Ivan Pokrajac

Acoustic signal analysis serves as an essential approach for detecting drones operated via optical fibers, which evade conventional radio-frequency detection methods. Nonetheless, achieving reliable acoustic-based detection poses notable challenges due to the pronounced sensitivity of classification models to environmental noise and low signal-to-noise ratios. While deep transfer learning has emerged as a powerful tool for audio classification, the choice of optimal time-frequency representation remains a critical factor in model performance. This paper presents a comparative evaluation of spectrogramand bispectrum-based transfer learning for drone audio classification. Utilizing a comprehensive dataset of drone and non-drone acoustic signals, raw audio data is transformed into standard 2D spectrograms and higher-order bispectrum representations, which are then classified using three pre-trained models: ResNet-18, ResNet-50 and ViT Base 16. The results of this analysis suggest that higher-order spectral analysis provides a more robust and dependable feature representation in case of smaller model architectures, while larger networks struggle to generalize on the nonlinear features of the bispectrum and favor spectrogram representations.

Researchers are actively pursuing advancements in convolutional neural networks and their application in anti-drone systems for drone classification tasks. Our study investigates the hypothesis that the accuracy of drone classification in the radio frequency domain can be enhanced through a hybrid approach. Specifically, we aim to combine fuzzy logic for edge detection in images (the spectrograms of drone radio signals) with convolutional and convolutional recurrent neural networks for classification tasks. The proposed FLEDNet approach introduces a tailored engineering strategy designed to tackle classification challenges in the radio frequency domain, particularly concerning drone detection, the identification of drone types, and multiple drone detection, even within varying signal-to-noise ratios. The strength of this tailored approach lies in implementing a straightforward edge detection method based on fuzzy logic and simple convolutional and convolutional recurrent neural networks. The effectiveness of this approach is validated using the publicly available VTI_DroneSET dataset across two different frequency bands and confirmed through practical inference on the embedded computer NVIDIA Jetson Orin NX with radio frequency receiver USRP-2954. Compared to other approaches, FLEDNet demonstrated a 4.87% increase in accuracy for drone detection, a 13.41% enhancement in drone-type identification, and a 7.26% rise in detecting multiple drones. This enhancement was achieved by integrating straightforward fuzzy logic-based edge detection methods and neural networks, which led to improved accuracy and a reduction in false alarms of the proposed approach, with potential applications in real-world anti-drone systems. The FLEDNet approach contrasts with other research efforts that have employed more complex image processing methodologies alongside sophisticated classification models.

Signal detection theory, a fundamental concept in various scientific disciplines, involves mandatory measuring of the signal features. This theory finds applications in telecommunications, radar technology, medical devices, automation and process control, geophysical research, biometric systems, and security systems, emphasizing its broad significance. Likewise, drone detection in the radio-frequency domain is necessary for signal detection and ensures efficient and reliable communication, surveillance, and security. The recent conflict between Russia and Ukraine has underscored the crucial role of drones in modern warfare. Our research can improve the detection of any malicious drones that pose a threat, thereby underlining the significance of the proposed methodology in modern electronic warfare. This is a specialized approach to drone signal detection based on two-stage classification with two key components: a method based on spectrogram energy detection and deep learning classification. Energy detection on the spectrograms is particularly effective when the signal's energy characteristics differ significantly from the surrounding noise. The practical applicability of our proposed method was evaluated using the publicly available VTI_DroneSET dataset, which contains a diverse range of signals from three types of drones. Furthermore, we conducted tests with the VTI_DroneUSRP dataset and signals from the NI-USRP-2954 receiver, demonstrating the effectiveness and practicality of the proposed method in real situations. The successful detection and identification of Wi-Fi, Bluetooth, and drone signals in both ISM frequency bands were performed, proving the method's reliability.The proposed approach improved execution times and energy savings, indicating that applying the energy detector on the spectrograms in a two-stage classification significantly enhances the performance of ADRO applications for real-time drone detection. Furthermore, we conducted a comparative analysis of different deep learning algorithms at the outset of twostage classification, which is a potential basis for adopting this approach.

Signal detection theory, a fundamental concept in various scientific disciplines, involves mandatory measuring of the signal features. This theory finds applications in telecommunications, radar technology, medical devices, automation and process control, geophysical research, biometric systems, and security systems, emphasizing its broad significance. Likewise, drone detection in the radio-frequency domain is necessary for signal detection and ensures efficient and reliable communication, surveillance, and security. The recent conflict between Russia and Ukraine has underscored the crucial role of drones in modern warfare. Our research can improve the detection of any malicious drones that pose a threat, thereby underlining the significance of the proposed methodology in modern electronic warfare. This is a specialized approach to drone signal detection based on two-stage classification with two key components: a method based on spectrogram energy detection and deep learning classification. Energy detection on the spectrograms is particularly effective when the signal's energy characteristics differ significantly from the surrounding noise. The practical applicability of our proposed method was evaluated using the publicly available VTI_DroneSET dataset, which contains a diverse range of signals from three types of drones. Furthermore, we conducted tests with the VTI_DroneUSRP dataset and signals from the NI-USRP-2954 receiver, demonstrating the effectiveness and practicality of the proposed method in real situations. The successful detection and identification of Wi-Fi, Bluetooth, and drone signals in both ISM frequency bands were performed, proving the method's reliability. The proposed approach improved execution times and energy savings, indicating that applying the energy detector on the spectrograms in a two-stage classification significantly enhances the performance of ADRO applications for real-time drone detection. Furthermore, we conducted a comparative analysis of different deep learning algorithms at the outset of two-stage classification, which is a potential basis for adopting this approach.

Introduction/purpose: This article analyzes the publicly available literature on drone classification in the radio frequency domain, focusing on detection and identification. Drones are increasingly used for illegal purposes, making classification techniques crucial. This review paper covers passive radio frequency sensors, classification techniques, and datasets that highlight the challenges. Methods: Researchers are developing antidrone solutions because drones have become valuable tools and targets for illegal activities. Due to the scope of the subject matter, the review included only the classification of drones via passive radio frequency sensors with a description of the classification techniques (set of algorithms, methods, and procedures) and the datasets used for performance testing. This study introduces a new categorization and offers deeper insights into publicly available drone classification techniques. Results: Based on the results of this study, it is apparent that deep learning algorithms are presently the most effective approach to addressing the challenge of drone classification within the radio frequency domain. One of the primary obstacles is the absence of a comprehensive standard for classifying drones in the radio frequency domain, which should be based on end-user requirements. Additionally, the results of two ablative experiments highlight the preprocessing of raw I/Q radio signals as an essential step in drone classification. Conclusion: In summary, the proposed categorization provides a valuable tool for literature review. Deep learning is the most effective technique for drone classification, but publicly available datasets with drone radio signals are limited. The key strength of this study is that it represents the first review of publicly available datasets with drone radio signals.

We presented a method based on multigraphs to mathematically define a distribution function in time for the generation of data exchange in a special-purpose communication network. This is needed for the modeling and design of communication networks (CNs) consisting of integrated telecommunications and computer networks (ITCN). Simulation models require a precise definition of network traffic communication. An additional problem for describing the network traffic in simulation models is the mathematical model of data distribution, according to which the generation and exchange of certain types and quantities of data are realized. The application of multigraphs enabled the time and quantity of the data distribution to be displayed as operational procedures for a special-purpose communication unit. A multigraph was formed for each data-exchange time and allowed its associated adjacency matrix to be defined. Using the matrix estimation method allowed the mathematical definition of the distribution function values. The application of the described method for the use of multigraphs enabled a more accurate mathematical description of real traffic in communication networks.

Considering that networks based on New Radio (NR) technology are oriented to provide services of desired quality (QoS), it becomes questionable how to model and predict targeted QoS values, especially if the physical channel is dynamically changing. In order to overcome mobility issues, we aim to support the evaluation of second-order statistics of signal, namely level-crossing rate (LCR) and average fade duration (AFD) that is missing in general channel 5G models. Presenting results from our symbolic encapsulation point 5G (SEP5G) additional tool, we fill this gap and motivate further extensions on current general channel 5G. As a matter of contribution, we clearly propose: (i) anadditional tool for encapsulating different mobile 5G modeling approaches; (ii) extended, wideband, LCR, and AFD evaluation for optimal radio resource allocation modeling; and (iii) lower computational complexity and simulation time regarding analytical expression simulations in related scenario-specific 5G channel models. Using our deterministic channel model for selected scenarios and comparing it with stochastic models, we show steps towards higherlevel finite state Markov chain (FSMC) modeling, where mentioned QoS parameters become more feasible, placing symbolic encapsulation at the center of cross-layer design. Furthermore, we generate values within a specified 5G passband, indicating how it can be used for provisioningoptimal radio resource allocation.

The localization of various acoustic sources in a battlefield (such as weapon rounds, mortars, rockets, mines, improvised explosive devices, vehicle-borne improvised explosive devices and airborne vehicles) nowadays has a significant history. In acoustic source localization systems of multiple sensors, networked and placed at known positions are used to detect signals emitted from the source and perform localization of the source. Time of arrival estimation of the gun fire shock waves at large distances is significant when the purpose is the gun localization. At large distances shock wave signal is significantly deformed and therefore it is relatively difficult to accurately estimate the time of arrival. In this paper methods for TOA estimation are proposed, which are based on cumulants of the acoustic signal which originate from distant gunfire events. Localization of acoustic sources is performed in two steps by time difference estimation of acoustic signal arrival and using Discrete Probability Density method for positioning. Numerous field experiments have been conducted in order to verify performance of the proposed approach.

GNSS technology has been used for many applications. Beside military applications, GNSS technology is used for mini UAVs - drones. One of the possible approaches for achieving anti-drone capabilities is a jamming receiver of GNSS at drone. The GNNS jammers broadcast jamming signal in the frequency band used for satellite navigation in order to deny service of GNSS. In this paper we have considered the possibility to generate optimal jamming signal to deny service of all GNSS. Some of the results are shown in this paper.

Modern systems for electronic warfare, especially systems for electronic support based on fast scanning wideband direction finders, are usually designed to achieve high scanning speed in order to provide high probability of detection of signals of interest. To achieve high scanning speed two parameters are important: the first is instantaneous bandwidth of a receiver and the second is processing speed. By increasing instantaneous bandwidth of the receiver, it is possible to increase scanning speed. However, increasing of instantaneous bandwidth has a direct impact at amount of processing power to keep the same or to increase processing speed. In modern systems for electronic support, scanning speed depends on instantaneous bandwidth of the receiver, frequency resolution, implemented methods for estimation of direction of arrival and required preprocessing and processing power of hardware. The latest computer system architectures being developed by the computer industry offers more processing power in a smaller footprint. Parallel computing has been rejuvenated with the development of multicore technologies such as Multicore Processors, Multicore Digital Signal Processors and General Purpose Graphics Processing Units. In this paper an approach of using GPGPUs in the wideband fast scanning direction finder has been considered. Some of the obtained results are also presented.