The Open Radio Access Network (O-RAN) transforms mobile networks by disaggregating hardware and software, utilizing open interfaces, and deploying commercial off-the-shelf equipment. While enhancing interoperability and vendor diversity, O-RAN faces challenges in energy efficiency and resource utilization, especially in dense urban areas with high user density and variable traffic loads. Radio Units (RUs), which handle radio frequency functions, contribute significantly to energy consumption due to continuous operation, while Distributed Units (DUs), responsible for higher-layer processing, may lack sufficient computational resources for radio functions. Together, these components complicate effective resource and energy management. To address these issues, we propose DRAGON (Deep Reinforcement learning for Autonomous Green and cOmpute-aware RANs), a DRL-based two-stage design with a Deep Q-Network (DQN) for energy-aware RU activation and a Twin Delayed Deep Deterministic Policy Gradient (TD3) agent for user-RU association under DU compute constraints and user QoS guarantees. Our performance analysis shows that DRAGON achieves 71% energy efficiency and 78.6% user rate satisfaction, yielding gains of 31.48% and 40.6%, respectively, with respect to state-of-the-art. These findings demonstrate DRAGON’s potential for sustainable and intelligent O-RAN optimization.

The combination of service virtualization and edge computing allows for low latency services, while keeping data storage and processing local. However, given the limited resources available at the edge, a conflict in resource usage arises when both virtualized user applications and network functions need to be supported. Further, the concurrent resource request by user applications and network functions is often entangled, since the data generated by the former has to be transferred by the latter, and vice versa. In this paper, we first show through experimental tests the correlation between a video-based application and a vRAN. Then, owing to the complex involved dynamics, we develop a scalable reinforcement learning framework for resource orchestration at the edge, which leverages a Pareto analysis for provable fair and efficient decisions. We validate our framework, named VERA, through a real-time proof-of-concept implementation, which we also use to obtain datasets reporting real-world operational conditions and performance. Using such experimental datasets, we demonstrate that VERA meets the KPI targets for over$96\%$of the observation period and performs similarly when executed in our real-time implementation, with KPI differences below 12.4%. Further, its scaling cost is$54\%$lower than a centralized framework based on deep-Q networks.

The efficient allocation of radio resources is an essential trait of 5G/6G radio access networks (RANs), as they are called to meet diverse QoS requirements of highly demanding applications. To equip RANs with such an ability and, at the same time, meet their function split constraints, we envision a distributed learning approach for radio resource allocation that makes the most out of the Central Unit (CU) and Distributed Unit (DU) components by effectively exploiting their synergy. On the one hand, our solution, named MERGE, leverages the knowledge of the radio connectivity dynamics that each DU can acquire through the local use of a deep reinforcement learning radio agent. On the other hand, it lets the CU collect such agents in a crowdsourcing fashion, and, then, thanks to a meta-learning policy, properly select and aggregate them to create up-to-date radio agents of the right size (hence, complexity level) to fit the computing constraints of the individual DUs. Our results show that MERGE can match the performance of the highest-complexity radio model in [1] with 25% less computational requirements, and, for a given computational resource, it outperforms a single pruned model with a 19% increase in QoS.

The combination of service virtualization and edge computing allows mobile users to enjoy low latency services, while keeping data storage and processing local. However, the network edge has limited resource availability, and when both virtualized user applications and network functions need to be supported concurrently, a natural conflict in resource usage arises. In this paper, we focus on computing and radio resources and develop a framework for resource orchestration at the edge that leverages a model-free reinforcement learning approach and a Pareto analysis, which is proved to make fair and efficient decisions. Through our testbed, we demonstrate the effectiveness of our solution in resource-limited scenarios, and show an improvement of around 60% in the CPU budget violation rate with respect to RL based standard multi-agent framework.

Recent advances in electric metering infrastructure have given rise to the generation of gigantic chunks of data. Transmission of all of these data certainly poses a significant challenge in bandwidth and storage constrained Internet of Things (IoT), where smart meters act as sensors. In this work, a novel multivariate data compression scheme is proposed for smart metering IoT. The proposed algorithm exploits the cross correlation between different variables sensed by smart meters to reduce the dimension of data. Subsequently, sparsity in each of the decorrelated streams is utilized for temporal compression. To examine the quality of compression, the multivariate data is characterized using multivariate normal-autoregressive integrated moving average modeling before compression as well as after reconstruction of the compressed data. Our performance studies indicate that compared to the state-of-the-art, the proposed technique is able to achieve impressive bandwidth saving for transmission of data over communication network without compromising faithful reconstruction of data at the receiver. The proposed algorithm is tested in a real smart metering setup and its time complexity is also analyzed.

This article presents a new paradigm for channel dynamics adaptive transmission of intermittent data in smart grid IoT communication networks, wherein novel channel prediction frameworks using stochastic modeling as well as data-driven learning of channel variability are proposed. A probing-based transmission is also proposed as a benchmark. These prediction frameworks are complemented with an adaptive channel coding scheme to increase the transmission reliability of time-critical grid monitoring data over a wireless channel. Through analyzing the prediction and packet loss performance at varying SNR and fading conditions, it is noted that the stochastic modeling framework is efficient when the fading correlation in the channel is high while the learning-based approach is more adaptive to channel dynamics as the correlation reduces. The proposed frameworks are easily implementable on low-cost end nodes, owing to the optimal selection of parameters for low runtime complexity. When compared to probing-based data transmission for a given fading in the channel, the packet loss probability of the learning-based transmission closely matches while with stochastic model loss probability is found to be 12.3% higher. However, their respective signaling overheads are 38% and 98% lower with respect to the probing-based approach, which is a significant gain at the cost of marginally additional computation complexity.

This paper presents a novel data driven framework based on $\epsilon$ -Support Vector Regression to reduce the bandwidth requirement for transmission of phasor measurement unit (PMU) data. This is achieved by judicious elimination of redundant data at the PMU before transmission. Simultaneously, the missing samples are predicted at PDC to ensure faithful identification of impending disturbances in the power system. Due to inherent nonstationary nature of PMU data, the hyperparameters are dynamically recomputed as necessary, thereby maintaining the accuracy of prediction and robustness of the algorithm. Performance of the proposed algorithm is evaluated via large scale simulations using powerline frequency data. A trade-off between prediction quality and runtime of the algorithm is observed, which is addressed by suitable selection of hyperparameters. Compared to the competitive data reduction scheme, the proposed algorithm saves around 60% bandwidth and identifies power system disturbances 73% more accurately.

In this paper, a novel characterization of smart meter data based on Gaussian mixture (GM) model is presented. It is shown that compared to the existing characterization models, the proposed GM model provides a significantly better fit for smart meter data. Furthermore, at each smart meter, sparsity of data is exploited to devise an adaptive data reduction algorithm using compressive sampling technique such that the bandwidth requirement for smart meter data transmission is reduced with minimum loss of information. When compared to the closest competitive scheme, the proposed compressive sampling based data reduction algorithm is found to be noise robust and offers 12.8% and 7.4% higher bandwidth saving, respectively, at 1 s and 30 s sampling intervals for comparable reconstruction accuracy. Proposed scheme is tested in real-time using RT-LAB.