Zdenek Becvar

In this paper, we focus on a cache-enabled networks, where a content requested by user equipments (UEs) can be delivered by means of multi-hop relaying via unmanned aerial vehicles (UAVs) and/or via other UEs exploiting device-to-device (D2D) communication. For such scenario, we optimize transmission power allocation, bandwidth allocation, and route selection to minimize the content delivery duration. The formulated problem is a mixed-integer nonlinear programming (MINLP) problem and, thus, it is hard to solve. Hence, we first solve the power and bandwidth allocation by a cooperative game among all transmitting nodes involved in the content delivery process and we prove that this game is a potential game. Then, we introduce a fast-converging iterative algorithm employing an approximate better-response mechanism for the resource allocation (i.e., power and bandwidth). After that, we design a low-complexity greedy algorithm jointly handling route selection and power and bandwidth allocation. The algorithm selects transmission routes to deliver the requested contents to the UEs and, after each selection, the resource allocation at the transmitting nodes involved in the content delivery process (i.e., GBS, UAVs, and relaying UEs) is updated. The simulation results demonstrate that the proposed scheme reduces the average content delivery duration by up to 23.2% compared to the best-performing benchmark algorithm.

In this letter, we propose a coordinated federated learning framework for joint optimization of multiple radio resource management parameters in wireless networks. Like in a traditional federated learning, each user equipment (UE) trains a local machine learning model and shares parameters of the local models. However, the proposed approach also involves the communication-related aspects in the process of the federated learning and considers mutual dependencies among optimized variables. To this end, the UEs also share values of a new designed local loss functions specifically tailored to capture performance of the communication network to enable a local coordination for capturing mutual relation between optimized parameters. Then, base stations (BSs) leverage the received values of the new loss functions to produce a weighted global machine learning model. The weighting is based on the loss values so that the models of the UEs with low loss values are prioritized, since the low loss values indicate a proper setting of radio resource management parameters. The simulation results demonstrate a significant improvement from 10.6% to 18.9% and from 3.1% to 11.2% in sum capacity and ratio of UEs satisfied with the capacity, respectively, compared to the best performing state-of-the-art federated learning solution.

Low-latency Vehicular Edge Computing (VEC) applications require an efficient VEC resource allocation considering all components contributing to application latency, i.e., computation, communication, and storage. While the optimization of communication and computation resources is broadly addressed in literature, storage, a significant source of latency in the computing stack, is often ignored in existing works. Thus, in this paper, we optimize the communication and computation resources together with the storage resources to minimize the latency of VEC applications. The problem of jointly minimizing communication, computation, and storage latency under practical constraints is NP-hard. Hence, we employ dual decomposition and Lagrangian relaxation to achieve a computationally viable solution for the joint communication, computing, and storage resource allocation to VEC applications. To this end, we define a dual problem of the assignment of VEC applications to base stations. This problem corresponds to the perfect matching problem in a weighted bipartite graph and optimally solvable by algorithms with polynomial computation complexity. Then, as the solution to the dual problem may violate some constraints of the main resource allocation problem, we find a feasible solution to the main resource allocation problem using Lagrangian relaxation. We show that the joint optimization of all three aspects, i.e., communication, computation, and storage, reduces the overall offloading latency up to 60% compared to state-of-the-art works.

Hyperconverged Infrastructures (HCIs) combine processing and storage elements to meet the requirements of data-intensive applications in performance, scalability, and quality of service. As an emerging paradigm, HCI should couple with a variety of traditional performance improvement approaches such as I/O caching in virtualized platforms. Contemporary I/O caching schemes are optimized for traditional single-node storage architectures and suffer from two major shortcomings for multi-node architectures: a) imbalanced cache space requirement and b) imbalanced I/O traffic and load. This makes existing schemes inefficient in distributing cache resources over an array of separate physical nodes. In this paper, we propose an Efficient and Load Balanced I/O Cache Architecture (ELICA), managing the solid-state drive (SSD) cache resources across HCI nodes to enhance I/O performance. ELICA dynamically reconfigures and distributes the SSD cache resources throughout the array of HCI nodes and also balances the network traffic and I/O cache load by dynamic reallocation of cache resources. To maximize the performance, we further present an optimization problem defined by Integer Linear Programming to efficiently distribute cache resources and balance the network traffic and I/O cache relocations. Our experimental results on a real platform show that ELICA improves quality of service in terms of average and worst-case latency in HCIs by 3.1× and 23%, respectively, compared to the state-of-the-art.

We develop a framework to simultaneously optimize the artificial noise (AN) covariance of eavesdroppers, the high-dimensional continuous phase shifts of Unmanned Aerial Vehicles (UAV)-borne reconfigurable intelligent surfaces (RISs), and beamforming of base stations (BSs) in vehicular networks. This optimization aims to maximize the worst-case secrecy rate for vehicular users (VUs) amidst eavesdroppers considering inaccurate channel state information (CSI) and VUs' quality-of-service (QoS) requirements. We propose a predictive localization-based actor-critic deep reinforcement learning (DRL) solution. We employ a deep neural network (DNN) to predict the future spatial location of VUs and eavesdroppers, since these are apriory unknown. Then, the predicted positions are fed to a Deep Deterministic Policy Gradient (DDPG) algorithm, which determines the AN matrix and RIS beamforming vector. The DDPG integrated with DNN-based predictive localization offers a significant advantage over the standard DDPG, as additional states are incorporated into the system, thereby enhancing the model's capability to coordinate DDPG based on predicted location and to capture more complex and abstract features of the highly dynamic vehicular network. Simulations demonstrate the proposed DDPG with predictive localization ensures secure and reliable communications, surpassing state-f-the-art baselines.

Mobility management and radio resource management are critical aspects of device-to-device (D2D) communication due to the coexistence of cellular user equipments (CUEs) and D2D user equipments (DUEs) operating within the same spectrum. Therefore, in this paper, we target a problem of D2D communication for mobile users and we address it via joint selection of communication mode (i.e., cellular mode, relay mode, D2D mode), selection of relay, and allocation of resources to maximize the sum rate of D2D pairs while ensuring quality of service for all CUEs. As the formulated problem is a binary integer nonlinear programming problem, which is of very high complexity, we first develop a preliminary greedy approach virtually pre-selecting only mode and relay for each D2D pair. Then, we enhance the preliminary approach towards a comprehensive greedy algorithm for joint mode selection, relay selection, and resource allocation. We also address implementation aspects, including complexity, channel quality acquisition, and signaling message exchange for the management of the proposed solution for mobile users. The simulations demonstrate that our proposed scheme increases the sum rate of D2D pairs by approximately 20% if compared to the best performing related work.

We focus on handover management in networks integrating traditional terrestrial ground base stations (GBSs) and non-terrestrial unmanned aerial vehicles (UAVs) serving users. In such scenario, we propose a joint management of handover of users equipment (UEs) between UAVs and GBSs as well as handover of UAVs between GBSs. Our goal is to maximize the sum capacity for UEs while avoiding redundant handovers. As an impact of handover on the future network performance is not explicit, we adopt deep deterministic policy gradient (DDPG). Furthermore, since the UAVs are usually energy constrained, we consider an energy efficient transparent relaying mode for the UAVs. However, in the transparent relaying mode, the access channel quality between the UAV relay and the UE is unknown even if such information is essential for handover. Thus, we further employ deep neural network (DNN) to predict the access channel quality. An incorporation of DDPG for handover management together with DNN for channel quality prediction can impair the sum capacity of the UEs due to an accumulation of inherent small prediction errors of DNN and DDPG. Hence, we also introduce a coordination between DNN and DDPG to suppress the accumulation of the prediction errors. Simulations demonstrate that the proposed handover management and the coordination of DDPG and DNN increase the sum capacity by up to 63% while notably reducing the number of handovers and handover failure ratio compared to the state-of-the-art works.

Computer vision plays a crucial role in enabling connected autonomous vehicles (CAVs) to observe and comprehend their surroundings. The computer vision tasks are typically based on convolutional neural networks (CNNs). However, CNNs often require significant processing power. Techniques like early exiting and split computing enhance CNN task execution latency and adaptability to varying environmental conditions. Since the split computing introduces additional overhead for offloading of the task from the CAV to an edge servers, we incorporate multiple autoencoders within each split point to enhance the adaptability of splitting under varying environmental conditions. However, the autoencoders introduce an additional layer of complexity related to the selection of the optimal compression strategy alongside the splitting and exiting decisions. To tackle this challenge, we introduce a novel approach based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. This algorithm dynamically and jointly determines the most suitable exit point, split point, and autoencoder. Furthermore, the MADDPG-based approach considers other CAVs when selecting action, promoting cooperation among CAVs. Our results demonstrate that the proposed approach reduces latency up to 44.4% while maintaining at least comparable or even higher accuracy of the computed vision outcome compared to the state-of-the-art solutions.