In the fight against cyber attacks, Network Softwarization (NS) is a flexible and adaptable shield, using advanced software to spot malicious activity in regular network traffic. However, the availability of comprehensive datasets for mobile networks, which are fundamental for the development of Machine Learning (ML) solutions for attack detection near their source, is still limited. Cross-Domain Artificial Intelligence (AI) can be the key to address this, although its application in Open Radio Access Network (O-RAN) is still at its infancy. To address these challenges, we deployed an end-to-end O-RAN network, that was used to collect data from the RAN and the transport network. These datasets allow us to combine the knowledge from an in-network ML traffic classifier for attack detection to bolster the training of an ML-based traffic classifier specifically tailored for the RAN. Our results demonstrate the potential of the proposed approach, achieving an accuracy rate of 93%. This approach not only bridges critical gaps in mobile network security but also showcases the potential of cross-domain AI in enhancing the efficacy of network security measures.

Relevant services envisaged for beyond 5G (B5G) systems, such as extended reality and holographic communications, have extremely demanding user experience requirements with significant computational and communication demands. While edge computing aims to address the computation requirements by offloading the computational tasks to edge servers near the user, the communication will take advantage of the technologies developed for 5G New Radio jointly with an never-before-seen degree of network densification. This paper proposes the use of relays with edge computing capabilities. The approach's potential for B5G are identified, and a system model is defined to characterize both computational and communications viewpoints. Based on this, results are provided to highlight the gains and limitations of the proposed approach from a system-level perspective. Finally, the main challenges for enabling relays with computing capabilities in B5G deployments are discussed.

Radio Resource Management (RRM) strategies are essential components in mobile wireless networks, such as Long Term Evolution (LTE) or 5G New Radio (NR)[1]. The design of RRM algorithmic solutions is an area of research that has received a lot of attention for decades (see e.g. surveys [2–4] and references therein). Given that RRM strategies need to handle traffic dynamicity in the Radio Access Network (RAN), the availability of realistic user distributions in space and time can be very useful to support a more realistic performance assessment of RRM solutions. For this purpose, this article introduces a dataset containing real measurements of the number of users connected to the different Wifi Access Points (APs) at the Campus Nord facilities of the Universitat Politècnica de Catalunya (UPC) in Barcelona. The Wifi network is composed of 247 APs. To characterize the temporal variations, the data was collected for each AP every 1000 s (approximately) during 62 days. Besides the number of users connected to each AP (including users connected to both 2.4 GHz and 5 GHz bands), the dataset also contains information about the theoretical coverage area of each AP, so that the number of users connected to each AP can be associated to a specific geographical area. In this way, the dataset captures the spatio-temporal variations of users in the Campus at different times of the day, different days of the week and different periods of the academic year.

Relevant services envisaged for beyond 5G(B5G) systems, such as extended reality and holographic communications, present stringent user experience requirements with high computational and communication demands. While edge computing aims to address the computation requirements by offloading the computational tasks to edge servers close to the user, the communication will leverage the technologies developed for 5G New Radio together with an unprecedented level of network densification. This paper advocates for deploying relays equipped with edge computing capabilities. The potentials of this approach for B5G are identified and a system model is presented to characterize both computational and communications perspectives. Based on this, results are provided to show the benefits and limitations of the proposed approach from a system-level perspective.

This paper proposes a model to detect and solve coverage holes in 5G Radio Access Network (RAN) deployments operating with millimeter waves. The proposed model utilizes a DBSCAN-based detector to identify coverage constrained areas and then proposes the use of Relay User Equipment (RUE) capabilities to extend the RAN coverage. To optimize the activation and deactivation of RUEs, a Deep-Q-Network-based algorithm is proposed, aiming to improve spectral efficiency and decrease the outage probability experienced by network users. The obtained results demonstrate the effectiveness of the model in accurately detecting coverage constrained areas and efficiently solving these issues by means of an effective RUE activation, leading to significant improvements in network performance while minimizing the time that RUEs remain in active mode, which implies potential benefits for MNOs and UE holders and significant energy savings.

Deep Reinforcement Learning (DRL) has recently emerged as a promising technique to deal with different problems in the 5G and beyond Radio Access Network (RAN). The practical implementation of DRL solutions in the real network embraces a training process that is fundamental to materialize the expected benefits associated to these techniques. However, little effort has been devoted in the literature to analyze this training process when applying DRL in the RAN. In an effort to contribute to fill this gap, this paper presents an impact analysis of the training process on the obtained performance by a DRL solution for RAN slicing. To this end, a methodology to specify the training dataset is introduced together with the definition of relevant metrics. Then, the paper presents different simulation results to determine the features of the training dataset that allow a satisfactory training and a high performance of the obtained policy when applied during the inference stage.

Reinforcement Learning (RL)-based algorithmic solutions have been profusely proposed in recent years for addressing multiple problems in the Radio Access Network (RAN). However, how RL algorithms have to be trained for a successful exploitation has not received sufficient attention. To address this limitation, which is particularly relevant given the peculiarities of wireless communications, this paper proposes a functional framework for training RL strategies in the RAN. The framework is aligned with the O-RAN Alliance machine learning workflow and introduces specific functionalities for RL, such as the way of specifying the training datasets, the mechanisms to monitor the performance of the trained policies during inference in the real network, and the capability to conduct a retraining if necessary. The proposed framework is illustrated with a relevant use case in 5G, namely RAN slicing, by considering a Deep Q-Network algorithm for capacity sharing. Finally, insights on other possible applicability examples of the proposed framework are provided.

Multi-connectivity emerges as a useful feature to handle the traffic in heterogeneous cellular scenarios and fulfill the demanding requirements in terms of data rate and reliability. It allows a device to be simultaneously connected to multiple cells belonging to different radio access network nodes from a single or multiple radio access technologies. This paper addresses the problem of optimally splitting the traffic among cells when multi-connectivity is used. For this purpose, it proposes the use of deep learning to determine the optimum amount of traffic of a device that needs to be sent through one or another cell depending on the current traffic and radio conditions. Obtained results reveal a promising capability of the proposed Deep Q Network solution to select quasi optimum traffic splits in the considered scenario.

One of the key features of the 5G architecture is network slicing, which allows the simultaneous support of diverse service types with heterogeneous requirements over a common network infrastructure. In order to support this feature in the Radio Access Network (RAN), it is required to have capacity sharing mechanisms that distribute the available capacity in each cell among the existing RAN slices while satisfying their requirements and efficiently using the available resources. Deep Reinforcement Learning (DRL) techniques are good candidates to deal with the complexity of capacity sharing in multi-cell scenarios where the traffic in the different cells can be heterogeneously distributed in the time and space domains. In this paper, a multi-agent reinforcement learning-based solution for capacity sharing in multi-cell scenarios is discussed and assessed under heterogeneous traffic conditions. Results show the capability of the solution to satisfy the requirements of the RAN slices while using the resources in the different cells efficiently.

Network slicing is included as a key feature of the 5G architecture in order to simultaneously support diverse service types with heterogeneous requirements. The deployment of network slicing in the Radio Access Network (RAN) needs mechanisms that allow the distribution of the available capacity in the system in an efficient manner while satisfying the requirements of the different services. In this paper, a capacity sharing function is proposed, which is approached as a multi-agent reinforcement learning based on the Deep Reinforcement Learning (DRL) algorithm Deep Q-Network (DQN). The proposed algorithm provides the capacity to be assigned to each RAN slice. Performance assessment reveals the promising behaviour of the proposed solution.

The network slicing capability introduced in 5G systems facilitates the realisation of flexible multi-tenant networks. Network slicing enables the partition of a common shared network in several logical networks, each configured to fulfil specific service requirements. In scenarios where the lifecycle of network slices has to be managed dynamically (e.g. allocation, modification and deallocation of network slices in response to changing tenants’ needs), slice admission control becomes a central function to assure that the set of slices concurrently activated count with the sufficient resources to fulfil their service requirements. Slice admission control is particularly challenging for the Radio Access Network (RAN) part of a slice, because the amount of required radio resources is highly dependent on the characteristics of the deployment environment and type of cells. In this context, this paper presents a functional data-analytics framework along with a new analytical methodology for estimating the radio resource requirements for RAN slice admission control. Specifically, in order to characterise the propagation and interference conditions in each cell, the proposed resource estimation method leverages statistical information extracted via data analytics from the cell performance measurements collected at the management plane. Results show the benefits of the proposed estimation method under different types of cell deployments.