environment. The first stage of the classifier is a binary line-of-sight (LoS) / non-line-of-sight (NLoS) classifier, followed by region-specific multi-class classifiers for fine-grained identification of the specific LoS or NLoS tracks. We evaluate our approach using a real-world massive multiple-input multiple-output (mMIMO) orthogonal frequency division multiplexing (OFDM) outdoor CSI dataset collected at the Nokia campus in Stuttgart, Germany. Experimental results show that i) the LoS / NLoS identification accuracy can reach 100%, and ii) the proposed two-stage approach significantly outperforms a single-stage multi-class baseline, achieving accuracy over 98% in LoS regions and 95% in NLoS regions. These findings demonstrate the potential of combining AoA with ML for robust localization in outdoor mMIMO propagation environments.

Open radio access network (O-RAN) is one of the key solutions envisaged to meet the ever increasing and stringent quality of service (QoS) demands of emerging applications and services supported by beyond 5G networks. In addition to meeting these QoS demands, O-RAN also promotes vendor diversity and interoperability by disaggregating the radio access network (RAN) into different units, namely the central unit (CU), distributed unit (DU), and radio unit (RU). However, the disaggregation introduces a new challenge of ensuring seamless interoperability between the various units with a view to ensuring that the QoS of the user equipments (UEs) is met and the energy consumption of the system is managed efficiently. In this paper, we focus on jointly optimizing UE association and RU energy consumption. We express the problem as an optimization problem and propose a scheme that determines the optimal RU among potential RUs that a UE should be associated to while minimizing the total used RUs. Moreover, we employ a power control mechanism that enables idle RUs without associated UEs to be put in sleep mode. To show the efficiency of the proposed algorithm, we evaluated it against some state of the art (SoA) solutions. The results indicate that the proposed solution minimizes energy consumption by about 171 % and 32 % under low and high number of UEs, respectively, while incurring a paltry 3.2% increase in access network delay for UEs without violating their delay requirements.

Network slicing is a promising solution to handle the multiple use cases of the 5G system with their diverse requirements. However, applying a static slicing scheme could cause waste of resource and affect the performance. Therefore, dynamic slicing and resource management are necessary to manage the network's resources efficiently among different slices fulfilling their respective requirements. Network slicing approaches proposed in the literature usually consider a small number of slices and do not account for the numerous 5G services simultaneously. In this paper, we study the network slicing problem at the level of the network's computing and storage resources at the edge to support a large number of slices. We model the network and formulate the slicing problem as an integer linear program to maximize the estimated volume of accepted requests of diverse services. We propose a deep reinforcement learning (DRL) based approach to find a near-optimal solution for the NP-hard optimization problem. We assess the performance and the convergence of the proposed algorithm considering multiple configurations. Moreover, we apply the results to a heuristic task offloading that employs network slicing and compare the performance with static slicing. The simulation results show, i) the impact and importance of the parameters during the model's training and building, ii) the proposed algorithm's efficiency in presence of a large number of slices, iii) the importance of dynamic resource management in network slicing.

Vehicular Edge Computing (VEC) technology allows vehicles demanding significant computation and storage resources to offload their demands to the nearest edge computing node, aiming at reducing data transfer latency and enhancing the Quality of Service (QoS). Moreover, the heterogeneous applications of Vehicle-to-Everything (V2X) communications need an efficient management of the nodes' resources to satisfy the diverse requirements of the vehicles' demands. To this end, network slicing could be a promising solution. Task offloading algorithms in VEC proposed in the literature usually rely on offloading to a 5G base station (gNodeB) or a Road Side Unit (RSU) and do not differentiate between the various vehicular demands. In this paper, we study the task offloading problem with network slicing in V2X communications from vehicles to edge computing nodes hosted at gNodeBs, RSUs, and nearby vehicles. We model the network and formulate the problem as an integer linear program, with the objective of maximizing the volume of offloaded tasks from diverse services. We propose a heuristic algorithm and slicing schemes to find a near-optimal solution to the NP hard optimization problem. The simulation results show that considering offloading at nearby vehicles in addition to RSUs and gNodeBs yields better results in terms of acceptance ratio and resource utilization. Furthermore, it is found that it is beneficial to use an adaptive slicing scheme instead of relying on a fixed slicing; in particular, when the number of slices is large.

In vehicular networks, the benefits of jointly-optimizing data prefetching and caching with broadcast transmission scheduling and rate adaptation strategies at road-side units (RSUs) have already been demonstrated in the literature. Nevertheless, the effectiveness of the solution depends greatly on the accurate knowledge of vehicular trajectories. In practice, as specified in the standards, this is at odds with the important issue of privacy. One of the main contributions of this work is to address this issue by providing a scheme that jointly optimizes the throughput of broadcast-transmission scheduling from RSUs to vehicles, and the privacy of vehicles’ locations, by enabling them to disseminate obfuscated locations to the server from time to time. We formulate this problem as a reinforcement learning (RL) problem, where the vehicles learn when to report obfuscated locations, in order to maximize a utility that encompasses both the network capacity (related to vehicles’ throughput) and the level of privacy achieved. The proposed scheme is shown to consistently outperform alternative randomized schemes considered in past work, and in particular it proves to be robust against prediction errors of the future locations of the vehicles.

Vehicular edge computing (VEC) provides an environment for offloading tasks from vehicles. Indeed, the advantage through VEC is to push power computational and storage capacities at the edge nodes near the vehicles to handle the enormous resources required by some applications. On the other hand, in order to manage efficiently these resources, it would be necessary to partition them into several parts, each dedicated to a specific service. Thus, integrating network slicing in VEC appears to be relevant. Therefore, in this paper we study the task offloading problem from vehicles to wireless 5G new generation nodes (gNBs) and road side units (RSUs) hosting sliced edge computing servers. We formulate the problem as an integer linear programming problem and propose a new algorithm, which follows a centralized control strategy to holistically view and manage the whole network, and the sliced edge nodes. In addition, it follows network function virtualization framework to separate the logical network from the physical resources. The simulation results show that, in terms of acceptance ratio, the proposed algorithm provides very close results to the optimal solution, and when compared to state-of-art algorithm, integrating slicing is better when there is enough resources on the hosting nodes, but it still guarantees the differentiation among services.

Prefetching and caching content at road-side units (RSUs) and broadcast-transmission scheduling can be shown to improve the throughput in vehicular networks considerably, provided vehicular trajectories prediction are accurate enough. As such many studies assume implicitly that such trajectories are accurately available via GPS or location tracking. In practice, as discussed in the standards, this raises a big issue of privacy. In this paper, we focus on jointly optimizing the broadcastscheduling throughput from RSUs to vehicles, while preserving the privacy of vehicles by enabling them to disseminate obfuscated location information to the server. As it is difficult to predict the vehicular throughput according to its disseminated obfuscated locations, we propose to use the capacity based on reported information as an approximation. We formulate the problem as a reinforcement learning (RL) problem, where the decision variables concern the action to obfuscate disseminated location according to current location and the last disseminated location, with the objective of maximizing a utility function that consists of a weighted sum of the capacity and the level of privacy. Simulation results show that, the proposed scheme consistently outperforms the randomized benchmark, and is insensitive to the prediction accuracy of vehicles' future locations.

Prefetching data at the road side units (RSUs) and transmitting them to interested vehicles can help reduce the traffic load in vehicular networks, and improve the data retrieval time. Most prior work in this area opted for supporting data delivery by using a single data rate, in stark contrast to common networking knowledge that good tradeoffs can be achieved in terms of performance by adopting different, more appropriate data rates for different users. In particular, in general wireless networks it is well known that when the data rate is small, the transmission is robust but takes too long; while when the data rate is high, the data delivery time is smaller but the coverage area of the RSU becomes smaller and so does the robustness. In this paper, we study the joint problem of caching and scheduling to decide, what data to prefetch, when to deliver it, and which data rate to use, with the objective of maximizing the volume of data delivered to the mobile vehicles in the network. This problem is subject to many constraints imposed by the real system limits, such as the mobility pattern of the vehicles, the distribution of data requests by the vehicles, and the limited buffer space available at the RSUs. We formulate the problem as an integer linear programming problem, and propose a heuristic caching algorithm, and a heuristic scheduling algorithm to find an approximate solution that is shown via simulation to improve the throughput, compared to standard alternative approaches.

In this paper, we propose a distributed implementation of the opportunistic interference alignment (OIA) technique developed in [1]. Therein, global channel state information (CSI) is assumed at the secondary transmitter that is, the knowledge of all the channel transfer matrices is required, which may be difficult or impossible to obtain in practice. Therefore, we propose to relax this assumption by only assuming local CSI at the transmitters and that the covariance of the received secondary signal is available at the secondary transmitter; this setup is quite similar to the one assumed for the MIMO iterative water-filling algorithm [2]. One of the key ingredients which allows the secondary transmitter to implement the OIA condition of [1] by only having this reduced knowledge is that the primary transmitter reveals information about its local channels to the secondary transmitter by embedding (for one time-slot typically, supposedly among many others) in its pre-processing matrix the information needed at the secondary. The proposed distributed implementation is proved to be effective analytically but simulations are also provided to prove that it seems to be robust against imperfect covariance feedback.