In the beyond 5G era, AI/ML empowered real-world digital twins (DTs) will enable diverse network operators to collaboratively optimize their networks, ultimately improving end-user experience. Although centralized AI-based learning techniques have been shown to achieve significant network traffic accuracy, resulting in efficient network operations, they require sharing of sensitive data among operators, leading to privacy and security concerns. Distributed learning, and specifically federated learning (FL), that keeps data isolated at local clients, has emerged as an effective and promising solution for mitigating such concerns. Federated learning poses, however, new challenges in ensuring fairness both in terms of collaborative training contributions from heterogeneous data and in mitigating bias in model predictions with respect to sensitive attributes. To address these challenges, a fair FL framework is proposed for collaborative network traffic prediction and resource allocation. To demonstrate the effectiveness of the proposed approach, noniid and imbalanced federated datasets based on real-word traffic traces are utilized for an elastic optical network. The assumption is that different optical nodes may be managed by different operators. Fairness is evaluated according to the coefficient of variations measure in terms of accuracy across the operators and in terms of quality-of-service across the connections (i.e., reflecting end-user experience). It is shown that fair traffic prediction across the operators result in fairer resource allocations across the connections.

Charging electric vehicles (EVs) within parking infrastructures with a small number of charging spots is a problem that needs addressing in order to facilitate the wide adoption of EVs. Fairness and efficiency during the charging process are two important parameter that are examined in this work for this specific use case. The α-fair and fair-enough approaches are utilized, demonstrating that for the EV charging scenario considered the fair-enough approach is better applicable for practical applications, as it does not have the computational complexity of the α-fair scheme by achieving the fairest energy resource allocation possible prior to the system efficiency starts degrading. Specifically, it is demonstrated that the fair-enough scheme approximates the α-fair allocation, while also reducing processing time by up to 94%.

Traditionally, the bandwidth allocation problem is solved by maximizing network efficiency, which may however leave some connections unserved. This clearly leads to an unfair solution from the user’s point of view, rendering fair bandwidth allocation algorithms of paramount importance, especially in the presence of congested network links. Specifically, fair bandwidth allocation algorithms are necessary to effectively control the achievable quality-of-service (QoS) of end-users, or equivalently to effectively control the fairness of the bandwidth allocation decisions. As the fair bandwidth allocation problem has long concerned network operators, various measures of fairness have been proposed. Amongst the most widely applied measures is the$\alpha $-fair scheme that also captures proportional and max-min fairness by appropriately tuning the inequality aversion parameter$\alpha $. Even though this scheme allows a network operator to control the arising fairness-efficiency trade-off, the extensive processing time required for iterating over several$\alpha $-fair solutions often hinders its applicability. Committing, instead, to a single measure of fairness (e.g., proportional or max-min fairness) allows to fast approximate a fair bandwidth allocation but this may lead to either conservative QoS fairness levels or to a heavily degraded system efficiency. To alleviate limitations of known fairness measures, this work proposes a multi-objective optimization function that simultaneously optimizes both QoS fairness and network efficiency, aiming to derive an allocation that is as fair as possible to the extent that network utilization is not degraded for the sake of fairness. To evaluate the performance of the proposed function an optical network environment is considered where connections contend for the spectrum resources, demonstrating that the proposed optimization function significantly outperforms the$\alpha $-fairness scheme in terms of processing time by up to 95% and its special cases in terms of fairness and system efficiency, thus alleviating also the limitations of committing to a single measure of fairness.

In this work, the capabilities of an encoder-decoder learning framework are leveraged to predict soft-failure evolution over a long future horizon. This enables the triggering of timely repair actions with low quality-of-transmission (QoT) margins before a costly hard-failure occurs, ultimately reducing the frequency of repair actions and associated operational expenses. Specifically, it is shown that the proposed scheme is capable of triggering a repair action several days prior to the expected day of a hard-failure, contrary to soft-failure detection schemes utilizing rule-based fixed QoT margins, that may lead either to premature repair actions (i.e., several months before the event of a hard-failure) or to repair actions that are taken too late (i.e., after the hard failure has occurred). Both frameworks are evaluated and compared for a lightpath established in an elastic optical network, where soft-failure evolution can be modeled by analyzing bit-error-rate information monitored at the coherent receivers.

This work proposes a reward-based framework for fair resource allocation in reconfigurable elastic optical networks (EONs) under modeled traffic demand fluctuations. The routing and spectrum allocation (RSA) problem is formulated based on the maximization of the reward-based constant elasticity social welfare function, with rewards received by connections when allocated spectrum slots, given their a-priori modeled traffic distributions. Connections are essentially contending for link resources, with the connections’ preferences depending on the reward function used for scoring these resources. Two reward functions are utilized, considering different attributes of the optimization problem, according to fairness- and efficiency-related measures. The proposed framework can be used by a network operator for approximating in advance the provisioning of the resources for each connection that best meets predefined targets. Performance results demonstrate that both reward functions result in spectrum allocations where the quality-of-service (QoS) is as similar as possible for all connections, significantly improving connection over- and under-provisioning. Additionally, the reward function that also accounts for the modeled traffic further improves fairness and connection over-provisioning, while incurring only a small penalty in terms of the aggregate unserved traffic for all destinations.

This work examines deep quantile regression for quality-of-transmission (QoT) estimation and accurate decision making in optical networks. Quantile regression is applied to approximate QoT models capable of inferring QoT bounds for any future lightpath, according to a predefined level of certainty, for confident decision making, without the need to consider traditional margins at decision time. It is shown, that quantile regression automatically accounts for such margins, in a discriminative fashion, leading to a significant margin reduction and subsequently to more accurate inference of the QoT of unestablished lightpaths, when compared to the traditional margin-based decision approaches. Specifically, deep quantile regression for QoT estimation ensures that lightpaths with insufficient QoT will be accurately identified and rejected, while also identifying correctly lightpaths with sufficient QoT, making it a confident decision making tool for the planning of optical networks.

Trajectory prediction is crucial in assisting both human-driven and autonomous vehicles. Most of the existing approaches, however, focus on straight stretches of road and do not address trajectory prediction at intersections. This work aims to fill this gap by proposing a solution that copes with the higher complexity exhibited for the intersection scenario, leveraging the 5G-MEC capabilities. In particular, the reduced latency and edge computational power are exploited to centrally collect and process measurements from both vehicles (e.g., odometry) and road infrastructure (e.g., traffic light phases). Based on such a holistic system view, we develop a Long Short Term Memory (LSTM) recurrent neural network which, as shown through simulations using a real-world dataset, provides high-accuracy trajectory predictions. The encountered challenges and advantages of the presented approach are analyzed in detail, paving the way for a new vehicle trajectory prediction methodology.

We propose an α-fair routing and spectrum allocation (RSA) framework for reconIlgurable elastic optical networks under modeled tidal trafIlc, that is based on the maximization of the social welfare function parameterized by a scalar α (the inequality aversion parameter). The objective is to approximate an egalitarian spectrum allocation (SA) that maximizes the minimum possible SA over all connections contending for the network resources, shifting from the widely used utilitarian SA that merely maximizes the network efIlciency. A set of existing metrics are examined (i.e., connection blocking, resource utilization, coefIlcient of variation (CV) of utilities), and a set of new measures are also introduced (i.e., improvement on connection over(COP) and under-provisioning (CUP), CV of unserved trafIlc), allowing a network operator to derive and evaluate in advance a set of α-fair RSA solutions and select the one that best Ilts the performance requirements of both the individual connections and the overall network. We show that an egalitarian SA better utilizes the network resources by signiIlcantly improving both COP (up to 20%) and CUP (up to 80%), compared to the utilitarian allocation, while attaining zero blocking. Importantly, the CVs of utilities and unserved trafIlc indicate that a SA that is fairest with respect to the amount of utilities allocated to the connections does not imply that the SA is also fairest with respect to the achievable QoS of the connections, while an egalitarian SA better approximates a fairest QoS-based SA.

Dynamic network slicing has emerged as a promising and fundamental framework for meeting 5G’s diverse use cases. As machine learning (ML) is expected to play a pivotal role in the efficient control and management of these networks, in this work we examine the ML-based Quality-of-Transmission (QoT) estimation problem under the dynamic network slicing context, where each slice has to meet a different QoT requirement. We examine ML-based QoT frameworks with the aim of finding QoT model/s that are fine-tuned according to the diverse QoT requirements. Centralized and distributed frameworks are examined and compared according to their accuracy and training time. We show that the distributed QoT models outperform the centralized QoT model, especially as the number of diverse QoT requirements increases.

This paper investigates the problem of multicast traffic grooming in transparent optical networks utilizing a novel grooming approach that is based on routing/grooming of multicast calls on hybrid graphs. The proposed approach exhibits improved performance when compared to existing grooming schemes especially when the physical layer impairments are taken into account.