Dragan Samardzija
Research Subjects
Dragan Samardzija has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
Abdelrahman Ibrahim
Qualcomm Technologies, Inc., San Diego, CA, USA
1 shared publicationAleksa Arsic
RT-RK, Institute for Computer Based Systems, Novi Sad, Serbia
1 shared publicationAndreas Maeder
Nokia Standards, Munich, Germany
1 shared publicationBartek Kozicki
Nokia Bell Labs, Antwerp, Belgium
1 shared publication- Bhargav Gokalgandhi
Nokia Bell Labs, 600 Mountain Ave Bldg 5, New Providence, NJ 07974, USA
1 shared publication - Bogdan Pavković
Faculty of Technical Sciences, University of Novi Sad, 21000 Novi Sad, Serbia
1 shared publication
Latest Publications
- 18 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications
- Vol. 33 (3) , 126-133
Demonstrating Interoperable Channel State Feedback Compression With Machine Learning
Neural network-based compression and decompression of channel state feedback has been one of the most widely studied applications of machine learning (ML) in wireless networks. Various simulation-based studies have shown that ML-based feedback compression can result in reduced overhead and more accurate channel information. However, to the best of our knowledge, there are no real-life proofs of concepts demonstrating the benefits of ML-based channel feedback compression in a practical setting, where the user equipment (UE) and base station have no access to each others’ ML models. In this paper, we present a novel approach for training interoperable compression and decompression ML models in a confidential manner, and demonstrate the accuracy of the ensuing models using prototype UEs and base stations. The performance of the ML-based channel feedback is measured both in terms of the accuracy of the reconstructed channel information and achieved downlink throughput gains when using the channel information for beamforming. The reported measurement results demonstrate that it is possible to develop an accurate ML-based channel feedback link without having to share ML models between device and network vendors. These results pave the way for a practical implementation of ML-based channel feedback in commercial 6G networks.
- 18 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Wireless Communications
- Vol. 33 (3) , 126-133
Demonstrating Interoperable Channel State Feedback Compression With Machine Learning
Neural network-based compression and decompression of channel state feedback has been one of the most widely studied applications of machine learning (ML) in wireless networks. Various simulation-based studies have shown that ML-based feedback compression can result in reduced overhead and more accurate channel information. However, to the best of our knowledge, there are no real-life proofs of concepts demonstrating the benefits of ML-based channel feedback compression in a practical setting, where the user equipment (UE) and base station have no access to each others’ ML models. In this paper, we present a novel approach for training interoperable compression and decompression ML models in a confidential manner, and demonstrate the accuracy of the ensuing models using prototype UEs and base stations. The performance of the ML-based channel feedback is measured both in terms of the accuracy of the reconstructed channel information and achieved downlink throughput gains when using the channel information for beamforming. The reported measurement results demonstrate that it is possible to develop an accurate ML-based channel feedback link without having to share ML models between device and network vendors. These results pave the way for a practical implementation of ML-based channel feedback in commercial 6G networks.
- 23 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
System for detecting driver’s drowsiness, fatigue and inattention
High percentage of road accidents with even fatal outcomes are caused by the driver’s drowsiness and other factors that can be controlled by the driver himself. Thus, for modern Advanced Driving Assistance Systems (ADAS) it would be of great use to implement a reliable system for detecting driver’s drowsiness, fatigue, and inattention. One way this could be achieved is by using convolutional neural networks (CNN) and machine learning (ML) principles. In this paper, we present academic research on the topic which is based on three CNN’s used for monitoring the driver with a possibility to dispatch notification when concluded that his state of attention is not suitable for operating a motorized vehicle. Each of the three CNN’s processes different parts of the image from the inside of the vehicle and are connected in a series in a way that the outputs of the previous are used as the inputs for the next CNN model in the series.
- 23 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
System for detecting driver’s drowsiness, fatigue and inattention
High percentage of road accidents with even fatal outcomes are caused by the driver’s drowsiness and other factors that can be controlled by the driver himself. Thus, for modern Advanced Driving Assistance Systems (ADAS) it would be of great use to implement a reliable system for detecting driver’s drowsiness, fatigue, and inattention. One way this could be achieved is by using convolutional neural networks (CNN) and machine learning (ML) principles. In this paper, we present academic research on the topic which is based on three CNN’s used for monitoring the driver with a possibility to dispatch notification when concluded that his state of attention is not suitable for operating a motorized vehicle. Each of the three CNN’s processes different parts of the image from the inside of the vehicle and are connected in a series in a way that the outputs of the previous are used as the inputs for the next CNN model in the series.
- 23 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
Mobile Smartphone as a Security and Surveillance Device
A general problem that comes with setting up a security system is the price of the hardware that is used for surveillance, the ease of setting up the hardware required for the system that is in use and the accessibility of the application controlling the system from the user’s perspective. This paper presents two different methods of solving this problem. Each method was done as a separate app. The main idea is to find a device that can run these apps without costing too much and that doesn’t require much knowledge from the end user. When implementing our security system we also have to look at two different possibilities of conducting the system. The first possibility is to use an external server for analyzing the video stream that was captured from our device, while the second possibility is to implement the entire process on the same device as the surveillance system. There are also two different methods in creating our automated security system. The first method is to implement it using conventional computer vision techniques, while the second method relies on neural networks for detecting and tracking our intruder. The first app was created using the Java programming language with the help of the OpenCV library and the SMTP library. While the second app uses both Java and Python programming languages with the help of the YOLO library.
- 23 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
Mobile Smartphone as a Security and Surveillance Device
A general problem that comes with setting up a security system is the price of the hardware that is used for surveillance, the ease of setting up the hardware required for the system that is in use and the accessibility of the application controlling the system from the user’s perspective. This paper presents two different methods of solving this problem. Each method was done as a separate app. The main idea is to find a device that can run these apps without costing too much and that doesn’t require much knowledge from the end user. When implementing our security system we also have to look at two different possibilities of conducting the system. The first possibility is to use an external server for analyzing the video stream that was captured from our device, while the second possibility is to implement the entire process on the same device as the surveillance system. There are also two different methods in creating our automated security system. The first method is to implement it using conventional computer vision techniques, while the second method relies on neural networks for detecting and tracking our intruder. The first app was created using the Java programming language with the help of the OpenCV library and the SMTP library. While the second app uses both Java and Python programming languages with the help of the YOLO library.
- 15 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
Multi-camera object localization as a layer of a High-Definition map
People use conventional maps for orientation in various environments. With the increase of traffic participant number, restrictions of conventional maps result in traffic congestion and an increased number of accidents. We could overcome such restrictions of conventional maps with High Definition (HD) maps. On top of conventional maps, HD maps provide detailed real-time information about traffic participants in monitored areas. Consequently, HD maps enable faster deployment of autonomous driving systems and provide significant improvement to Smart Cities. We can provide such additional real-time information as a result of an object detection algorithm. We propose a novel solution in developing an HD map layer by combining multiple object detection results. Traffic is monitored from multiple cameras observing vehicles and pedestrians from different points of view. We tested our solution on an urban environment, upgrading the public Geolocation map to an HD map.
- 15 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
Multi-camera object localization as a layer of a High-Definition map
People use conventional maps for orientation in various environments. With the increase of traffic participant number, restrictions of conventional maps result in traffic congestion and an increased number of accidents. We could overcome such restrictions of conventional maps with High Definition (HD) maps. On top of conventional maps, HD maps provide detailed real-time information about traffic participants in monitored areas. Consequently, HD maps enable faster deployment of autonomous driving systems and provide significant improvement to Smart Cities. We can provide such additional real-time information as a result of an object detection algorithm. We propose a novel solution in developing an HD map layer by combining multiple object detection results. Traffic is monitored from multiple cameras observing vehicles and pedestrians from different points of view. We tested our solution on an urban environment, upgrading the public Geolocation map to an HD map.
- 19 August 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 9 (6) , 4533-4553
Reliable Low-Latency Wi-Fi Mesh Networks
We propose reliability and latency quantities as metrics to be used in the routing tree optimization procedure for Wi-Fi mesh networks. In contrast to state-of-the-art routing optimization methods, our proposal involves directly optimizing the data rates of individual mesh links according to underlying channel conditions such that reliability and latency requirements are satisfied for entire mesh paths. Moreover, to mitigate the channel contention problem that is common in Wi-Fi networks, we propose a multichannel (MC) assignment method. In this method, bandwidth is allocated to the individual mesh nodes based on the expected traffic load that they are expected to handle. Once the bandwidth for each node is determined, specific channels are assigned in a way to avoid co-channel interference. Furthermore, considerable efforts were spent for developing a system-level simulator that captures the features of the physical (PHY) layer and medium access layer defined in the IEEE 802.11 standard (Wi-Fi). Using this simulator, we were able to show that Wi-Fi mesh networks using the proposed routing metric based on reliability and latency quantities significantly outperform the state of the art. Finally, the mitigation of channel contention through the proposed MC assignment method results in further dramatic gains in performance.
- 19 August 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 9 (6) , 4533-4553
Reliable Low-Latency Wi-Fi Mesh Networks
We propose reliability and latency quantities as metrics to be used in the routing tree optimization procedure for Wi-Fi mesh networks. In contrast to state-of-the-art routing optimization methods, our proposal involves directly optimizing the data rates of individual mesh links according to underlying channel conditions such that reliability and latency requirements are satisfied for entire mesh paths. Moreover, to mitigate the channel contention problem that is common in Wi-Fi networks, we propose a multichannel (MC) assignment method. In this method, bandwidth is allocated to the individual mesh nodes based on the expected traffic load that they are expected to handle. Once the bandwidth for each node is determined, specific channels are assigned in a way to avoid co-channel interference. Furthermore, considerable efforts were spent for developing a system-level simulator that captures the features of the physical (PHY) layer and medium access layer defined in the IEEE 802.11 standard (Wi-Fi). Using this simulator, we were able to show that Wi-Fi mesh networks using the proposed routing metric based on reliability and latency quantities significantly outperform the state of the art. Finally, the mitigation of channel contention through the proposed MC assignment method results in further dramatic gains in performance.
- Open Access
- 1 January 2017
- Published by National Library of Serbia in Computer Science and Information Systems
- Vol. 14 (1) , 153-173
Fast edge-preserving gravity-like image interpolation
In this paper we propose a novel image interpolation algorithm which preserves edges and keeps a natural texture of interpolated images. The algorithm is based on an idea that only pixels that belong to the same side of an edge should be used in interpolation of pixels that belong to an edge. Beside similarity-based separation of known interpolation pixels a gravity-like interpolation coefficient set is also introduced in order to support different number of interpolation pixels and their location in two dimensional plane. The algorithm also applies arbitrary scaling factors, thus offering a broader scope of applications. Use of a local set of interpolating points makes the proposed algorithm suitable for applications on resource-limited platforms. The edge performance is demonstrated for structured geometric forms, while a general interpolated image quality is evaluated using objective measures and subjective comparisons. A comparison with some relevant interpolation algorithms shows the desirable tradeoff between image quality (sharpness and texture) and requested computing power (run-time).
- 1 January 2017
- Published by National Library of Serbia in Computer Science and Information Systems
- Vol. 14 (1) , 153-173
Open AccessFast edge-preserving gravity-like image interpolation
In this paper we propose a novel image interpolation algorithm which preserves edges and keeps a natural texture of interpolated images. The algorithm is based on an idea that only pixels that belong to the same side of an edge should be used in interpolation of pixels that belong to an edge. Beside similarity-based separation of known interpolation pixels a gravity-like interpolation coefficient set is also introduced in order to support different number of interpolation pixels and their location in two dimensional plane. The algorithm also applies arbitrary scaling factors, thus offering a broader scope of applications. Use of a local set of interpolating points makes the proposed algorithm suitable for applications on resource-limited platforms. The edge performance is demonstrated for structured geometric forms, while a general interpolated image quality is evaluated using objective measures and subjective comparisons. A comparison with some relevant interpolation algorithms shows the desirable tradeoff between image quality (sharpness and texture) and requested computing power (run-time).
- Open Access
- 1 January 2012
- Published by Walter de Gruyter GmbH in Journal of Electrical Engineering
- Vol. 63 (2) , 65-74
Road Nail: Experimental Solar Powered Intelligent Road Marking System
- 1 January 2012
- Published by Walter de Gruyter GmbH in Journal of Electrical Engineering
- Vol. 63 (2) , 65-74
Open AccessRoad Nail: Experimental Solar Powered Intelligent Road Marking System
- 1 January 2003
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 1 (10903038) , 399-403 Vol.1
Multiple antenna transmitter optimization schemes for multiuser systems
We present multiple antenna transmitter optimization (i.e., spatial prefiltering) schemes that are based on linear transformations and transmit power optimization (keeping average transmit power conserved). We consider the downlink of a wireless system with multiple transmit antennas at the base station and a number of mobile terminals (i.e., users) each with a single receive antenna. We consider the maximum achievable data rates in the case of the zero-forcing and triangularization spatial prefiltering coupled with a dirty paper coding transmission scheme. We also present the effects of channel mismatch.
- 1 January 2003
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 1 (10903038) , 399-403 Vol.1
Multiple antenna transmitter optimization schemes for multiuser systems
We present multiple antenna transmitter optimization (i.e., spatial prefiltering) schemes that are based on linear transformations and transmit power optimization (keeping average transmit power conserved). We consider the downlink of a wireless system with multiple transmit antennas at the base station and a number of mobile terminals (i.e., users) each with a single receive antenna. We consider the maximum achievable data rates in the case of the zero-forcing and triangularization spatial prefiltering coupled with a dirty paper coding transmission scheme. We also present the effects of channel mismatch.