Indoor positioning systems (IPS) are essential for modern smart home services, yet conventional Bluetooth Low Energy (BLE) techniques often incur high infrastructure costs and energy overhead due to the requirement for dedicated beacons. This paper proposes a opportunistic positioning framework that leverages existing operational traffic, specifically periodic heartbeat messages, within BLE Mesh automation networks. To address signal instability and the irregular traffic patterns inherent in mesh environments, we implement a data processing pipeline integrating Kalman filtering and temporal aggregation. These stabilized Received Signal Strength Indicator (RSSI) features are then utilized by a Deep Neural Network (DNN) to map sparse signal fingerprints to spatial coordinates. Experimental validation in a realistic office environment demonstrates that the proposed DNN model achieves high localization precision, with a Mean Absolute Error (MAE) of 37.5 cm and 46.7 cm for the X and Y coordinates, respectively. Our findings suggest that existing smart home infrastructures can be transformed into dual-purpose networks, enabling zero-overhead positioning while maintaining primary communication tasks.

Wearable IMU–based fall detection systems often exhibit significant subject-specific biases, which can compromise their reliability for new users. This paper proposed a deployment-oriented approach to enhance cross-subject generalization by implementing independent z-score normalization at the subject level. Utilizing the KFall dataset with 32 participants and a lightweight ConvLSTM model, we evaluate conventional global normalization against our proposed per-subject scheme using a LOSO protocol. The results indicate that per-subject normalization provides modest yet consistent improvements, increasing precision from 95.86% to 96.45%, and F1-score from 96.30% to 96.59%. More importantly, it substantially tightens the distribution of cross-subject results, reducing the standard deviation of accuracy and F1-score by 34.9\% and 33.6\%, respectively, thereby mitigating the impact of low-performing outlier users and leading to more stable behavior across unseen subjects. Building on this observation, we present an on-device implementation workflow that leverages short calibration sequences to estimate user-specific normalization parameters, which are then stored and reused during inference. This proposed approach is deterministic, label-free, and computationally efficient, making it well-suited for resource-constrained wearable devices, such as those based on microcontrollers.

The learning ability of lightweight CNN-based models is usually modest due to lack of spatial diversity in feature extraction as well as the imperfection of aggregated spatial information for identity mappings. To deal with these problems, we introduce an efficient lightweight model by addressing three novel concepts as follows. i) A novel perceptive block is proposed to extract discriminative moment-residual-coherent features (named MRCF) from depthwise-based tensors. ii) To adapt to the channel-elasticity moments of MRCF in a shallow backbone, two novel adaptive residual mechanisms are presented: an increase-moment residual is based on the expanding flexibility of a pointwise operator, while the decrease-moment one is on the aggregated spatial patterns of a fused tensor. To the best of our knowledge, it is the first time that an identity-mapping mechanism is structured for condensed-spatial information without increasing the model complexity. iii) A lightweight network is introduced by addressing three robust caret-shape segments of MRCFs. Experiments on various benchmark datasets have verified the efficacy of our proposals. All codes are available at https://github.com/nttbdrk25/CaretNet

Web crawling is a fundamental process in many applications such as search engines, data mining, and content integration. Traditional web parsers often collect directly to the websites or struggle with modern web content’s dynamic and heterogeneous nature, leading to inaccuracies and inefficiencies. This paper examines the implementation and evaluation of machine learning-based parsers aimed at enhancing the accuracy and adaptability of web crawling systems. By combining Google search and machine learning models, we aim to enhance the ability of parsers to understand and extract relevant information from diverse web pages. We integrate state-of-the-art natural language processing techniques and semantic analysis to develop parsers capable of handling complex and varied content structures. Our model demonstrates the superiority of machine learning-based parsers over conventional methods through extensive experiments and evaluations of real-world web data. The results show significant improvements in parsing accuracy and efficiency, improving the potential of machine learning to transform web crawling practices. In the applications, we perform the scrawling of trending majors that high school students are interested in joining at universities in Vietnam. The suggested websites are related to university training to make decisions, and students can understand and select appropriately to continue their higher education.

To achieve agricultural automation, deep learning applications for early and accurate disease detection in tomato plants have been extensively developed. However, there is a fundamental trade-off between computational efficiency and diagnostic accuracy in resource-constrained agricultural edge environments. This paper proposes an evaluation framework for seven architectures that represent standard, efficient, and hybrid CNN structures to assess their implementation potential. Through evaluations of explainability, computational efficiency, and diagnostic performance, seven lightweight architectures (ShuffleNetV2, MobileNetV3-Small, SqueezeNet, MobilePlantViT, DenseNet121, ResNet50, and VGG16) are thoroughly examined. Three significant findings are derived from experiments conducted on a subset of tomato diseases in the PlantVillage dataset. First, the MobilePlantViT architecture accurately strikes the ideal balance between efficiency and performance. Second, in order to quantitatively assess the explainability of XAI models (Grad-CAM, SHAP, and LIME) and identify the best option for edge devices, we propose the perturbation stability score (PSS) metric. Third, we test CPU inference measurements to better reflect the actual scenario and find that the hybrid design effectively leverages parallel computing. According to these findings, MobilePlantViT is the ideal architecture for applications that require operation on edge devices with limited resources and achieve high diagnosis accuracy (above 99.5%).

Quantum key distribution (QKD) provides a foundation for information-theoretic security based on quantum mechanics, yet its practical deployment is often constrained by intrinsically low secure key generation rates, particularly in high-bandwidth or low-latency settings. This work introduces a hybrid cryptographic technique that integrates conventional QKD with deterministic chaos, modeled using the Lorenz attractor, to provide a software-based enhancement of the effective key expansion rate. From a short 20-bit QKD seed, the system generates long bitstreams within milliseconds; although these streams exhibit high empirical randomness, their fundamental entropy remains bounded by the seed, consistent with standard cryptographic principles. The method employs the exponential divergence of chaotic trajectories, such that even minute uncertainties in an adversary’s estimate of the initial state lead to rapid desynchronization and, as established in Appendix A, an exponential decay of Eve’s mutual information with respect to the expanded key. Simulation results confirm this theoretical behavior and demonstrate an effective rate amplification exceeding two orders of magnitude over the baseline QKD seed rate. The proposed chaotic expansion operates entirely in software and requires no modifications to existing QKD hardware, offering a practical pathway to enhance throughput for applications ranging from secure video communication to low-latency IoT and edge-computing environments.

Cell-free massive multiple-input multiple-output (MIMO)-aided integrated sensing and communication (ISAC) systems are investigated where distributed access points jointly serve users and sensing targets. We demonstrate that only a subset of access points (APs) has to be activated for both tasks, while deactivating redundant APs is essential for power savings. This motivates joint active AP selection and power control for optimizing energy efficiency. The resultant problem is a mixed-integer nonlinear program (MINLP). To address this, we propose a model-based Branch-and-Bound approach as a strong baseline to guide a semi-supervised heterogeneous graph neural network (HetGNN) for selecting the best active APs and the power allocation. Comprehensive numerical results demonstrate that the proposed HetGNN reduces power consumption by 20−25% and runs nearly 10,000 times faster than model-based benchmarks.

The emergence of palm payment systems signals a transformative shift in digital transaction. While widely adopted in developed countries, their presence in developing economies remains limited and under-researched. This study examines factors influencing the adoption intention of palm payment through the UTAUT2 framework, integrating perceived trust as a key determinant. Data from 413 respondents were analyzed mix methods (PLS-SEM, ANN, and fsQCA). Firstly, this study employs the PLS-SEM approach due to its suitability for small sample sizes and its robustness against violations of multivariate normality assumptions. However, a major limitation in prior research using single-method approaches, particularly PLS-SEM, is the inability to fully address measurement errors. This study overcomes this issue by integrating ANN and fsQCA to validate and complement PLS-SEM results. ANN helps mitigate measurement error concerns by identifying patterns beyond linear relationships, reducing reliance on strict parametric assumptions. Meanwhile, fsQCA provides a configurational perspective, testing how different combinations of factors lead to adoption, thereby compensating for potential biases in individual variable estimates. The combined use of methods strengthens result reliability by cross-verifying findings through distinct analytical lenses, reducing the risk of over-or underestimating effects due to measurement inconsistencies. Findings reveal that hedonic motivation, price value, habit, perceived trust, social influence, facilitating conditions, performance expectancy, and effort expectancy significantly drive adoption intention. Theoretically, this study extends UTAUT2 by revealing the role of biometric trust in contactless payments. Methodologically, it confirms the value of a mixed-methods approach. Practically, it offers insights for enhancing biometric payment adoption in emerging markets.

Potassium–lithium aluminoborate glasses with the composition (30-x)K2O-xLi2O-5Al2O3-63B2O3-1Ce2O3-1Dy2O3 glasses (labeled as KLABx:Ce,Dy, where x = 5, 10, 15, 20, and 25) were successfully synthesized via the melt–quenching method. The glass–ceramic phase was obtained by annealing the precursor glass at the crystallization temperature. The phase structure of the materials was confirmed by X-ray diffraction. The mixedalkali effect was clearly observed through FTIR, absorption, and emission spectra, as well as the thermoluminescence process. The emission intensity and color coordinates of the KLABx:Ce,Dy glass could be tuned by varying the Li2O/K2O concentration ratio. The changes in the physical properties and network structure of the glasses, as a function of the proportion of each alkali metal, were also discussed in detail. Thermoluminescence (TL) properties and dosimetric properties such as sensitivity, dose response, reusability, and fading, etc., have also been studied towards the application of the material for the manufacture of TLDs. The ceramicization of glass by annealing at the crystallization temperature has significantly improved the dosimetric performance of the material. The research results show the prospect of applying this material to the manufacture of environmental radiation dosimeters and radiation field determination.

Overlapping illumination in multi-LED visible light communication (VLC) networks introduces cross-LED coupling that reshapes the received-signal composition and may trigger error propagation in successive interference cancellation (SIC) for layered ADO-OFDM NOMA. This work employs an overlap factor 𝛽 ∈[0,1] to quantify the severity of overlap-induced cross-LED coupling and develops a 𝛽-aware resource-allocation framework for a dual-LED indoor downlink. The proposed design integrates channel-aware MCGAD user grouping with three-level coefficient adaptation, including the inter-LED power split 𝜂, the inter-layer ACO/DCO split 𝜌, and the intra-layer two-user NOMA coefficients 𝛼. Monte Carlo evaluations over 𝛽 ∈{0,0.2,0.5} show that stronger coupling drives the system into an interference-limited regime with a pronounced high-SNR BER floor for strong users after SIC; the proposed 𝛽-aware design consistently reduces this floor relative to a 𝛽-blind fixed-coefficient baseline. Meanwhile, the spectral-efficiency curves remain close to the baseline, with only a minor gap at moderate-to-high SNR, and the Shannon-rate energy-efficiency trends remain comparable across coupling scenarios. The grouping-and-allocation procedure is dominated by sorting and deterministic pairing, exhibiting 𝒪⁡(𝑈⁢log⁡𝑈) complexity and avoiding the combinatorial growth of exhaustive grouping.

Green finance has become a key instrument for promoting sustainable development, particularly in emerging economies where environmental investment demands are rapidly increasing. This study examines the influence of financial capacity, market support, and sustainability culture on green bond issuance decisions (GBI) among listed construction and real estate firms in Vietnam. Using primary data collected through a structured survey and analyzed through multiple regression techniques in SPSS, the study evaluates the relationships between financial capacity (ACC), market support (SUP), sustainability culture (CUL), and green bond issuance decisions (GBI). Diagnostic tests confirm that the regression model satisfies classical assumptions, with variance inflationfactors (VIFs) below 3 and statistically significant ANOVA results. The model explains 29.4% of the variance in green bond issuance decisions. The findings indicate that financial capacity has a statistically significant positive effect on green bond issuance decisions, whereas market support and sustainability culture do not demonstrate statistically significant effects. These findings suggest that internal financial readiness remains the primary determinant of green financing adoption in Vietnam’s construction and real estate sector

This study explores the stabilization of a rotating parallel inverted pendulum, a nonlinear, underactuated system challenging conventional control methods. Unlike prior work focusing on linear techniques like Linear Quadratic Regulator, which lack robustness to uncertainties, this research introduces a Sliding Mode Control (SMC) strategy. Utilizing a linear sliding surface, the proposed SMC ensures finite-time convergence and robust performance despite model uncertainties and disturbances. Numerical simulations and experimental results validate the controller’s ability to stabilize the pendulum and rotating base, demonstrating its superiority over traditional methods and potential applicability to similar complex mechanical systems.

This study investigates the antecedents and outcomes of customer engagement in AI-driven digital platforms, focusing on the roles of Digital Trust, Content Engagement, Content Engagement AI, Emotion, and Platform Familiarity. Despite increasing activity on social media, previous findings indicate that user engagement remains largely superficial and unsustainable. Thus, this study aims to explore whether satisfaction derived from AI personalization and emotional connection can lead to long-term digital relationships. Using a purposive sampling method, data were collected from 430 respondents, predominantly young users aged 12–27 years in Jakarta and surrounding areas, during April 2025. Respondents were active users of social media platforms who engaged with influencer-generated content. Structural Equation Modeling using Partial Least Squares (SEM-PLS) was applied to analyze the relationships among variables. The results confirm that Digital Trust significantly drives Content Engagement, Emotion, and Customer Engagement, while Platform Familiarity enhances trust. Moreover, AI-driven personalization and emotional resonance play pivotal mediating roles in shaping deeper, sustainable engagement. This study contributes theoretically by elucidating the multidimensional process of engagement in AI-mediated contexts and practically by guiding marketers and policymakers to design more authentic and trustworthy digital ecosystems. Keywords: Customer Engagement, Digital Trust, Content Engagement AI, Thomas Tandewijaya Information Systems Department School of Information Systems Bina Nusantara University Jakarta 11480, Indonesia thomas.tandewijaya@binus.ac.id Nguyen Minh Tuan* Faculty of Information Technologyt, Posts and Telecommunications Institute of Technology Ho Chi Minh city, Viet Nam minhtuan@ptit.edu.vn consumers make to interactions, has become a central theme of marketing and communication research [1]. Unlike superficial measures such as clicks or views, engagement reflects a deeper psychological affinity that is reflected in loyalty and advocacy. In the past two years, artificial intelligence has enabled digital platforms to suggest highly personalized content, making it more engaging and resulting in more screen time [2]. The younger generations, particularly the digitally born, have been highly sensitive to such personalization. Such findings suggest that digital platforms are not merely spreading content but also shaping the way audiences connect with what they consume [3]. Emotion, Platform Familiarity

Searching for d0 magnetism in two-dimensional materials has attracted great attention because of its importance for next-generation spintronics. In this work, doping with pnictogen (X = P and As) atoms is proposed for engineering the magnetic properties of a SnI2 monolayer. The pristine monolayer is a nonmagnetic semiconductor with a band gap of 2.03 eV. Magnetic states are induced by doping with a single pnictogen atom, where an overall magnetic moment of 2.00 mB is produced primarily by the impurity. In addition, the emergence of magnetic semiconducting behavior and in-plane magnetic anisotropy (IMA) is also confirmed. The study of spin coupling in pnictogen-doped SnI2 systems demonstrates the effectiveness of X–X separation for controlling the electronic and magnetic properties. Specifically, an antiferromagnetic semiconducting nature is found to be stable with a small interatomic distance, while increasing the separation between the impurities induces a transition in the electronic behavior, resulting in ferromagnetic half-metallicity or ferromagnetic semiconducting character. In the latter case, large Curie temperatures of 282.20 and 462.66 K are obtained by P and As doping, respectively, indicating the robustness of the ferromagnetism. In addition, the X–X separation also significantly influences the magnetic anisotropy of the doped system. Specifically, perpendicular magnetic anisotropy (PMA) is obtained when two impurities are close to each other, whereas increasing the distance between them induces a PMA-to-IMA transition. Our findings provide insights into the electronic and magnetic properties of the SnI2 monolayer upon doping with pnictogen atoms, suggesting that these efficient doping approaches can induce d0 magnetism and enable promising applications in magnetic-field sensing and magnetoresistive random-access memory (MRAM) fabrication.

Existing smart contract vulnerability datasets exhibit over 34% train–test overlap due to repeated function-level code, causing models to favor structural memorization over semantic generalization. To mitigate this issue, we construct a benchmark dataset with zero function overlap between the training and test partitions. Furthermore, we introduce GraphFusionDetect (GFD), a novel approach that integrates fine-tuned CodeBERT embeddings with Graph Neural Networks (GNNs) to capture inter-function dependencies. GFD achieves F1-scores of 80% for detecting reentrancy vulnerabilities and 89% for timestamp dependency vulnerabilities, surpassing baseline methods and enabling more robust and generalizable vulnerability detection.

Emergency fire suppression activities subject rescue personnel to severe thermal conditions, hazardous fumes, and blast risks, creating extremely perilous environments for human operators. Rapid urban development has amplified fire emergency occurrences, necessitating the deployment of advanced autonomous firefighting platforms. This study presents an innovative tracked firefighting robot designed to navigate complex terrain and autonomously detect and approach fire sources. The system integrates a You Only Look Once version 8 (YOLOv8)-based deep learning model for real-time fire detection and employs depth imaging to calculate angular deviation and distance to the fire. These measurements are transmitted to a Programmable Logic Controller (PLC)-based control unit via a Modbus RS485 interface for responsive control. To enable autonomous navigation, the proposed robot combines an enhanced Bug-2 pathfinding algorithm with LiDAR-based environmental mapping and Hector Simultaneous Localization and Mapping (SLAM) for real-time localization and mapping. The core innovation lies in the integration of YOLOv8-based fire detection with deviation-angle-optimized Bug-2 navigation and a PLC-Robot Operating System (ROS) control architecture, enabling precise fire localization and obstacle avoidance in dynamic environments. Experimental validation confirms the effectiveness of the proposed firefighting robot in identifying fire sources and navigating around obstacles, demonstrating its potential as a reliable solution for autonomous firefighting in hazardous scenarios.

This paper is an attempt to evaluate ASEAN universities’ research impact and more specifically that of Vietnam by analyzing research-oriented key performance indicators that drive rankings. According to data, these universities have large gaps between the average reputations for academia and employers, faculty-student ratios, and research output. The experiment's three models named LSTM, GRU, and Hybrid predicted rankings on citation performance. The results revealed that the Hybrid model produces the highest accuracy, reiterating that the citation per paper variable tends to most strongly drive the positive correlation between citations per paper and university ranking. In addition, the study also indicated that Hue University cited 175% more research opened up to gain below a rank of 55. This point emphasized that research output is a major driver of rankings and institutional prestige. At the same time, the study offered several recommendations for universities, particularly those looking to increase their rankings, including focusing on high-impact publications, developing global research partnerships, and investing in faculty development.

This study aims to use the 2025 Times Higher Education (THE) Asia University Rankings as a frame of reference to consider what drives and dictates university rankings in the ASEAN region. Their analysis looks at six major indicators – Teaching, Research, Citations, Industry Income, International Outlook and Rank.. Comparative analysis confirmed that while Vietnam universities are purely citation-dependent in their quest to enhance higher ranked tables, some universities of our Southeast Asian neighbors at least start from a more balanced and diverse evaluation. The study deployed three deep learning models BiLSTM, BiGRU, and a Hybrid BiGRU-BiLSTM to predict university rankings. Among all the models, the Hybrid model performed the best with the lowest RMSE (3.72) and MAE (2.61), which indicated the highest predictive accuracy and stability. The analysis additionally showed the power of higher citation scores to boost rankings, with the greatest gains accruing to institutions with the lowest ranking. Feature importance analysis indicated that collaboration between industry and research is very important for ASEAN universities. Quality of teaching was still a main focus in Vietnam universities. The results highlight the role of research impact and international collaboration to go up in ranking. The analysis cautioned that universities and colleges looking to secure their position on the world stage needed to prioritize the volume of research produced, degree of international engagement, and depth of industry collaboration. The findings guaranteed the suitability of deep learning models as a robust methodological toolbox for ranking prediction and data-driven strategic improvement of university performance.

Accurate obstetric ultrasound segmentation is hampered by speckle noise and scarce annotations. We propose OASIS-Net, a dual-space adversarial semi-supervised framework that trains a single DeepLabV3$+$ backbone by minimizing one unified consistency loss. The loss couples input-space adversaries (iterative FGSM with $K=3$ steps, $\epsilon =4/255$) and weight-space gradient-aligned perturbations (DGAP, weight scale $=0.5$) whose influence grows with a sigmoid ramp ($T_{\text{ramp}}=20$, $\alpha _{\max }=1.0$). Pseudo-labels are accepted with a confidence threshold of 0.95 and the unlabeled loss weight is 1.0. We evaluate OASIS-Net on two public obstetric benchmarks: FUGC (50 labeled, 450 unlabeled) and PSFH (5,101 frames, 70% unlabeled). Using 20% of labels, the method attains Dice = 96.53% and HD$_{95}$ = 3.86 px on FUGC, and Dice = 97.16% and HD$_{95}$ = 2.34 px on PSFH. Ablation shows that removing either perturbation stream reduces Dice by up to 1.8 percentage points. The trained model runs at 18.96 frames s$^{-1}$ on a single RTX 4060 Ti and produces high-precision masks that enable automated cervical-length and angle-of-progression measurements for objective obstetric screening and intrapartum monitoring. These results demonstrate that jointly enforcing input- and parameter-space adversarial consistency yields a label-efficient, robust solution for obstetric ultrasound segmentation and supports real-time clinical use.

Reliable 3D object detection is critical for autonomous driving, yet LiDAR-only methods often fail under adverse weather, occlusion, or sensor degradation. We introduce TriFusion, a GNN-based multi-modal fusion framework that integrates LiDAR, camera, and radar for robust 3D detection. Our approach builds a heterogeneous graph with nodes representing modality-specific features and edges encoding spatial and cross-modal correspondences, enabling attention based message passing across sensors. Evaluated on the nuScenes benchmark against leading baselines (e.g., PointPainting, MVX-Net, BEVFusion), TriFusion achieves superior accuracy and robustness in challenging conditions while maintaining efficiency. These results underscore the promise of graph-based fusion for reliable perception in autonomous driving.

To enhance the effectiveness of vulnerability detection in software developed using C and C++ programming languages, our study introduces a novel correlation calculation method for analyzing and evaluating Code Property Graphs (CPG). The intelligent computation method proposed in this study comprises three key stages. In the first stage, we present a method for extracting features from the CPG source code. To accomplish this, we integrate three distinct data exploration methods: employing Graph Convolutional Neural (GCN) to extract node features from CPG, utilizing Convolutional Neural Network (CNN) to extract edge features from CPG, and finally employing the Doc2vec natural language processing algorithm to extract source code from CPG nodes. The second stage involves proposing a method for synthesizing CPG source code features. Building on the features acquired in the first stage, our paper introduces a synthesis and construction method to generate feature vectors for the source code. The final stage, stage three, executes the detection of source code vulnerabilities. The experimental results demonstrate that our proposed model in this study achieves higher efficiency compared to other studies, with an improvement ranging from 3% to 4%.

Facial Expression Recognition (FER) represents a crucial topic in computer vision and affective computing, focusing on automatically identifying human emotions through facial images. While recent developments in FER have been predominantly driven by deep learning architectures such as CNN-based and Transformer-based networks with promising results, these approaches primarily extract features directly from raw facial images. Our study reveals that incorporating facial landmark information in a meaningful way leads to improved performance. Specifically, using heatmaps generated from landmarks produces better results than using raw landmark coordinates. Secondly, the fusion approach significantly impacts performance, with early fusion yielding the best results. Finally, selective landmark points contribute more effectively to expression recognition than utilizing the complete set of facial landmarks. Through systematic experiments, we furthermore identify the optimal standard deviation value for Gaussian heatmap generation.

The application of machine learning in technical fields is developing in an increasingly large way. It offers significant benefits with notable results. In the field of diagnosis, machine learning is also widely used. Many studies have been proposed and successfully implemented. Recently, the iris has been used as a major instrument in personal identification. In the identification direction, numerous successful studies on the application of iris recognition have been commercialized for both civilian and military purposes. In addition, the iris also reflects the health status of the body, so many studies on the application of iris images in disease diagnosis have also been introduced. In this study, we examine recent results up to 2025 on the application of machine learning on iris images in disease diagnosis to provide a basic, multi-dimensional perspective. And we also discuss the challenges, opportunities, and potentials we are leveraging from this approach.

This paper investigates an in-band full-duplex (IBFD) unmanned aerial vehicle (UAV) communication system operating under realistic hardware impairments (HI). Although IBFD-UAV architectures offer strong potential for enhancing spectral efficiency and communication coverage, their practical deployment is fundamentally challenged by unavoidable hardware-induced distortions and residual self-interference (RSI). To counteract the negative effects of HI and RSI on system reliability, we derive exact and asymptotic expressions for the outage probability (OP) and throughput of the IBFD-UAV system (denoted as the IBFD-HI system). These expressions are general and can be adapted to ideal-hardware IBFD (IBFD-ID) and half-duplex (HD) systems with or without HI by appropriately modifying system parameters. To mitigate the detrimental effects of HI, RSI, and other performance-limiting factors, an optimal transmission power allocation scheme is recommended. Numerical results demonstrate that both HI and RSI dramatically degrade OP and throughput, while the proposed power optimization strategy effectively alleviates these impairments. Notably, an IBFD-HI system with optimal power allocation can even surpass an IBFD-ID system without power optimization, underscoring the practical importance of impairment-aware resource allocation. The analysis further examines the influence of key parameters, including HI level, RSI strength, data rate, UAV speed, operating frequency, and UAV positioning, on system behavior, from which several design insights are provided. All analytical findings are validated through Monte-Carlo simulations.