Dynamic Discontinuous Reception (D-DRX) is an innovative power-saving mechanism for mobile devices in cellular networks, enhancing traditional DRX by enabling real-time adjustments to DRX parameters based on network conditions and dynamic traffic patterns. In this demonstration, we showcase a novel implementation of D-DRX on top of the open-source OpenAirInterface (OAI) [1] 5G platform, namely Deep Reinforcement Learning based Energy Saver (DRL-ES). DRL-ES is running as an xApp within the open-source O-RAN compliant RIC, FlexRIC [2], allowing for seamless adaptation to changing network dynamics. The DRL-ES operates by continuously monitoring Radio Link Control (RLC) latency data as input from the Distributed Unit (DU). DRL-ES leverages the Deep Q-Network (DQN) algorithm, which is trained to select the best DRX parameters, specifically the on-duration parameter for each User Equipment (UE). This focused approach optimizes the UE's active listening period within each DRX cycle based on observed RLC latency and achieves a balance between energy saving and Low latency. The DRL-ES utilizes two O-RAN compliant Service Models (SM): the Key Performance Measurement (KPM) SM to obtain RLC latency input and the RAN Control (RC) SM to send the on-duration values. This implementation adheres to existing 3GPP and O-RAN standards while demonstrating the potential of Artificial Intelligence (AI) driven optimization in 5G networks and beyond. Using DRL-ES, network operators can enhance battery life for mobile devices without compromising performance, particularly critical services.

The emergence of new computing paradigms such as Edge Computing, Fog Computing, and Far-Edge Computing is driven by the increasing demands of modern applications. Together, these paradigms form the Cloud-Edge Computing Continuum (CECC), presenting new challenges in resource allocation and incentive-driven interactions. New stakeholders are joining the business market to make a profit by selling their services (i.e., infrastructure resources, applications, or virtual resources). These actors, namely, infrastructure providers and service providers, have conflicting goals in terms of making a profit. There is a need to study and model the business interaction between these actors, especially considering the distributed nature of continuum. In this paper, we tackle the resource allocation and pricing problem in the context of CECC. We first propose a system model of the incentive interactions between actors of the continuum, where the price of resources varies based on different factors. Then, we formulate a budget-aware resource bidding problem where the objective is to jointly maximize the budget of a service provider and minimize Service Level Agreement (SLA) violations. To address this challenge, we propose a Deep Reinforcement Learning (DRL) approach that efficiently balances budget expenditure and SLA compliance. Our experimental results demonstrate that the proposed method effectively achieves a favorable trade-off between budget management and SLA satisfaction.

Multi-Access Edge Computing (MEC) is a key technology in the field of telecommunications and computing. It brings computing and storage resources closer to the edge of the network, typically at or near base stations and hence reduces the access latency of User Equipment (UE) to applications hosted at the edge. However, mobility of UEs brings challenging issues for service continuity and Service Level Agreement (SLA) fulfilment of 5G services. To solve these issues, 3GPP introduced a new Network Function (NF) called the Edge Application Server Discovery Function (EASDF) [1]. The latter aims to support session breakouts by dynamically resolving the Domain Name Service (DNS) of MEC applications to application servers closer to the UE's physical location. However, the 3GPP specifications [1] do not provide details about how the EASDF handles the UE's mobility. To fill this gap, we propose a novel design and implementation of the EASDF on the top of OpenAirInterface (OAI) open-source 5G network [2]. Simulation results show the efficiency of the EASDF in reducing the access latency during the UE's mobility with a small overhead of less than 4ms in high-load scenarios.

The mobile industry - via forums such as the O-RAN Alliance and Linux Foundation CAMARA - is working on network APIs that allow a mobile network operator to expose network capabilities to application developers. One of these APIs is the Quality on Demand (QoD) API, which enables the application to ask for additional network resources for improved latency or bandwidth. In this work, we show how an intelligent content delivery network (CDN) can exploit these APIs to improve the quality of experience (QoE) of video streaming despite difficult network conditions by boosting the available network bandwidth at precise moments in time. As the bandwidth boost is only applied whenever necessary, we avoid the caveat of constantly and statically assigning network resources to a service. We propose two boosting strategies both relying on information provided by the video player via Common Media Client Data (CMCD). We implemented the approach and evaluated it on an emulation testbed and on top of an actual 5G O-RAN compliant network capable of running xApps and the CAMARA QoD API. Our evaluation shows the gains in terms of QoE but also highlights possible caveats and adverse interactions with the ABR algorithm of the video player.

Dynamic Time Division Duplex (D-TDD) is a promising solution to address newly emerging 5G and 6G services characterized by asymmetric and dynamic uplink (UL) and downlink (DL) traffic demands. However, there are two major issues: (i) determining the TDD scheme (i.e., the number of slots devoted to UL and DL) to meet the dynamic traffic demands of the Users Equipment (UE); (ii) cross-link interference between cells that use different TDD schemes. The 3GPP standard neither specifies algorithms or solutions to derive the TDD configuration nor solves the cross-link interference. To fill this gap, we model the dynamic TDD problem in 5G NR as a linear programming problem. Then, we design Multi-Agent Deep Reinforcement Learning based 5G RAN TDD Pattern (MADRP), a fully decentralized solution based on the Multi-Agent Deep Reinforcement Learning (MADRL) approach. Based on the simulation results, the algorithm effectively prevents buffer overflows, avoids cross-link interference, and adapts to changes in the traffic pattern, ensuring its versatility. We compared our solution with the optimal solution and different static TDD configurations. We found that MADRP outperforms the static TDD configurations. We finally discuss the algorithm's limitations in terms of the number of cells, traffic variance, and cross-link interference probability.

The Network Data Analytics Function (NWDAF) is a key component of the 5G Core Network (CN) architecture whose role is to generate analytics and insights from the network data to accommodate end users and improve the network performance. NWDAF allows the collection, processing, and analysis of network data to enable a variety of applications, such as User Equipment (UE) mobility analytics and UE abnormal behaviour. Although defined by 3GPP, realizing these applications is still an open problem. To fill this gap: (i) we propose a microservices architecture of NWDAF to plug the 3GPP applications as mi-croservices enabling greater flexibility and scalability of NWDAF; (ii) devise a Machine Learning (ML) algorithm, specifically an LSTM Auto-encoder whose role is to detect abnormal traffic events using real network data extracted from the Milano dataset [1]; (iii) we integrate and test the abnormal traffic detection algorithm in the NWDAF based on OpenAirInterface (OAI) 5G CN and RAN [2]. The experimental results show the ability of NWDAF to collect data from a real 5G CN using 3GPP-compliant interfaces and detect abnormal traffic generated by a real UE using ML.

Future generation cellular networks consider Power Consumption (PC) as a key concern in designing and operating wireless communication systems. In this context, 3GPP has proposed several techniques to reduce User Equipment (UE) PC, such as Connected-mode Discontinuous Reception (C-DRX), with a new set of parameters introduced by 5G New Radio (NR) and BandWidth Part (BWP) adaptation. However, they did not specify how to derive the C-DRX parameters and BWP configuration that reduce the PC while avoiding latency overflow. To address this shortcoming, we propose a novel solution to jointly derive the C-DRX parameters and the BWP configuration to find a trade-off between low PC and low latency. Given the inherent dynamics and uncertainty in wireless network environments, our solution relies on Deep Reinforcement Learning (DRL) to learn from the dynamic traffic pattern and derive the best C-DRX and BWP configuration that minimizes PC while achieving low latency. Simulation results demonstrate the effectiveness of the proposed methodology in reducing the PC (i.e., 50-95% power gain) while avoiding latency overflow for a different number of connected UEs (i.e., 1 to 20 UEs).

Dynamic Time Duplex Division (D-TDD) is a promising solution to accommodate the new emerging 5G and 6G services characterised by asymmetric and dynamic Uplink (UL) and Downlink (DL) traffic demands. D-TDD dynamically changes the TDD configuration of a cell without interrupting users’ connectivity, hence balancing the bandwidth for UL or DL communication according to the traffic pattern. However, 3GPP standard does not specify algorithms or solutions to derive the TDD configuration, i.e., the number of slots to dedicate to UL and DL. In [1], we have proposed a Machine Learning (ML)-based solution relaying on Deep Reinforcement Learning (DRL) to allow the base station (or gNB) to self-adapt to the traffic pattern of the cell by periodically adapting the number of slots dedicated to UL and DL. In this work, we implemented the DRL algorithm on top of an open-source gNB based on OpenAirInterface (OAI) [2] to demonstrate its efficiency. To this end, we relied on the O-RAN architecture [3], where the proposed DRL algorithm is deployed as xApp at the Near Real-time RAN Intelligent Controller (RIC) and communicates with the base station using O-RAN E2 interface. We developed xTDD Service Model (SM) following the E2SM standard [3], allowing the DRL solution to monitor DL and UL buffers from the gNB to deduce the optimal TDD configuration that accommodates the current traffic. Then, the decision (i.e., TDD configuration) is pushed to the base station. We implemented the solution on top of the OAI 5G StandAlone (SA) platform and Flexric RIC [4]. To the best of our knowledge, this is the first demonstration of a ML-based D-TDD on top of a real 5G network, showing the advantage of O-RAN architecture to building Self Organized Network (SON) function for dynamic configuration of D-TDD.

5G networks and beyond are shifting from dominant Downlink (DL) traffic to a more equilibrate DL/UpLink (UL) and dominant UL traffic for specific emerging services. Particularly for ultra-Reliable and Low Latency Communications (uRLLC) services, the UL latency becomes an essential factor to consider. However, current UL scheduling methods are not efficient in terms of Physical Resource Blocks (PRBs) allocation, latency, or link adaptation. In this paper, we address the emerging challenge related to the UL latency in 5G networks and beyond. We introduce a solution based on Deep Reinforcement Learning (DRL) to dynamically allocate the future UL grant by learning from the dynamic traffic pattern. Simulation results demonstrate the efficiency of the proposed methodology in reducing the UL latency down to 0.25 ms and ensuring the generality by reacting to different traffic models.

5G New Radio (NR) introduces several key features to support the new emerging vertical industry use-cases, mainly: (1) Different numerology that gives more flexibility in managing time slot duration, and hence satisfying different delay requirements; (2) Bandwidth part that permits dedicating parts of the bandwidth to ensure different data rate requirements. However, although 5G NR introduces several enhancements, it makes radio resource management, more precisely resource scheduling, more complex and challenging. In this paper, we address the challenge of radio resource management in 5G NR featuring network slicing. We introduce a novel scheduling solution based on Deep Reinforcement Learning (DRL) to allocate resources and numerology for UEs to satisfy their different requirements. We evaluated the solution for different network configurations and compared its performance with the maximum achievable throughput. Simulation results demonstrated the efficiency of the proposed algorithm to allocate resources and the ability to scale for larger bandwidths covering both Frequency Range 1 (FR1) and FR2, as well as serving a higher number of User Equipment (UE).

The deployment of 5G and 6G is highly motivated by the emerging network services that demand more band-width and very low latency. Besides, these services are shifting from dominant Downlink (DL) Traffic to a more equilibrate DL/UpLink (UL) and dominant UL traffic for specific emerging services. One option to accommodate this new behavior is to use Time Duplex Division (TDD), where the radio frame is shared between UL and DL time slots, namely UL/DL pattern. While 4G TDD has a fixed number of configurations that cannot be updated on runtime, 5GNR allows complete flexibility to define the UL/DL pattern. Therefore, 5G base stations can dynamically change the pattern to adapt to the type of traffic (i.e., UL or DL). However, the 5G standard does not specify algorithms or solutions to derive the UL/DL pattern. To fill this gap, we propose a Deep Reinforcement Learning (DRL) that adds intelligence to the base station to self-adapt to the traffic pattern of the cell type. The proposed DRL algorithm monitors UL and DL buffers at the 5G base station to derive the optimal UL/DL pattern in respect to the current traffic configuration. The proposed solution delivers the optimal configuration in a timely and efficient manner. Simulation results demonstrated the efficiency of the proposed algorithm to avoid buffer overflow and ensure the generality by reacting to traffic pattern changes.

Radio Link Failure (RLF) is a challenging problem in 5G networks as it may decrease communication reliability and increases latency. This is against the objectives of 5G, particularly for the ultra-Reliable Low Latency Communications (uRLLC) traffic class. RLF can be predicted using radio measurements reported by User Equipment (UE)s, such as Reference Signal Receive Power (RSRP), Reference Signal Receive Quality (RSRQ), Channel Quality Indicator (CQI), and Power HeadRoom (PHR). However, it is very challenging to derive a closed-form model that derives RLF from these measurements. To fill this gap, we propose to use Machine Learning (ML) techniques, and specifically, a combination of Long Short Term Memory (LSTM) and Support Vector Machine (SVM), to find the correlation between these measurements and RLF. The RLF prediction model was trained with real data obtained from a 5G testbed. The validation process of the model showed an accuracy of 98% when predicting the connection status (i.e., RLF). Moreover, to illustrate the usage of the RLF prediction model, we introduced two use-cases: handover optimization and UAV trajectory adjustment.