The Open Radio Access Network (O-RAN) architecture is key to enabling AI-driven dynamic network management. However, the complexity of this architecture introduces challenges, especially in managing conflicts between different AI-driven applications that operate concurrently within the network. These conflicts, if left unchecked, can lead to degraded network performance and service disruptions. To address this issue, we propose XAI4C (Explainable AI for Conflict Detection), a framework that leverages the SHAP (SHapley Additive exPlanations) explainable AI technique. XAI4C enhances transparency and interpretability in AI decision-making by helping network operators understand the factors driving AI decisions across different network components thereby allowing for early detection of conflicts between applications. In this paper, we first present the architecture and operation of the XAI4C framework. We then demonstrate its effectiveness in conflict detection through two case studies related to network slicing. Our results demonstrate that XAI4C outperforms the state-of-the-art PACIFISTA providing a detection accuracy increase up to 30%, while reducing the number of samples required for conflict detection by 41.17%.

Emerging mobile applications often require the execution of computer vision (CV) tasks based on compute-and memory-intensive deep neural networks (DNNs). Although offloading CV tasks to edge servers can decrease resource consumption at the mobile devices, it poses the challenge of handling multiple concurrent tasks with limited computing and memory capacity. In stark opposition with the existing state of the art, we tackle this challenge by jointly optimizing (i) the utilization of resources at the edge, among which memory - so far widely overlooked - and the radio resources used for task offloading; (ii) which and how many offloaded tasks should be executed; and (iii) the structure of the DNNs. First, we formulate the DNN for scalable Offloading of Tasks (DOT) problem, prove that it is NP-hard, and envision a weighted-tree-based heuristic solution, named OffloaDNN, that efficiently solves the DOT problem. We evaluate OffloaDNN through extensive numerical analysis using state-of-the-art image classification ResNet-18, as well as real-world experiments on the Colosseum emulator. The numerical results show that, in small-scale scenarios, OffloaDNN matches the optimum very closely, and, in larger-scale scenarios, increases the number of admitted offloaded tasks by 26.9 % with respect to the state of the art, while saving 82.5 % memory and 77.4% per-inference computing time. The numerical results are confirmed by the real-world validation on Colosseum.

Owing to large spectrum availability, millimeter wave (mmWave) communications are being proposed for 5th generation and beyond networks. However, the mmWaves are easily obstructed by objects, resulting in complete link blockage, which is more common in indoor communication. In this study, we propose to use the existing infrastructure of passive backscatter devices to establish links between the access point and a blocked legacy mobile user, with uncertain coordinates, in an indoor mmWave communication system. We set up backscatter devices in retro-reflective mode for the user to estimate the direction of arrival from them. To estimate the angles, we propose an estimator and derive its Cramer-Rao lower bound. Furthermore, while accounting for the antenna array configuration at both the user and the backscatter devices, we maximize the rate support at the user by optimizing the backscatter device reflection coefficient and the user's steering angle. The simulation results show that, by exploiting the backscatter devices already present in the environment with a sufficient number of antenna elements at the user, higher capacity is achieved as compared to that achieved by using a fixed re-configurable intelligent surface.

The existing literature on cellular multi-user mmWave communication focus on joint baseband and RF precoding designs to enable spatial multiple access with minimum interference. These studies either assume the number of users$M$is less than the number of RF units$N_{RF}$or schedule the users in time domain by dedicating one RF unit to each user if$M >N_{RF}$. It is expected that serving multiple users over OFDMA in each analog beam will offer better utilization of the wideband channel. To this end, for the scenario with$M \gg N_{RF}$we propose a sectored-cell model that is supported by multi-RF chains over the wideband mmWave channel, with each beam serving multiple users within a sector and the sectors being scheduled in round-robin fashion. We also propose a variable time frame structure that conducts sector-wise initial access, with simultaneous access to all users within a sector. It provides improved spectrum efficiency and decreased initial access delay as compared to the initial access using exhaustive beam search method. We then jointly estimate the optimum beamwidth and optimum$N_{RF}$that offer maximum average long-run user rate. Further, we introduce a reduced-complexity sector sojourn time optimization for non-homogeneously distributed users, that improves fairness of long-run user rates leveraging the variable time frame structure. The numerical results show that, while a high value of$N_{RF}$causes more interference to peak data rate, the average long-run user rate improves. Additionally, using a very narrow beam is also not optimal for providing maximum rate support. The proposed beamforming method offers a higher average long-run user rate over the competitive beamforming schemes while the complexity of user scheduling is independent of$M$.

With increasing density of data-hungry devices per unit area, allocating single highly-directed beam per user in millimeter-wave communications is not practical. Therefore, the requirement is to serve multiple users over a single beam. Considering a single-cell scenario with a fixed number of users, this paper addresses the problem of selection of optimal beamwidth depending on user density and distribution. First, by considering fixed beam service time in each sector, optimal beamwidth is estimated using exhaustive search for average long-run user rate and base station energy efficiency maximization. Based on the results of the average long-run user rate maximization using an exhaustive search, another method of reduced complexity is proposed to find sub-optimal beamwidth. Subsequently, optimum beamwidth is estimated with user density dependent variable time scheduling in a sector, that offers improved performances over fixed time scheduling. An efficient algorithm on variable time scheduling is also provided. Finally, the effect of localization error on optimal beamwidth estimation is investigated. The numerical results show that using the narrowest beam does not necessarily result in achieving a better average long-run user rate. Further, localization error does not affect the selection of optimal beamwidth, however, user Quality-of-Service degrades.