MR
M. Raftopoulou
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
6 records found
1
Journal article
(2025)
-
Maria Raftopoulou, José Mairton B. da Silva, Remco Litjens, H. Vincent Poor, Piet Van Mieghem
Many algorithms related to vehicular applications, such as enhanced perception of the environment, benefit from frequent updates and the use of data from multiple vehicles. Federated learning is a promising method to improve the accuracy of algorithms in the context of vehicular networks. However, limited communication bandwidth, varying wireless channel quality, and potential latency requirements may impact the number of vehicles selected for training per communication round and their assigned radio resources. In this work, we characterize the vehicles participating in federated learning based on their importance to the learning process and their use of wireless resources. We then address the joint vehicle selection and resource allocation problem, considering multi-cell networks with multi-user multiple-input multiple-output (MU-MIMO)-capable base stations and vehicles. We propose a “vehicle-beam-iterative” algorithm to approximate the solution to the resulting optimization problem. We then evaluate its performance through extensive simulations, using realistic road and mobility models, for the task of object classification of European traffic signs. Our results indicate that MU-MIMO improves the convergence time of the global model. Moreover, the application-specific accuracy targets are reached faster in scenarios where the vehicles have the same training data set sizes than in scenarios where the data set sizes differ.
...
Many algorithms related to vehicular applications, such as enhanced perception of the environment, benefit from frequent updates and the use of data from multiple vehicles. Federated learning is a promising method to improve the accuracy of algorithms in the context of vehicular networks. However, limited communication bandwidth, varying wireless channel quality, and potential latency requirements may impact the number of vehicles selected for training per communication round and their assigned radio resources. In this work, we characterize the vehicles participating in federated learning based on their importance to the learning process and their use of wireless resources. We then address the joint vehicle selection and resource allocation problem, considering multi-cell networks with multi-user multiple-input multiple-output (MU-MIMO)-capable base stations and vehicles. We propose a “vehicle-beam-iterative” algorithm to approximate the solution to the resulting optimization problem. We then evaluate its performance through extensive simulations, using realistic road and mobility models, for the task of object classification of European traffic signs. Our results indicate that MU-MIMO improves the convergence time of the global model. Moreover, the application-specific accuracy targets are reached faster in scenarios where the vehicles have the same training data set sizes than in scenarios where the data set sizes differ.
Journal article
(2024)
-
Maria Raftopoulou, José Mairton B. da Silva Jr. , Remco Litjens, H. Vincent Poor, Piet Van Mieghem
Federated learning is an effective method to train a machine learning model without requiring to aggregate the potentially sensitive data of agents in a central server. However, the limited communication bandwidth, the hardware of the agents and a potential application-specific latency requirement impact how many and which agents can participate in the learning process at each communication round. In this paper, we propose a selection metric characterizing each agent’s importance with respect to both the learning process and the resource efficiency of its wireless communication channel. Leveraging this importance metric, we formulate a general agent selection optimization problem, which can be adapted to different environments with latency or resource-oriented constraints. Considering an example wireless environment with latency constraints, the agent selection problem reduces to the 0/1 Knapsack problem, which we solve with a fully polynomial approximation. We then evaluate the agent selection policy in different scenarios, using extensive simulations for an example task of object classification of European traffic signs. The results indicate that agent selection policies which consider both learning and channel aspects provide benefits in terms of the attainable global model accuracy and/or the time needed to achieve a targeted accuracy level. However, in scenarios where agents have a limited number of data samples or where the latency requirement is very stringent, a pure learning-based agent selection policy is shown to be more beneficial during the early or late stages of the learning process.
...
Federated learning is an effective method to train a machine learning model without requiring to aggregate the potentially sensitive data of agents in a central server. However, the limited communication bandwidth, the hardware of the agents and a potential application-specific latency requirement impact how many and which agents can participate in the learning process at each communication round. In this paper, we propose a selection metric characterizing each agent’s importance with respect to both the learning process and the resource efficiency of its wireless communication channel. Leveraging this importance metric, we formulate a general agent selection optimization problem, which can be adapted to different environments with latency or resource-oriented constraints. Considering an example wireless environment with latency constraints, the agent selection problem reduces to the 0/1 Knapsack problem, which we solve with a fully polynomial approximation. We then evaluate the agent selection policy in different scenarios, using extensive simulations for an example task of object classification of European traffic signs. The results indicate that agent selection policies which consider both learning and channel aspects provide benefits in terms of the attainable global model accuracy and/or the time needed to achieve a targeted accuracy level. However, in scenarios where agents have a limited number of data samples or where the latency requirement is very stringent, a pure learning-based agent selection policy is shown to be more beneficial during the early or late stages of the learning process.
Following the trend of previous years, the number of devices, and hence the traffic in cellular networks is increasing. Moreover, new applications with stringent requirements are envisioned. Examples of such applications include collaborative learning and coverage extension with drones. To accommodate the traffic with its respective Quality of Service (QoS) requirements and to support new challenging applications in the Radio Access Network (RAN), we need to develop new algorithms and tools for efficient resource management. In this dissertation, resource management in the RAN is considered in three distinct areas.
In Chapter 2 we provide an introduction to the key concepts, which establish the technological context of the following chapters. The first part of this dissertation focuses on serving traffic with diverse requirements in the context of 5G networks. In 5G, RAN slicing has been introduced, to support services with diverse QoS requirements in the same network infrastructure. Moreover, RAN slicing allows the Mobile Network Operators (MNOs) to configure customer-specific slices. In Chapter 3, we assess RAN slicing in terms of the traffic handling capacity for an Industry 4.0-inspired scenario. For the assessment, we compare a network with isolated slices and a non-sliced network. Extensive simulations show that the non-sliced network can serve more traffic than the sliced network while satisfying the same class-specific QoS requirements. Considering that RAN slicing will be adopted by the MNOs, this result highlights that additional radio resource management mechanisms are needed when RAN slicing is configured. To that end, in Chapter 4 we evaluate RAN slicing in combination with allowing slices to use idle resources of other slices, in a realistic smart city environment. The results show that idle resource sharing significantly improves the traffic performance. However, it is not until RAN slicing is further combined with other technology features, i.e. flexible numerology and mini-slots that it provides better traffic performance than non-sliced networks.
The second part of this dissertation focuses on the application of collaborative learning, and more specifically on Federated Learning (FL) in resource-constrained wireless networks. In Chapter 5, we characterise agents by their importance in the learning process and the resource efficiency of their wireless channel. Then, we provide a general agent selection framework to indicate which agents should participate in the learning process. Extensive simulations in various scenarios verify the potential of the proposed framework. Additionally, it is revealed that in scenarios where agents have small data sets or the latency requirement is stringent, it is more beneficial to perform pure learning-based agent selection. In Chapter 6 we extend the previously proposed framework to perform joint agent selection and resource allocation. We describe the problem in resource-constrained vehicular wireless networks with Multi-User Multiple Input Multiple Output (MU-MIMO) capable base stations. To approximate the optimal solution of the problem, we propose the Vehicle-Beam-Iterative (VBI) algorithm. Then, we evaluate the VBI algorithm in scenarios related to vehicular communications. The results show that in scenarios where the vehicles have the same data set sizes, the application-specific accuracy targets are achieved faster than in scenarios where the data set sizes are different. Additionally, it is shown that MU-MIMO improves the convergence time of the global FL model.
In the third part of this dissertation, the deployment of a drone swarm is addressed. In Chapter 7 we study the link density is Random Geometric Graphs (RGGs). Specifically, we very accurately approximate the link density in any two- and three-dimensional rectangular spaces with the Fréchet distribution. Then, we express the minimum number of nodes needed to ensure network connectivity in terms of the link density. Finally, we model a drone swarm with a RGG and we estimate the required size of the swarm such that communication among all drones can be ensured.
The conclusions of this dissertation and the directions for future work are presented in Chapter 8.
...
Following the trend of previous years, the number of devices, and hence the traffic in cellular networks is increasing. Moreover, new applications with stringent requirements are envisioned. Examples of such applications include collaborative learning and coverage extension with drones. To accommodate the traffic with its respective Quality of Service (QoS) requirements and to support new challenging applications in the Radio Access Network (RAN), we need to develop new algorithms and tools for efficient resource management. In this dissertation, resource management in the RAN is considered in three distinct areas.
In Chapter 2 we provide an introduction to the key concepts, which establish the technological context of the following chapters. The first part of this dissertation focuses on serving traffic with diverse requirements in the context of 5G networks. In 5G, RAN slicing has been introduced, to support services with diverse QoS requirements in the same network infrastructure. Moreover, RAN slicing allows the Mobile Network Operators (MNOs) to configure customer-specific slices. In Chapter 3, we assess RAN slicing in terms of the traffic handling capacity for an Industry 4.0-inspired scenario. For the assessment, we compare a network with isolated slices and a non-sliced network. Extensive simulations show that the non-sliced network can serve more traffic than the sliced network while satisfying the same class-specific QoS requirements. Considering that RAN slicing will be adopted by the MNOs, this result highlights that additional radio resource management mechanisms are needed when RAN slicing is configured. To that end, in Chapter 4 we evaluate RAN slicing in combination with allowing slices to use idle resources of other slices, in a realistic smart city environment. The results show that idle resource sharing significantly improves the traffic performance. However, it is not until RAN slicing is further combined with other technology features, i.e. flexible numerology and mini-slots that it provides better traffic performance than non-sliced networks.
The second part of this dissertation focuses on the application of collaborative learning, and more specifically on Federated Learning (FL) in resource-constrained wireless networks. In Chapter 5, we characterise agents by their importance in the learning process and the resource efficiency of their wireless channel. Then, we provide a general agent selection framework to indicate which agents should participate in the learning process. Extensive simulations in various scenarios verify the potential of the proposed framework. Additionally, it is revealed that in scenarios where agents have small data sets or the latency requirement is stringent, it is more beneficial to perform pure learning-based agent selection. In Chapter 6 we extend the previously proposed framework to perform joint agent selection and resource allocation. We describe the problem in resource-constrained vehicular wireless networks with Multi-User Multiple Input Multiple Output (MU-MIMO) capable base stations. To approximate the optimal solution of the problem, we propose the Vehicle-Beam-Iterative (VBI) algorithm. Then, we evaluate the VBI algorithm in scenarios related to vehicular communications. The results show that in scenarios where the vehicles have the same data set sizes, the application-specific accuracy targets are achieved faster than in scenarios where the data set sizes are different. Additionally, it is shown that MU-MIMO improves the convergence time of the global FL model.
In the third part of this dissertation, the deployment of a drone swarm is addressed. In Chapter 7 we study the link density is Random Geometric Graphs (RGGs). Specifically, we very accurately approximate the link density in any two- and three-dimensional rectangular spaces with the Fréchet distribution. Then, we express the minimum number of nodes needed to ensure network connectivity in terms of the link density. Finally, we model a drone swarm with a RGG and we estimate the required size of the swarm such that communication among all drones can be ensured.
The conclusions of this dissertation and the directions for future work are presented in Chapter 8.
In this paper, we focus on the link density in random geometric graphs (RGGs) with a distance-based connection function. After deriving the link density in D dimensions, we focus on the two-dimensional (2D) and three-dimensional (3D) space and show that the link density is accurately approximated by the Fréchet distribution, for any rectangular space. We derive expressions, in terms of the link density, for the minimum number of nodes needed in the 2D and 3D spaces to ensure network connectivity. These results provide first-order estimates for, e.g., a swarm of drones to provide coverage in a disaster or crowded area.
...
In this paper, we focus on the link density in random geometric graphs (RGGs) with a distance-based connection function. After deriving the link density in D dimensions, we focus on the two-dimensional (2D) and three-dimensional (3D) space and show that the link density is accurately approximated by the Fréchet distribution, for any rectangular space. We derive expressions, in terms of the link density, for the minimum number of nodes needed in the 2D and 3D spaces to ensure network connectivity. These results provide first-order estimates for, e.g., a swarm of drones to provide coverage in a disaster or crowded area.
In a 5G Radio Access Network (RAN), different features are offered as solutions to serve traffic with diverse characteristics and requirements, including flexible numerology, (non-)pre-emptive mini-slot based scheduling and network slicing. In this paper, we present an extensive simulation-based assessment of the relative merit of these distinct 5G features in the context of a smart city environment. We further derive the optimal feature combination and associated configuration which best handles the services related to the smart city environment given their performance requirements. The obtained insights confirm the commonly argued potential of slicing, emphasizing that the optimal configuration of the slice-specific numerology depends not only on the nature of the handled services but also on the selected RAN features. Among these features, non-preemptive mini-slot based scheduling and idle resource sharing reveal significant performance potential.
...
In a 5G Radio Access Network (RAN), different features are offered as solutions to serve traffic with diverse characteristics and requirements, including flexible numerology, (non-)pre-emptive mini-slot based scheduling and network slicing. In this paper, we present an extensive simulation-based assessment of the relative merit of these distinct 5G features in the context of a smart city environment. We further derive the optimal feature combination and associated configuration which best handles the services related to the smart city environment given their performance requirements. The obtained insights confirm the commonly argued potential of slicing, emphasizing that the optimal configuration of the slice-specific numerology depends not only on the nature of the handled services but also on the selected RAN features. Among these features, non-preemptive mini-slot based scheduling and idle resource sharing reveal significant performance potential.
Optimisation of Numerology and Packet Scheduling in 5G Networks
To Slice or not to Slice?
Network slicing has been introduced in 5G networks as an enabling feature for the effective Quality of Service (QoS) provisioning to multiple service classes with distinct performance requirements. When applied in the Radio Access Network (RAN), a class-specific slice is assigned a set of radio resources and can furthermore be optimally configured in terms of the applied numerology and packet scheduler. As both the optimal numerology and the most suitable packet scheduler may be different for e.g. a class of Latency-Constrained (LC) and a class of Throughput-Oriented (TO) services, the potential of slicing is clear. However, the inherent trunking loss incurred when applying slicing with dedicated resources provides an argument against such slicing. In this paper we demonstrate that the performance and traffic handling capacity in an optimally configured non-sliced scenario may exceed that attained when using segregated individually optimised slices. To that end, we use simulations to assess the best-performing numerology and packet scheduler for a sliced scenario with LC and TO services. We then compare the thus optimised sliced scenario with an optimal non-sliced scenario and show that the non-sliced scenario can serve about 20% more traffic than the sliced scenario while satisfying the same class-specific QoS requirements.
...
Network slicing has been introduced in 5G networks as an enabling feature for the effective Quality of Service (QoS) provisioning to multiple service classes with distinct performance requirements. When applied in the Radio Access Network (RAN), a class-specific slice is assigned a set of radio resources and can furthermore be optimally configured in terms of the applied numerology and packet scheduler. As both the optimal numerology and the most suitable packet scheduler may be different for e.g. a class of Latency-Constrained (LC) and a class of Throughput-Oriented (TO) services, the potential of slicing is clear. However, the inherent trunking loss incurred when applying slicing with dedicated resources provides an argument against such slicing. In this paper we demonstrate that the performance and traffic handling capacity in an optimally configured non-sliced scenario may exceed that attained when using segregated individually optimised slices. To that end, we use simulations to assess the best-performing numerology and packet scheduler for a sliced scenario with LC and TO services. We then compare the thus optimised sliced scenario with an optimal non-sliced scenario and show that the non-sliced scenario can serve about 20% more traffic than the sliced scenario while satisfying the same class-specific QoS requirements.