Elmin Marevac
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Elmin Marevac has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Esad Kadušić
Faculty of Educational Sciences, University of Sarajevo, 71000 Sarajevo, Bosnia and Herzegovina
7 shared publications - Natasa Zivic
Faculty of Digital Transformation, Leipzig University of Applied Sciences, 04277 Leipzig, Germany
7 shared publications - Christoph Ruland
Chair of Digital Communication Systems, University of Siegen, 57076 Siegen, Germany
4 shared publications Selman Patković
Polytechnic Faculty of the University of Zenica
3 shared publicationsChristoph Ruland
Institute for Data Comm. Systems, University of Siegen,Siegen,Germany
2 shared publicationsEsad Kadusic
University of Sarajevo,Faculty of Educational Sciences,Sarajevo,Bosnia and Herzegovina
2 shared publications
Latest Publications
- Open Access
- 30 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (21) , 5032
GC-Assisted Runtime Integrity Verification for Heap-Resident Bytecode in an FPGA-Based Virtual Machine
Dynamic-language virtual machines store compiled functions as garbage-collected heap objects, yet embedded and field-programmable gate array (FPGA) integrity mechanisms check code only at load time or ignore whether memory holds live code. picoStella, a proof-of-concept FPGA-targeted virtual machine, reauthenticates live code in every garbage-collection pass: after marking, a verification pass uses the mark bits to re-hash each reachable function with SHA3-256 (Secure Hash Algorithm 3) against a write-once trusted digest. The design passed 36 testbenches with identical results across two simulators, was synthesized for a Cyclone IV E FPGA, and its verifier was tested on a DE2-115 board. A closed-form model predicts per-function cost exactly; verification adds 167–8914% to the mark-sweep cost for up to eight 256-byte functions. Reusing the collector’s reachability saves 0.40% over a verifier with its own traversal, but up to 83% over a liveness-oblivious scrubber. Detection takes 42–314 cycles, all 5770 injected bytecode faults were detected, and board-level debug-register injection reproduced these latencies. On-chip clock glitching defeated detection in 4.5% of corrupted trials. Register-level soft-error injection shows SHA3-state upsets fail safe, whereas control or metadata upsets can hang a pass or skip a function. The time-of-check-to-time-of-use window is narrowed, not closed; the reset-cleared lock is not a root of trust; larger-scale costs are projections.
- 30 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (21) , 5032
Open AccessGC-Assisted Runtime Integrity Verification for Heap-Resident Bytecode in an FPGA-Based Virtual Machine
Dynamic-language virtual machines store compiled functions as garbage-collected heap objects, yet embedded and field-programmable gate array (FPGA) integrity mechanisms check code only at load time or ignore whether memory holds live code. picoStella, a proof-of-concept FPGA-targeted virtual machine, reauthenticates live code in every garbage-collection pass: after marking, a verification pass uses the mark bits to re-hash each reachable function with SHA3-256 (Secure Hash Algorithm 3) against a write-once trusted digest. The design passed 36 testbenches with identical results across two simulators, was synthesized for a Cyclone IV E FPGA, and its verifier was tested on a DE2-115 board. A closed-form model predicts per-function cost exactly; verification adds 167–8914% to the mark-sweep cost for up to eight 256-byte functions. Reusing the collector’s reachability saves 0.40% over a verifier with its own traversal, but up to 83% over a liveness-oblivious scrubber. Detection takes 42–314 cycles, all 5770 injected bytecode faults were detected, and board-level debug-register injection reproduced these latencies. On-chip clock glitching defeated detection in 4.5% of corrupted trials. Register-level soft-error injection shows SHA3-state upsets fail safe, whereas control or metadata upsets can hang a pass or skip a function. The time-of-check-to-time-of-use window is narrowed, not closed; the reset-cleared lock is not a root of trust; larger-scale costs are projections.
- Open Access
- 10 July 2026
- Published by MDPI AG in Electronics
- Vol. 15 (14) , 3043
A Low-Latency Embedded Inertial Motion Tracking System for Real-Time Applications
Inertial motion sensing plays an important role in real-time human–machine interaction applications, including interactive systems, virtual environments, and motion-controlled interfaces, where low latency and accurate motion tracking are important. This paper presents the design and implementation of an embedded inertial sensing system that acquires, processes, and transmits motion data from consumer-grade inertial measurement units (IMUs) under real-time constraints. Rather than introducing a new sensor fusion algorithm, the contribution lies in a systems-oriented methodology comprising a predictive clock-advancement mechanism that prevents cumulative timing drift, an automated matrix-based calibration procedure for hardware-agnostic deployment, and a benchmarking framework for end-to-end real-time system evaluation. Implemented on a resource-constrained embedded platform, the framework integrates sensor acquisition, lightweight filtering, sensor fusion, and real-time orientation estimation within a single processing pipeline. Motion data are transmitted using the CemuHook UDP (User Datagram Protocol) motion protocol (DSU) to demonstrate interoperability with existing motion-control software while maintaining low end-to-end latency and stable throughput. Experimental results show stable sampling frequency, low communication latency, accurate orientation estimation, and low computational overhead. The presented system provides an embedded inertial sensing framework that can be adapted to a range of real-time motion-sensing applications beyond the communication protocol used for demonstration.
- 10 July 2026
- Published by MDPI AG in Electronics
- Vol. 15 (14) , 3043
Open AccessA Low-Latency Embedded Inertial Motion Tracking System for Real-Time Applications
Inertial motion sensing plays an important role in real-time human–machine interaction applications, including interactive systems, virtual environments, and motion-controlled interfaces, where low latency and accurate motion tracking are important. This paper presents the design and implementation of an embedded inertial sensing system that acquires, processes, and transmits motion data from consumer-grade inertial measurement units (IMUs) under real-time constraints. Rather than introducing a new sensor fusion algorithm, the contribution lies in a systems-oriented methodology comprising a predictive clock-advancement mechanism that prevents cumulative timing drift, an automated matrix-based calibration procedure for hardware-agnostic deployment, and a benchmarking framework for end-to-end real-time system evaluation. Implemented on a resource-constrained embedded platform, the framework integrates sensor acquisition, lightweight filtering, sensor fusion, and real-time orientation estimation within a single processing pipeline. Motion data are transmitted using the CemuHook UDP (User Datagram Protocol) motion protocol (DSU) to demonstrate interoperability with existing motion-control software while maintaining low end-to-end latency and stable throughput. Experimental results show stable sampling frequency, low communication latency, accurate orientation estimation, and low computational overhead. The presented system provides an embedded inertial sensing framework that can be adapted to a range of real-time motion-sensing applications beyond the communication protocol used for demonstration.
- Open Access
- 3 May 2026
- Published by MDPI AG in Cryptography
- Vol. 10 (3) , 30
A Hybrid Module-LWE and Hash-Based Framework for Memory-Efficient Post-Quantum Key Encapsulation
Deploying post-quantum cryptography on highly constrained devices remains challenging due to the large key sizes and substantial storage and memory-traffic demands of leading lattice-based schemes. Although constructions such as Kyber, Dilithium, and NTRU offer strong resistance against quantum adversaries, their multi-kilobyte public keys and intensive memory access patterns limit practical adoption in microcontrollers, smart cards, and low-power edge environments. This work proposes a hybrid key-encapsulation mechanism that integrates a compact, seed-generated Module-LWE structure with a quantum-secure hash-based authentication layer. The design employs a small public seed to instantiate lattice matrices on demand via a lightweight pseudorandom generator and incorporates a Merkle-tree commitment to represent compressed auxiliary error information. Additional design considerations—including sparsity-aware secret keys, SIMD-friendly polynomial operations, and cache-efficient decryption paths—are intended to reduce runtime memory usage and computational overhead. The security of the proposed construction is analysed under both Module-LWE and hash-based one-way assumptions, with further consideration of constant-time execution and cache-line alignment to mitigate side-channel risks. This hybrid approach outlines a design pathway toward post-quantum key-encapsulation mechanisms suitable for deployment on memory-limited and energy-constrained platforms.
- 3 May 2026
- Published by MDPI AG in Cryptography
- Vol. 10 (3) , 30
Open AccessA Hybrid Module-LWE and Hash-Based Framework for Memory-Efficient Post-Quantum Key Encapsulation
Deploying post-quantum cryptography on highly constrained devices remains challenging due to the large key sizes and substantial storage and memory-traffic demands of leading lattice-based schemes. Although constructions such as Kyber, Dilithium, and NTRU offer strong resistance against quantum adversaries, their multi-kilobyte public keys and intensive memory access patterns limit practical adoption in microcontrollers, smart cards, and low-power edge environments. This work proposes a hybrid key-encapsulation mechanism that integrates a compact, seed-generated Module-LWE structure with a quantum-secure hash-based authentication layer. The design employs a small public seed to instantiate lattice matrices on demand via a lightweight pseudorandom generator and incorporates a Merkle-tree commitment to represent compressed auxiliary error information. Additional design considerations—including sparsity-aware secret keys, SIMD-friendly polynomial operations, and cache-efficient decryption paths—are intended to reduce runtime memory usage and computational overhead. The security of the proposed construction is analysed under both Module-LWE and hash-based one-way assumptions, with further consideration of constant-time execution and cache-line alignment to mitigate side-channel risks. This hybrid approach outlines a design pathway toward post-quantum key-encapsulation mechanisms suitable for deployment on memory-limited and energy-constrained platforms.
- Open Access
- 1 January 2026
- Published by Engineering and Technology Publishing in Journal of Advances in Information Technology
- Vol. 17 (6) , 1188-1210
Accelerated Quantum Algorithm Prototyping: Modular Simulation and Noise Modelling with CUDA Support
This work presents a portable, modular quantum-circuit emulator designed to accelerate algorithm prototyping by leveraging Central Processing Unit (CPU) and Compute Unified Device Architecture (CUDA)-enabled Graphic Processing Unit (GPU) backends for dense linear-algebra workloads. The emulator provides a lightweight backend abstraction that unifies NumPy and CuPy, supports both statevector and density-matrix representations, and implements efficient multi-target unitary embedding and Kraus-map noise channels. These capabilities enable rapid exploration of quantum algorithms that require full 2n × 2n operators, such as quantum phase estimation, full Quantum Fourier Transform (QFT), and noise-aware variational circuits, while minimizing host/device transfer and preserving numerical parity between backends. Implementation strategies are described for embedding single- and multi-qubit gates, parameterized rotations, and controlled unitaries, along with an “apply_unitary_on_targets” routine that employs reshape/matrix-multiply/reshape patterns to map arbitrary target sets to dense GPU kernels for high throughput. Using a curated benchmark suite of dense quantum workloads and microbenchmarks, including large matrix multiplies and Fast Fourier Transforms (FFT), performance is quantified across a range of qubit counts and precisions. The results demonstrate substantial speedups on modern NVIDIA GPUs for workloads dominated by dense linear algebra, identify cross-over points where GPU acceleration becomes beneficial, and analyse memory/precision trade-offs for complex64 versus complex128. The presented emulator lowers the barrier for noise-aware algorithm prototyping by combining practical software ergonomics with GPU performance, enabling faster iteration on algorithm design and noise-mitigation strategies on commodity single-node hardware. While the abstraction layer ensures portability, achieving peak performance currently requires hardware-specific tuning—a limitation we address through future automation strategies.
- 1 January 2026
- Published by Engineering and Technology Publishing in Journal of Advances in Information Technology
- Vol. 17 (6) , 1188-1210
Open AccessAccelerated Quantum Algorithm Prototyping: Modular Simulation and Noise Modelling with CUDA Support
This work presents a portable, modular quantum-circuit emulator designed to accelerate algorithm prototyping by leveraging Central Processing Unit (CPU) and Compute Unified Device Architecture (CUDA)-enabled Graphic Processing Unit (GPU) backends for dense linear-algebra workloads. The emulator provides a lightweight backend abstraction that unifies NumPy and CuPy, supports both statevector and density-matrix representations, and implements efficient multi-target unitary embedding and Kraus-map noise channels. These capabilities enable rapid exploration of quantum algorithms that require full 2n × 2n operators, such as quantum phase estimation, full Quantum Fourier Transform (QFT), and noise-aware variational circuits, while minimizing host/device transfer and preserving numerical parity between backends. Implementation strategies are described for embedding single- and multi-qubit gates, parameterized rotations, and controlled unitaries, along with an “apply_unitary_on_targets” routine that employs reshape/matrix-multiply/reshape patterns to map arbitrary target sets to dense GPU kernels for high throughput. Using a curated benchmark suite of dense quantum workloads and microbenchmarks, including large matrix multiplies and Fast Fourier Transforms (FFT), performance is quantified across a range of qubit counts and precisions. The results demonstrate substantial speedups on modern NVIDIA GPUs for workloads dominated by dense linear algebra, identify cross-over points where GPU acceleration becomes beneficial, and analyse memory/precision trade-offs for complex64 versus complex128. The presented emulator lowers the barrier for noise-aware algorithm prototyping by combining practical software ergonomics with GPU performance, enabling faster iteration on algorithm design and noise-mitigation strategies on commodity single-node hardware. While the abstraction layer ensures portability, achieving peak performance currently requires hardware-specific tuning—a limitation we address through future automation strategies.
- Open Access
- 4 November 2025
- Published by MDPI AG in Future Internet
- Vol. 17 (11) , 508
Data-Driven Predictive Analytics for Dynamic Aviation Systems: Optimising Fleet Maintenance and Flight Operations Through Machine Learning
The aviation industry operates as a complex, dynamic system generating vast volumes of data from aircraft sensors, flight schedules, and external sources. Managing this data is critical for mitigating disruptive and costly events such as mechanical failures and flight delays. This paper presents a comprehensive application of predictive analytics and machine learning to enhance aviation safety and operational efficiency. We address two core challenges: predictive maintenance of aircraft engines and forecasting flight delays. For maintenance, we utilise NASA’s C-MAPSS simulation dataset to develop and compare models, including one-dimensional convolutional neural networks (1D CNNs) and long short-term memory networks (LSTMs), for classifying engine health status and predicting the Remaining Useful Life (RUL), achieving classification accuracy up to 97%. For operational efficiency, we analyse historical flight data to build regression models for predicting departure delays, identifying key contributing factors such as airline, origin airport, and scheduled time. Our methodology highlights the critical role of Exploratory Data Analysis (EDA), feature selection, and data preprocessing in managing high-volume, heterogeneous data sources. The results demonstrate the significant potential of integrating these predictive models into aviation Business Intelligence (BI) systems to transition from reactive to proactive decision-making. The study concludes by discussing the integration challenges within existing data architectures and the future potential of these approaches for optimising complex, networked transportation systems.
- 4 November 2025
- Published by MDPI AG in Future Internet
- Vol. 17 (11) , 508
Open AccessData-Driven Predictive Analytics for Dynamic Aviation Systems: Optimising Fleet Maintenance and Flight Operations Through Machine Learning
The aviation industry operates as a complex, dynamic system generating vast volumes of data from aircraft sensors, flight schedules, and external sources. Managing this data is critical for mitigating disruptive and costly events such as mechanical failures and flight delays. This paper presents a comprehensive application of predictive analytics and machine learning to enhance aviation safety and operational efficiency. We address two core challenges: predictive maintenance of aircraft engines and forecasting flight delays. For maintenance, we utilise NASA’s C-MAPSS simulation dataset to develop and compare models, including one-dimensional convolutional neural networks (1D CNNs) and long short-term memory networks (LSTMs), for classifying engine health status and predicting the Remaining Useful Life (RUL), achieving classification accuracy up to 97%. For operational efficiency, we analyse historical flight data to build regression models for predicting departure delays, identifying key contributing factors such as airline, origin airport, and scheduled time. Our methodology highlights the critical role of Exploratory Data Analysis (EDA), feature selection, and data preprocessing in managing high-volume, heterogeneous data sources. The results demonstrate the significant potential of integrating these predictive models into aviation Business Intelligence (BI) systems to transition from reactive to proactive decision-making. The study concludes by discussing the integration challenges within existing data architectures and the future potential of these approaches for optimising complex, networked transportation systems.
- Open Access
- 10 September 2025
- Published by MDPI AG in Algorithms
- Vol. 18 (9) , 572
Multimodal Video Summarization Using Machine Learning: A Comprehensive Benchmark of Feature Selection and Classifier Performance
The exponential growth of user-generated video content necessitates efficient summarization systems for improved accessibility, retrieval, and analysis. This study presents and benchmarks a multimodal video summarization framework that classifies segments as informative or non-informative using audio, visual, and fused features. Sixty hours of annotated video across ten diverse categories were analyzed. Audio features were extracted with pyAudioAnalysis, while visual features (colour histograms, optical flow, object detection, facial recognition) were derived using OpenCV. Six supervised classifiers—Naive Bayes, K-Nearest Neighbors, Logistic Regression, Decision Tree, Random Forest, and XGBoost—were evaluated, with hyperparameters optimized via grid search. Temporal coherence was enhanced using median filtering. Random Forest achieved the best performance, with 74% AUC on fused features and a 3% F1-score gain after post-processing. Spectral flux, grayscale histograms, and optical flow emerged as key discriminative features. The best model was deployed as a practical web service using TensorFlow and Flask, integrating informative segment detection with subtitle generation via beam search to ensure coherence and coverage. System-level evaluation demonstrated low latency and efficient resource utilization under load. Overall, the results confirm the strength of multimodal fusion and ensemble learning for video summarization and highlight their potential for real-world applications in surveillance, digital archiving, and online education.
- 10 September 2025
- Published by MDPI AG in Algorithms
- Vol. 18 (9) , 572
Open AccessMultimodal Video Summarization Using Machine Learning: A Comprehensive Benchmark of Feature Selection and Classifier Performance
The exponential growth of user-generated video content necessitates efficient summarization systems for improved accessibility, retrieval, and analysis. This study presents and benchmarks a multimodal video summarization framework that classifies segments as informative or non-informative using audio, visual, and fused features. Sixty hours of annotated video across ten diverse categories were analyzed. Audio features were extracted with pyAudioAnalysis, while visual features (colour histograms, optical flow, object detection, facial recognition) were derived using OpenCV. Six supervised classifiers—Naive Bayes, K-Nearest Neighbors, Logistic Regression, Decision Tree, Random Forest, and XGBoost—were evaluated, with hyperparameters optimized via grid search. Temporal coherence was enhanced using median filtering. Random Forest achieved the best performance, with 74% AUC on fused features and a 3% F1-score gain after post-processing. Spectral flux, grayscale histograms, and optical flow emerged as key discriminative features. The best model was deployed as a practical web service using TensorFlow and Flask, integrating informative segment detection with subtitle generation via beam search to ensure coherence and coverage. System-level evaluation demonstrated low latency and efficient resource utilization under load. Overall, the results confirm the strength of multimodal fusion and ensemble learning for video summarization and highlight their potential for real-world applications in surveillance, digital archiving, and online education.
- Open Access
- 7 January 2025
- Published by MDPI AG in Future Internet
- Vol. 17 (1) , 23
Framework Design for the Dynamic Reconfiguration of IoT-Enabled Embedded Systems and “On-the-Fly” Code Execution
Embedded systems, particularly when integrated into the Internet of Things (IoT) landscape, are critical for projects requiring robust, energy-efficient interfaces to collect real-time data from the environment. As these systems become complex, the need for dynamic reconfiguration, improved availability, and stability becomes increasingly important. This paper presents the design of a framework architecture that supports dynamic reconfiguration and “on-the-fly” code execution in IoT-enabled embedded systems, including a virtual machine capable of hot reloads, ensuring system availability even during configuration updates. A “hardware-in-the-loop” workflow manages communication between the embedded components, while low-level coding constraints are accessible through an additional abstraction layer, with examples such as MicroPython or Lua. The study results demonstrate the VM’s ability to handle serialization and deserialization with minimal impact on system performance, even under high workloads, with serialization having a median time of 160 microseconds and deserialization having a median of 964 microseconds. Both processes were fast and resource-efficient under normal conditions, supporting real-time updates with occasional outliers, suggesting room for optimization and also highlighting the advantages of VM-based firmware update methods, which outperform traditional approaches like Serial and OTA (Over-the-Air, the ability to update or configure firmware, software, or devices via wireless connection) updates by achieving lower latency and greater consistency. With these promising results, however, challenges like occasional deserialization time outliers and the need for optimization in memory management and network protocols remain for future work. This study also provides a comparative analysis of currently available commercial solutions, highlighting their strengths and weaknesses.
- 7 January 2025
- Published by MDPI AG in Future Internet
- Vol. 17 (1) , 23
Open AccessFramework Design for the Dynamic Reconfiguration of IoT-Enabled Embedded Systems and “On-the-Fly” Code Execution
Embedded systems, particularly when integrated into the Internet of Things (IoT) landscape, are critical for projects requiring robust, energy-efficient interfaces to collect real-time data from the environment. As these systems become complex, the need for dynamic reconfiguration, improved availability, and stability becomes increasingly important. This paper presents the design of a framework architecture that supports dynamic reconfiguration and “on-the-fly” code execution in IoT-enabled embedded systems, including a virtual machine capable of hot reloads, ensuring system availability even during configuration updates. A “hardware-in-the-loop” workflow manages communication between the embedded components, while low-level coding constraints are accessible through an additional abstraction layer, with examples such as MicroPython or Lua. The study results demonstrate the VM’s ability to handle serialization and deserialization with minimal impact on system performance, even under high workloads, with serialization having a median time of 160 microseconds and deserialization having a median of 964 microseconds. Both processes were fast and resource-efficient under normal conditions, supporting real-time updates with occasional outliers, suggesting room for optimization and also highlighting the advantages of VM-based firmware update methods, which outperform traditional approaches like Serial and OTA (Over-the-Air, the ability to update or configure firmware, software, or devices via wireless connection) updates by achieving lower latency and greater consistency. With these promising results, however, challenges like occasional deserialization time outliers and the need for optimization in memory management and network protocols remain for future work. This study also provides a comparative analysis of currently available commercial solutions, highlighting their strengths and weaknesses.
- 1 October 2024
- Published by Springer Nature
- p. 827-835
Optimizing Sudoku Solutions: Harnessing Parallelization for Lightning-Fast Results
- 1 October 2024
- Published by Springer Nature
- p. 827-835
Optimizing Sudoku Solutions: Harnessing Parallelization for Lightning-Fast Results
- 18 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 84-89
Application of Deep Learning and Neural Networks for Smart Grid Stability Predictions
Efficient and sustainable electrical grids are crucial for energy management in modern society and industry. Govern-ments recognize this and prioritize energy management in their plans, alongside significant progress made in theory and practice over the years. The complexity of power systems determines the unique nature of power communication networks, and most researches have been focusing on the dynamic nature of voltage stability, which led to the need for dynamic models of power systems. Control strategies based on stability assessments have become essential for managing grid stability, diverging from traditional methods and often leveraging advanced computational techniques based on deep learning algorithms and neural networks. This way, researchers can develop predictive models capable of forecasting voltage stability and detecting potential instability events in real-time, whereas neural networks can also optimize control strategies based on wide-area information and grid response, enabling more effective stability control measures, as well as detecting and classifying disturbances or faults in the grid. This paper explores the use of predictive models to assess smart grid stability, examining the benefits, risks, and comparing results to determine the most effective approach.
- 18 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 84-89
Application of Deep Learning and Neural Networks for Smart Grid Stability Predictions
Efficient and sustainable electrical grids are crucial for energy management in modern society and industry. Govern-ments recognize this and prioritize energy management in their plans, alongside significant progress made in theory and practice over the years. The complexity of power systems determines the unique nature of power communication networks, and most researches have been focusing on the dynamic nature of voltage stability, which led to the need for dynamic models of power systems. Control strategies based on stability assessments have become essential for managing grid stability, diverging from traditional methods and often leveraging advanced computational techniques based on deep learning algorithms and neural networks. This way, researchers can develop predictive models capable of forecasting voltage stability and detecting potential instability events in real-time, whereas neural networks can also optimize control strategies based on wide-area information and grid response, enabling more effective stability control measures, as well as detecting and classifying disturbances or faults in the grid. This paper explores the use of predictive models to assess smart grid stability, examining the benefits, risks, and comparing results to determine the most effective approach.