Directional antenna systems are gaining widespread adoption in wireless communication solutions, particularly using super-6 GHz bands in the electromagnetic spectrum. Hence, neighbor discovery and beam alignment in these directional wireless systems have attracted notable attention from researchers. However, fast neighbor discovery using directional wireless while maintaining covertness from eavesdroppers by minimizing the probability-of-intercept (POI) is an open research problem. We address this trade-off by proposing a sequential transceiver (or direction) selection protocol based on a tuning parameter (α) that guides the nodes to prioritize either goal by selecting the next operational transceiver subset for probing. We consider a 2-D multi-sector directional wireless system with electronic steering of transmission among sectors, assuming each sector is equipped with a transceiver and the sectors collectively cover the 2-D 360 • horizon around the node. We design a time-slotted neighbor discovery protocol that employs a probabilistic approach to select only one transceiver to use for the next time interval. By changing α, we study its capability to control the probing direction and impact on neighbor discovery speed and POI. Results show random selection offers fastest neighbor discovery but increases vulnerability to passive eavesdropping. Choosing from transceivers placed opposite to the active one accelerates discovery, while prioritizing the adjacent ones enhances covertness. During the search for the best α, we observe that exclusively prioritizing either rapid discovery or minimizing POI for a long stretch does not yield an optimal solution.

Monitoring the propagation of mechanical cardiac signals throughout the body is crucial for assessing cardiovascular health. A common drawback of current gold standard methods for vital sign monitoring is the necessity for continuous skin contact. Radar-based sensing offers a promising alternative by enabling contactless measurement of cardiac activity, including heart sound signals. As previous research has primarily focused on deriving signals from proximal body regions, insights into heart sound propagation to peripheral areas are lacking. To address this, we systematically investigated whether radar-based heart sound detection and propagation measurement is feasible across the whole body. Using a custom-built continuous-wave radar system and phonocardiogram as reference, we simultaneously recorded heart sounds in N=22 participants sequentially at eleven locations. The electrocardiogram served as reference for electrical heart activity. After synchronization and preprocessing, we manually segmented the heart sounds and extracted temporal characteristics from ensemble-averaged signals. Our findings show that heart sounds can be detected across the entire body with the radar-based as well as the gold standard system. Furthermore, the heart sounds' temporal characteristics vary between measurement locations. As the distance to the heart increases, we observed significantly increased propagation time intervals. This finding is consistent across both systems, exhibiting a high agreement with a Pearson Correlation Coefficient of .71 for the first and .68 for the second heart sound. In conclusion, our work is the first to demonstrate that radar-based systems are feasible for contactless detection of heart sounds and evaluating their propagation, offering new possibilities for research and health monitoring.

Ultrafast plane-wave ultrasound imaging is a versatile tool which has become increasingly relevant for blood flow imaging using speckle tracking, but suffers from a low signal-to-noise ratio (SNR). Cascaded dual-polarity waves (CDW) imaging can improve the SNR by transmitting pulse trains, which are subsequently decoded to recover the imaging resolution. However, the current decoding method (in the time domain) requires a set of two acquisitions, which introduces motion artifacts that result in incorrect speckle tracking at high flow velocities. Here, we propose a frequency-domain decoding approach that decodes acquisitions independently. Experiments using a disk phantom show that frequency-domain decoding of a four-pulse train achieves an SNR gain of up to 4.2 dB, versus 5.9 dB for conventional decoding. The benefit of frequency-domain decoding for flow quantification is assessed through experiments performed with a rotating disk phantom and a parabolic flow, and through matching linear simulations. Both CDW methods improve the tracking accuracy compared to single plane-wave imaging. Timedomain decoding outperforms frequency-domain decoding in low SNR conditions and low velocities (≤0.25 m/s), as a result of the higher SNR gain. In contrast, frequency-domain decoding outperforms time-domain decoding for high peak velocities in imaging of the rotating disk (1 m/s) and of the parabolic flow (2 m/s), when significant scatterer motion between acquisitions causes imperfect time-domain decoding. Its ability to decode individual acquisitions makes the proposed frequency-domain decoding of CDW a promising approach to improve the SNR and thereby the accuracy of flow quantification at high velocities.

This paper presents the tunable and switchable Perfect Absorbers (PAs), operating within the mid-infrared (mid-IR) spectrum, specifically targeting the 3 to 5 µm range with precise 0.25 µm intervals. This spectrum is particularly engineered for its minimal atmospheric absorption and unique atmospheric transmission characteristics. Our approach utilizes graphene-based nanophotonic aperiodic multilayer structures optimized through the synergy of micro-genetic optimization algorithm (GOA) within an inverse design framework. This strategic combination enables a predictive model-based strategy that broadens the design of space exploration, facilitating the discovery of PAs with highly accurate absorption control. Employing the Transfer-Matrix-Method (TMM) method for simulations, we manipulate the absorption characteristics, allowing for the precise tailoring of the desired spectral response while maintaining the multilayer structures' thickness under 2 µm. Our results demonstrate the tunability and switchability of PAs by adjusting graphene layers' chemical potentials, highlighting their dynamic optical behavior. For instance, a PA optimized for a 4 µm absorption peak can shift its absorption peak to 4.22 µm by merely changing the graphene layer's chemical potential from 0 eV to 1 eV, without compromising absorption efficiency. Additionally, our research uncovers the proposed absorbers' remarkable adaptability to various incident angles, maintaining 90% absorption up to 52 degrees. This adaptability demonstrates the versatility and robustness of our design across a broad spectrum of real-world applications, including thermal photovoltaics, sensors, and stealth technology, where angular independence significantly enhances device performance and efficiency. This research not only deepens the understanding of nanophotonic materials' capabilities but also paves the way for the design and development of highly efficient optical devices tailored for the mid-IR range.

Within Industry 4.0, efficient fault diagnosis plays a pivotal role in predictive maintenance of industrial machinery. However, the challenge lies in the significant domain shift between the source (training) and target (testing) domains, which hampers the application of machine learning in engineering practice. Several approaches based on transfer learning have been proposed to cope with the lack of training data in the target domain and the related domain adaptation challenges. Those approaches leverage the knowledge from similar source domains, including related real-world applications or lab machines. Unfortunately, access to sufficient faulty data from such source domains is often restricted due to insufficient history of faults in real machines, as well as difficulties to get labeled datasets from lab machines, which is time-consuming and sometimes unfeasible. To tackle those issues, this paper proposes a novel diagnostic framework integrating digital twins and transfer learning to mitigate the limitations posed by insufficient training datasets and domain discrepancies. By leveraging digital twins, training datasets are generated as the source domain, while introducing a model update strategy based on parameter sensitivity analysis to enhance adaptability. In addition, the partial transfer diagnostic model, incorporating a double-layer attention mechanism, enables to cope with data distribution discrepancies between digital twins and real machines, as well as inconsistencies in label spaces across domains. The diagnostic framework is validated on an industrial rotating machine case study, where faulty behaviors originated by defects on the inner race, outer race, and ball of the bearing are considered. Real data from two publicly available datasets are leveraged. The results of the experimental analysis have been compared with state-of-the-art methodologies: the proposed approach is able to improve the diagnostic accuracy by over 11% in the specific case study. Therefore, the approach can effectively increase equipment reliability, optimize maintenance, and enhance operational efficiency.

Intelligent job matching technology utilizes artificial intelligence to help people find suitable jobs more effectively. However, the development of this technology also brings a series of ethical and privacy issues, with data security and algorithm fairness being two main concerns. This paper synthesizes existing research, first introducing the background and current status of intelligent job matching technology, and then focusing on the challenges and solutions in data security and algorithm fairness, supported by relevant literature. Finally, through case analysis and future outlook, the development trends of ethical and privacy considerations in intelligent job matching technology are discussed.

Enabling wireless power transfer (WPT) over a large area has been proposed for charging moving devices. One of the main challenges in large-area WPT systems is the fluctuations of efficiency and power at certain receiver positions or orientations, which necessitates the implementation of complex receiver tracking and control algorithms. In this paper, we propose a planar three-coil receiver structure that inherently enables freepositioning movements in the transmitting area without any complex control or need for receiver position tracking. The results show that the proposed new receiver structure ensures almost constant DC-DC efficiency of approximately 85% throughout the full WPT area. The coil design guidelines are provided together with an optimization of 44% ferrite materials used in the WPT transmitter without affecting the performance. Finally, experimental results are presented to validate the proposed solution.

Exoskeleton robots hold immense promise in rehabilitation, serving as crucial aids for patient mobility and exercise. However, harnessing their capabilities requires overcoming significant control challenges arising from complex nonlinear dynamics and uncertainties in both models and actuators. This paper introduces a novel adaptive backstepping controller designed specifically for exoskeleton robots navigating uncertain dynamics and actuator parameters. Unlike conventional approaches that rely on basis functions, the proposed controller integrates a Modified Function Approximation Technique (MFAT) to approximate dynamic parameters without the need for such functions. The MFAT effectively manages mismatched perturbations, while the backstepping control compensates for uncertainties associated with state variables, enhancing resilience to disturbances, especially in scenarios where the exoskeleton's dynamics model is unknown. Lyapunov stability analysis ensures uniformly ultimately bounded signals within the closed-loop system. A comparative study conducted on the industrial robot IRB 120, validates the effectiveness of the proposed controller and highlights its superior performance. Real-time implementation on a 7 Degrees of Freedom (DOFs) wearable robot named ETS-MARSE confirms the efficiency of the control algorithm. The results from simulations and experiments underscore the efficacy of the proposed approach. The insights gained from this research pave the way to unlocking the full potential of exoskeleton robots in rehabilitation settings, promising improved patient outcomes and advancing human-technology interaction to new heights.

With the rapid development of artificial intelligence technology, the field of reinforcement learning has continuously achieved breakthroughs in both theory and practice. However, traditional reinforcement learning algorithms often entail high energy consumption during interactions with the environment. Spiking Neural Network (SNN), with their low energy consumption characteristics and performance comparable to deep neural networks, have garnered widespread attention. To reduce the energy consumption of practical applications of reinforcement learning, researchers have successively proposed the Pop-SAN and MDC-SAN algorithms. Nonetheless, these algorithms use rectangular functions to approximate the spike network during the training process, resulting in low sensitivity, thus indicating room for improvement in the training effectiveness of SNN. Based on this, we propose a trapezoidal approximation gradient method to replace the spike network, which not only preserves the original stable learning state but also enhances the model's adaptability and response sensitivity under various signal dynamics. Simulation results show that the improved algorithm, using the trapezoidal approximation gradient to replace the spike network, achieves better convergence speed and performance compared to the original algorithm and demonstrates good training stability.

Ransomware is one of the most significant cybersecurity threats to modern networks, leveraging advanced techniques to communicate with Command and Control (C&C) servers to coordinate attacks, exfiltrate data, and issue encryption commands. The novel approach presented in this study provides a comprehensive method for tracing ransomware network traffic to C&C servers, focusing on detecting the specific communication patterns, encryption protocols, and geolocation data that can enhance existing detection systems. The methodology integrates both signature-based and anomaly-based techniques, allowing for the identification of ransomware traffic even when advanced obfuscation methods, such as domain-flux and fastflux techniques, are employed. Through the use of controlled sandbox environments, custom-built C&C server emulators, and sophisticated network traffic analysis, the research outlines key indicators of ransomware communications and provides practical tools for improving network defenses. The results demonstrate how a hybrid approach to detection can significantly increase accuracy while maintaining operational efficiency, offering a tangible framework for mitigating ransomware threats through early identification of C&C traffic. The study also highlights the importance of understanding the geographical distribution of C&C infrastructure and its implications for network security, particularly in regions with lenient cybersecurity regulations. Overall, the findings contribute to a deeper understanding of the ransomware ecosystem and its critical dependence on covert C&C communications, providing a foundation for developing more effective cybersecurity defenses.

Efficient and informative representation of system topologies is critical in AI-driven power system analysis (PSA). Despite a major breakthrough, recent approaches employing Graph Neural Networks (GNNs) face significant challenges in large-scale PSA, including high computational demands for sufficient labeled data and poor generalization to unseen topologies. To tackle these issues, we propose a self-supervised strategy for pre-training GNNs that enhances their expressiveness at both the individual node feature level and the whole graph structure. Integrating physics-informed techniques, our strategy allows GNNs to internalize fundamental principles applicable to multiple downstream tasks. We demonstrate that our method enables efficient training of GNNs on extensive topology datasets without supervision, effectively addressing the noted challenges. By pre-training GNNs with 145 million parameters on 20 million unlabeled topologies and subsequently fine-tuning them, we observe a significant performance improvement, averaging over 13%, compared to existing state-of-the-art (SOTA) methods across four challenging tasks.

Current approaches to prosthetic control are limited by their reliance on traditional methods, which lack real-time adaptability and intuitive responsiveness. These limitations are particularly pronounced in assistive technologies designed for individuals with diverse cognitive states and motor intentions. In this paper, we introduce a framework that leverages Reinforcement Learning (RL) with Deep Q-Networks (DQN) for classification tasks. Additionally, we present a preprocessing technique using the Common Spatial Pattern (CSP) for multiclass motor imagery (MI) classification in a One-Versus-The-Rest (OVR) manner. The subsequent 'csp space' transformation retains the temporal dimension of EEG signals, crucial for extracting discriminative features. The integration of DQN with a 1D-CNN-LSTM architecture optimizes the decision-making process in real-time, thereby enhancing the system's adaptability to the user's evolving needs and intentions. We elaborate on the data processing methods for two EEG motor imagery datasets. Our innovative model, RLEEGNet, incorporates a 1D-CNN-LSTM architecture as the Online Q-Network within the DQN, facilitating continuous adaptation and optimization of control strategies through feedback. This mechanism allows the system to learn optimal actions through trial and error, progressively improving its performance. RLEEGNet demonstrates high accuracy in classifying MI-EEG signals, achieving as high as 100% accuracy in MI tasks across both the GigaScience (3class) and BCI-IV-2a (4-class) datasets. These results highlight the potential of combining DQN with a 1D-CNN-LSTM architecture to significantly enhance the adaptability and responsiveness of BCI systems.

This study presents a comprehensive analysis of a two-level battery charger for electric vehicles, focusing on modeling, simulation, and performance evaluation. The proposed charger topology employs two switches operating complementarily, along with essential components such as inductors, capacitors, and batteries. Detailed modeling in steadystate and dynamic regimes reveals the influence of nonlinearities, particularly switch and energy storage element characteristics, on charger efficiency and performance. Efficiency calculations highlight the significance of precise modeling and control strategies. Controller synthesis involves designing a robust proportional-integral compensator for effective battery charging current regulation. Analysis of non-idealities underscores the need for accurate component sizing and control strategies. The findings provide valuable insights for optimizing charger design and control, with implications for enhancing electric vehicle performance and reliability.

This paper deals with the minimum mean square error (MMSE) digital beamforming techniques at both transmitter (precoding) and receiver (combining) sides in multiuser multiple input multiple output (MU-MIMO) systems, under errors of channel state information (CSI). We consider very general system models in both downlink and uplink cases, where the number of transmit and receive antennas can take any values, and the CSI errors are random samples. Based on this model, we analyze the performance of the MMSE beamforming methods considering imperfect CSI in the construction of the beamforming matrices, reflecting possible errors of channel estimation. We theoretically show that the additional inter-user interference power due to the CSI errors is quadratically proportional to the square of a parameter α quantifying the CSI errors. This analytic result is supported by simulations, and we also show the effect of the CSI errors on the achievable bit error rate (BER), considering multipath and free space line-of-sight channel models. In both cases, simulations and theoretical results match, and highlight that the MMSE beamforming methods experience few dB loss in presence of potentially high CSI errors compared to perfect CSI.

Over the last decade, the cost of designing and training a deep neural network (DNN) model has escalated to the point that lures DNN model intellectual property (IP) theft and misappropriation. Recent advancements in model extraction and side-channel attacks have further reduced the cost of illegally reverse engineering a deployed DNN model. Existing countermeasures either require white-box accessibility to the suspect model for ownership verification or lack buyer traceability in the blackbox setting. This paper presents a novel black-box tripartite verifiable DNN IP protection scheme that not only allows the model owner (i.e., the first party) or any trusted third party (TTP) to verify the model ownership and trace the buyers of the distributed models, but also enables buyers (i.e., the second party) to verify the authenticity of a received model IP to prevent malicious model substitution during the delivery or prevent a fake model from exploiting good brand name. This is made possible by combining source-specific backdoor-based watermarking with adversarial fingerprinting. The proposed watermarking uses a dual-key trigger to bind an image source class with a composite backdoor feature. Such composite backdoor feature is created by applying a different data augmentation (DA) method on each color channel of an image. The large number of different accuracy-preserving model instances for buyer traceability can be met by embedding multiple backdoors with different combinations of dirty labels through fine-tuning a common clean host model. The trigger is embedded by content-adaptive interpolation to enhance the stealth and triggering efficiency. Adversarial examples generated from the host model are used for all its generated watermarked model instances as a common fingerprint for the proof of authenticity by transferability. Six series of 30 watermarked instances are evaluated. Each series is generated from either ResNet-18 or GoogleNet host model trained on CIFAR-10, GTSRB or ImageNet-10. They are evaluated on worstcase watermarked model accuracy, piracy detection accuracy, traitor traceability and robustness against nine types of attacks to demonstrate the superiority of the proposed scheme over stateof-the-art DNN IP protection schemes.

The widespread use of lithium-ion batteries and the demand for high-performance battery packs have made battery thermal modelling a crucial research area. This field helps to understand and manage the heat generated and the temperature distribution during battery operation. Effective thermal management is essential to ensure safety, extend battery lifespan and optimise efficiency. In this work, heat generation is identified as the primary driver of temperature change and distribution within the cell. Various battery models are reviewed and classified, driving the selection of the right model according to the application. Several thermal characterisation methods are described in detail, with a focus on experimental techniques that can be applied in-situ. Finally, the challenges posed by temperature distribution are discussed extensively, with a focus on how these challenges could shape the future research and design of the next generation of lithium-ion cells and battery packs.  

The aim of this research was to exactly quantify pure sodium metamizole from tablets , using a spectrophotometric analysis in Visible range. The method applied has been subjected to a validation protocal which consisted in analyzing the following parameters: linearity of the method, detection limit (LD) , quantitation limit (LQ). Following actual dosing, pure sodium metamizole amount in tablet of pharmaceutical was found to be 477.477 mg assigned to a percentage content of 95.495 %, very close to official declared amount (500 mg), with an maximum average percentage deviation of only 4.505 % from the official declared active substance content. This value was situated below the maximum admissible percentage deviation from stated active substance content (± 5%), established by Romanian Pharmacopoeia, X-th Edition rules.

This study examines the fundamental theory necessary for comprehending Variational Autoencoders (VAEs) and generative latent time-series models. We will also explain how these models extend the principles of VAEs to the domain of time-series data by incorporating temporal dependencies into the latent space. By leveraging the probabilistic nature of VAEs and the temporal dependencies captured by generative latent time-series models—researchers and practitioners can generate synthetic data for various applications, ranging from image generation to time-series forecasting. Through experiments and examples, we showcase the efficacy of these models in generating synthetic data that closely resembles the characteristics of the original dataset.