IRIgor Radusinović

Igor Radusinović

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Igor Radusinović has a total of 7 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Asynchronous Low Duty-Cycling for Reduced Event-Reporting Delay in Underwater Acoustic Sensor Networks

    by Slavica Tomović, Benjamin Sherlock, Jeffrey Neasham, Igor Radusinović

    Event-driven Underwater Acoustic Sensor Networks (UASNs) face the critical challenge of minimizing event-reporting delay while maintaining energy efficiency in resource-constrained underwater environments. This paper presents a novel asynchronous duty-cycling communication protocol integrating anycast forwarding and collision avoidance mechanisms to address this challenge without requiring energy-intensive synchronization overhead. The proposed protocol enables nodes to perform independent sleep-wake scheduling while opportunistically leveraging multiple forwarding candidates to minimize transmission latency. To handle packet collisions inherent in asynchronous networks, we developed a specialized MAC-layer collision avoidance mechanism optimized for underwater acoustic communication characteristics, specifically accounting for long propagation delays and effectively mitigating collisions when multiple nodes detect events in close temporal proximity or when multiple forwarding candidates wake up concurrently. The protocol employs a sender-initiated approach that allows nodes to remain silent until events occur, thereby reducing acoustic emissions and making the network less susceptible to detection. We validate the protocol through a comprehensive evaluation comprising hardware implementation, hardware-in-the-loop simulations, and sea trials in the Adriatic Sea. Comparative analysis against a state-of-the-art benchmark demonstrates significantly lower event-reporting delay and reduced per-event energy consumption under equivalent duty-cycle conditions. Experimental results further confirm efficient power management and reliable data delivery even in highly challenging underwater acoustic environments, making the protocol particularly suitable for long-term covert underwater monitoring applications where both performance and stealth are essential.

  • Open Access

    BEKMP: A Blockchain-Enabled Key Management Protocol for Underwater Acoustic Sensor Networks

    by Slavica Tomović, Bogdan Krivokapić, Ðula Nađ, Igor Radusinović

    This paper introduces a new blockchain-assisted key management protocol specifically designed for clustered Underwater Acoustic Sensor Networks (UASNs). The protocol focuses on simplifying cross-cluster reauthentication for mobile underwater devices and reducing the risks of internal attacks caused by compromised nodes. To address the resource limitations of UASN devices, the protocol utilizes Elliptic Curve Qu Vanstone (ECQV) certificates, which considerably reduce Public Key Infrastructure (PKI) overhead. Moreover, it integrates the Hashed One-pass Menezes-QuVanstone (HOMQV) protocol with blockchain technology to enhance key exchange security against active attacks. The blockchain is hosted on resource-intensive surface nodes that serve as cluster heads and decentralized authorities for certificate management. Smart contracts embedded within the blockchain facilitate certificate-related operations and node trust scoring mechanism. We compared the proposed solution against a state-of-the-art benchmark and demonstrated that it imposes lower computational and communication overheads while exhibiting robustness against various attacks, as confirmed by the AVISPA tool. The solution is implemented on actual underwater/surface devices and validated experimentally. By leveraging the immutable, traceable, and fault-tolerant properties of blockchain, it establishes a secure, scalable, and efficient communication framework for mission-critical UASN applications.


  • Open Access

    DR-ALOHA-Q: A Q-Learning-Based Adaptive MAC Protocol for Underwater Acoustic Sensor Networks

    by Slavica Tomovic, Igor Radusinovic

    Underwater acoustic sensor networks (UASNs) are challenged by the dynamic nature of the underwater environment, large propagation delays, and global positioning system (GPS) signal unavailability, which make traditional medium access control (MAC) protocols less effective. These factors limit the channel utilization and performance of UASNs, making it difficult to achieve high data rates and handle congestion. To address these challenges, we propose a reinforcement learning (RL) MAC protocol that supports asynchronous network operation and leverages large propagation delays to improve the network throughput.he protocol is based on framed ALOHA and enables nodes to learn an optimal transmission strategy in a fully distributed manner without requiring detailed information about the external environment. The transmission strategy of sensor nodes is defined as a combination of time-slot and transmission-offset selection. By relying on the concept of learning through interaction with the environment, the proposed protocol enhances network resilience and adaptability. In both static and mobile network scenarios, it has been compared with the state-of-the-art framed ALOHA for the underwater environment (UW-ALOHA-Q), carrier-sensing ALOHA (CS-ALOHA), and delay-aware opportunistic transmission scheduling (DOTS) protocols. The simulation results show that the proposed solution leads to significant channel utilization gains, ranging from 13% to 106% in static network scenarios and from 23% to 126% in mobile network scenarios.oreover, using a more efficient learning strategy, it significantly reduces convergence time compared to UW-ALOHA-Q in larger networks, despite the increased action space.


  • Open Access

    A comparative study of deep learning and decision tree based ensemble learning algorithms for network traffic identification

    by Nedeljko Nikolić, Slavica Tomović, Igor Radusinović

    In this paper, we apply Deep Learning (DL) and decision-tree-based ensemble learning algorithms to classify network traffic by application. Various Deep Learning (DL) models for network traffic identification have been presented, implemented and compared, including 1D convolutional, stacked autoencoder, multi-layer perceptron, and combination of the aforementioned. Then the results of DL models have been compared to those obtained with two popular ensemble learning models based on decision trees-Random Forest and XGBoost. To train and test the classification models, a dataset containing both encrypted and unencrypted traffic has been collected in a real network, under normal operating conditions, and pre-processed in a way that ensures non-biased results. The classification uncertainties of the models have been also quantified on publicly available ISCX VPN-nonVPN dataset. The models have been compared in terms of precision, recall, F1 score and accuracy, for different levels of complexity and training dataset sizes. The evaluation results indicate that the decision-tree ensemble learning algorithms provide more accurate results and outperform the DL algorithms. The performance gap reduces with the dataset complexity.

  • Dynamic spectrum access with deep Q-learning in densely occupied and partially observable environments

    by Slavica Tomović, Igor Radusinović

    In this paper, we propose a new Dynamic Spectrum Access (DSA) method for multi-channel wireless networks. We assume that DSA nodes do not have prior knowledge of the system dynamics, and have only partial observability of the channels. Thus, the problem is formulated as a Partially Observable Markov Decision Process (POMDP) with exponential time complexity. We have developed a novel Deep Reinforcement Learning (DRL) based DSA method which combines a double deep Q-learning architecture with a recurrent neural network and takes advantage of a prioritized experience buffer. The simulation analysis shows that the proposed method accurately predicts a channel state based on the fixed-length history of partial observations. Compared with other DRL methods for DSA, the proposed solution can find a near-optimal policy in a smaller number of iterations and suits a wider range of communication environments, including dynamic ones, where channel occupancy pattern changes over time. The performance improvement increases with the number of channels and with a channel state transition uncertainty. To boost the performance of the algorithm in densely occupied environments, multiple DRL exploration strategies are examined and evaluation results are presented in the paper.


  • Toward a Scalable, Robust, and QoS-Aware Virtual-Link Provisioning in SDN-Based ISP Networks

    by Slavica Tomovic, Igor Radusinovic

    In this paper, we propose a control framework for virtual-link provisioning in SDN-based ISP networks. Within the problem’s scope, we have analysed the traffic engineering (TE), Quality of Service (QoS) provisioning, network stability and failure recovery issues. The proposed solution aims to optimize the network throughput subject to multiple QoS constraints and link-failure robustness constraints that are defined in service level agreements (SLAs), while minimizing the time-average rate of routing updates according to ISP preferences. In contrast to the conventional TE mechanisms which periodically optimize all the routing rules, our solution is based on a bi-objective optimization model which jointly optimizes the routing and the number of flow-table updates. In this way, the reconfiguration overhead of TE is reduced, thus the SDN controller could perform TE more frequently, in order to quickly and efficiently respond to traffic and topology changes. To generate a set of Pareto optimal solutions for the proposed optimization model we use the augmented -constrained method. The most efficient trade-off solution from the Pareto set is chosen by the Lypunov drift-plus-penalty algorithm. In order to make the proposed approach scalable enough for large network topologies, we also propose a heuristic TE algorithm that runs in polynomial time. The proposed solution is validated via simulations and experimentally, and it is shown that it outperforms conventional periodic TE techniques. The advantages of our proposal over a state-of-the-art TE technique in terms of the capability to adjust the reconfiguration rate based on the threshold set by ISP is demonstrated. In the analysis, a special attention is paid to the impact of proactive failure protection mechanisms on the network service capacity.


  • RO-RO: Routing Optimality - Reconfiguration Overhead Balance in Software-Defined ISP Networks

    by Slavica Tomovic, Igor Radusinovic

    Software-defined networking provides a promising solution for traffic engineering (TE) by utilizing a centralized controller to remotely configure programmable network switches. In order to achieve optimal traffic distribution and meet the Quality of Service (QoS) requirements, the SDN controller must frequently solve complex optimization models. Applying an optimization model in ISP (Internet Service Provider) networks implies abundant routing reconfigurations, which adversely impact the network stability and QoS parameters. Namely, since routing updates usually cannot be applied on all switches at the same time, transient congestion and loops often happen as a result. Here, in this paper, we propose a new control framework that strives to maximize the network throughput and provide QoS with minimal reconfiguration costs. In contrast to the conventional TE approaches, which perform the network optimizations periodically and control the side effects of reconfigurations by carefully choosing the period length between the optimization cycles, we propose a multi-objective optimization model which jointly minimizes the routing cost function and the reconfiguration overhead. The Pareto frontier of the optimization model is generated by the augmented ∈-constrained method, whereas a Lyapunov drift-plus-penalty function is used to select the best compromise solution from the Pareto set. Since the reconfiguration overhead is reduced, the network controller could be allowed to optimize resource allocation more frequently, in order to respond quickly and efficiently to the network changes. Considering that the proposed optimization model is computationally intractable in large-scale networks, we also propose a heuristic algorithm to efficiently solve large instances of the problem. In simulations and Mininet experiments, we show that our solution brings performance improvement over the conventional periodic TE techniques. Moreover, the proposed heuristic achieves better QoS request acceptance ratio than the state-of-the-art multi-objective optimization solution, despite the significantly reduced computational complexity.


  • Open Access

    Dynamic optimization of load-balancing and reconfiguration overhead in SD-ISP networks

    by Slavica Tomovic, Igor Radusinovic

    In this paper, we propose a new traffic engineering (TE) approach for software-defined Internet Service Provider (SD-ISP) networks that strives to maximize the network throughput and provide adequate QoS (Quality of Service) with a minimal reconfiguration cost. In contrast to the conventional TE approaches, which perform the network optimizations periodically and control the side effects of reconfigurations by carefully choosing the period length between the optimization cycles, we use a bi-objective optimization model that minimizes maximum link utilization and the reconfiguration overhead. A new heuristic algorithm has been proposed in order to generate the approximated Pareto frontier for the bi-objective optimization model, while the Lyapunov drift-plus-penalty algorithm is used to select the most appropriate solution from the approximated Pareto set. Our simulation study shows that the proposed approach adjusts to the ISP's constraint on time-average reconfiguration rate by trading the throughput performance efficiently. Since the reconfiguration overhead is reduced, the network controller could be allowed to optimize resource allocation more frequently, in order to quickly and efficiently respond to the network changes. The paper analyses the impact of the reconfiguration rate constraint, average flow duration and the frequency of TE on the overall network performance.


  • Exhaustive Limited Service Round Robin Matching Algorithm for Buffered Crossbar Switches

    by I. Radusinovic, Z. Veljovic

    In this paper a high performance and simple scheduling algorithm for combined input-crosspoint crossbar switches, called exhaustive round-robin with limited service (ELRR) is presented and analyzed. We propose using of this scheduling system for arbitration at inputs and crosspoints. If the virtual output queue (crosspoint buffer) becomes empty, the input (crosspoint) arbiter updates its pointer to the next location in a fixed order. Otherwise, the pointer remains at the current virtual output queue (crosspoint buffer). To avoid unfairness under some extremely unbalanced traffic pattern limited service matching is presented. It is shown that this new solution offers high performance for several admissible traffic patterns using only one-cell crosspoint buffers.