Network

Vehicular Ad Hoc Networks (VANETs) face significant challenges in high-mobility environments, where dynamic channel conditions, particularly Doppler Shift (DS), degrade communication reliability and increase latency, thereby undermining safety-critical applications. To address these limitations, this paper proposes a neural network (NN)-based link adaptation strategy trained using the Scaled Conjugate Gradient (SCG) algorithm. SCG is selected as a second-order approximation optimizer that leverages curvature information to produce well-conditioned weight updates particularly suited to the small, physics-constrained training dataset. The SCG-optimized model dynamically adjusts transmission parameters to mitigate DS effects, improving real-time adaptability by explicitly incorporating Doppler Shift as a key input feature. Simulation results demonstrate that the proposed approach outperforms both the conventional Auto Rate Fallback (ARF) method and the SampleRate baseline. Specifically, the SCG-based strategy achieves an overall throughput improvement of +34.6% relative to ARF (1.77 Mbps vs. 1.32 Mbps) across all tested conditions, with condition-specific gains of +16.1% at 5 Hz Doppler (0.9 km/h), +21.7% at 750 Hz (137.3 km/h), and +35.2% at 1500 Hz (274.6 km/h), while consistently reducing transmission duration. A formal ablation study confirms that the Doppler Shift feature alone contributes +67% to +78% throughput gain at high mobility (DS > 900 Hz) compared to an SNR-only model. The main contributions of this work are threefold: (i) the explicit integration of Doppler Shift as a first-class input feature for link adaptation; (ii) the application of SCG optimization for fast, stable training of a lightweight feedforward neural network on a compact, physics-constrained dataset; and (iii) the formal ablation study that isolates and quantifies the Doppler feature’s contribution, establishing that the performance gain is attributable to feature engineering rather than the neural network architecture alone. This approach offers a scalable, real-time solution for Doppler-resilient VANET link adaptation. Full article

A coupled multi-stage learning framework is presented for joint bit error rate (BER) prediction and beam angle optimization in massive multiple-input multiple-output (MIMO) systems under a controlled simulation protocol. Unlike purely sequential benchmarking pipelines, the proposed method jointly coordinates BER prediction and beam-angle selection through a shared latent representation, an uncertainty-guided refinement mechanism, a cross-stage consistency loss and alternating optimization. Ten diverse approaches are systematically evaluated across two task-specific stages: Stage 1 examines six classical and adapted methods for BER prediction, including polynomial regression and deep unfolding networks; Stage 2 investigates four machine-learning and generative adversarial network (GAN)-based approaches for angle optimization, including conditional GANs and the proposed Direct-Angle neural network. Stage 3 couples the best-performing methods into a unified hybrid architecture through a shared encoder, explicit consistency regularization and alternating cross-stage updates, thereby producing an integrated beamforming decision strategy rather than an independent cascade. It is shown through the evaluation that the coupled hybrid framework achieves 96.0% overall angle-selection accuracy, a mean BER of $8.0\times {10}^{-5}$ and 100% BER tolerance compliance within $\pm 3$ dB. In this framework, a differentiable BER surrogate initialized from a second-degree polynomial-regression teacher is coupled with the proposed Direct-Angle-NN for angle optimization. Relative to the strongest reimplemented literature baseline under the same controlled simulation assumptions, a 33.3% reduction in mean BER is achieved. Ablation experiments show that the coupling mechanism provides a modest but consistent improvement over the decoupled sequential baseline, increasing angle-selection accuracy from 93.5% to 96.0% and reducing mean BER from $1.05\times {10}^{-4}$ to $8.0\times {10}^{-5}$; the shared encoder accounts for the largest part of this gain while the consistency loss adds 0.6 percentage points. These results indicate that the shared encoder, consistency regularization and uncertainty-guided refinement improve the final beamforming decision, although the gain should be interpreted as incremental rather than as a large architectural breakthrough. A spectral efficiency of 38.0 bps/Hz and an energy efficiency of 0.466 Gbps/W are achieved with a power consumption of only 32.6 W. The theoretical discussion is presented as an analytical characterization of BER sensitivity, complemented by a computational-complexity assessment and empirical convergence diagnostics for the alternating optimization, rather than as a formal optimality proof. The effectiveness of the framework across multiple performance metrics is supported by Monte Carlo simulations, while the limitations of the current setup, including perfect CSI, uncoded QPSK, ideal hardware assumptions and a fixed beam codebook, are explicitly discussed. The complete simulation framework, including code and trained models, can be made available by the corresponding author upon reasonable request to facilitate reproducible research in massive MIMO optimization. Full article

Weather forecasting, given the ever-increasing occurrence of climate change-induced events, has been widely introduced as a method to offer accurate and timely forecasts for proactive measures and risk mitigation. Artificial intelligence of things (AIoT) offers promising solutions for short-term weather forecasting, contributing to the advancement of sustainable and efficient weather monitoring technologies. This work presents everWeather_2.0, a significantly enhanced low-cost and self-powered AIoT-based weather forecasting station, which addresses key challenges in power consumption, user engagement and forecasting accuracy. The proposed end-to-end Cloud-Edge-IoT (CEI) proof-of-concept solution improves upon its predecessor by combining a more robust renewable energy subsystem for complete power autonomy with a series of lightweight, adaptive statistical models for on-device forecasting and an integrated display for on-site user engagement. Deployed in a real-world scenario, the station demonstrated seamless operation and high short-term forecasting accuracy for the thermodynamic variables during the pilot deployment period, with model errors observed as low as 2% for 30 min forecasts to 4.3% for 120 min intervals, validating its applicability in real-time and continuous physical weather monitoring. While wind speed and rainfall were monitored, they were excluded from the current accuracy metrics due to their high volatility and the insufficient number of events recorded during the pilot period to ensure reliable modeling. Full article

Over the last two decades, the automotive industry has spearheaded a shift toward data-centric manufacturing, where Real-Time Ethernet (RTE) networks defined in IEC61784-2 serve as critical components for ensuring deterministic communication at the Operation Technology level. Although AI-based systems offer significant potential for predictive maintenance and cybersecurity, their effectiveness is currently limited by a lack of structured datasets from real-world industrial environments. Most existing research relies on small-scale simulations or laboratory setups that fail to capture the scale and complexity of actual production. To address this gap, this paper introduces a novel methodology for repurposing network data collected throughout a plant’s lifecycle, specifically during the commissioning and validation phases of RTE networks according to IEC61918. An additional important contribution is the creation of the first multi-plant dataset for real RTE (PROFINET) traffic in the automotive sector, aggregating 300 GB of data from 54,000+ devices across nearly 700 production lines in 17 industrial sites. The work defines standardized methodologies and replicable processes for systematic data acquisition, validation, and labeling to ensure long-term usability for training AI models. Finally, four case studies (focused on performance, maintenance, security, and machine learning) show how this dataset can be used to enhance the reliability of modern smart manufacturing. Full article

The growing sophistication of cyber threats has reduced the effectiveness of traditional cybersecurity tools in protecting modern organisations and complex networks. This challenge requires advanced solutions capable of real-time detection, rapid response, and efficient threat mitigation. In this context, AI-based approaches have emerged as a powerful enabler of intelligent, adaptive, and data-driven security operations. This study presents a comprehensive analysis of AI-driven threat detection combined with automated incident response mechanisms in modern cybersecurity architectures. The novelty of this work lies in the integration of advanced machine learning-based detection with real-time, automated response capabilities to address zero-day and previously unknown threats in heterogeneous digital environments. The paper examines system architecture design, implementation strategies, and performance evaluation across diverse deployment scenarios. Experimental results demonstrate that AI-driven detection with automated response significantly enhances cybersecurity effectiveness, achieving accuracies between 96% and 97%, dramatically reducing the mean response time from 45 min to less than 30 s, and substantially improving zero-day threat detection and containment success rates. Overall, the proposed approach achieves up to a 98.9% improvement in incident containment efficiency, highlighting the operational and defensive advantages of intelligent automation. Full article

Continuous and objective monitoring of livestock behavior plays a key role in precision farming, animal welfare assessment, and reproductive management. This study proposes a non-invasive framework for sheep behavior and reproductive activity monitoring that integrates wearable actigraphy, machine learning, and a cloud-based data processing architecture. Tri-axial accelerometer data were collected at 30 Hz using collar-mounted ActiGraph sensors under real farming conditions. Raw acceleration signals were processed without temporal aggregation, preserving full temporal resolution that includes axis-specific acceleration, vector magnitude, and delta magnitude features. Several supervised learning models were evaluated for behavior classification, including BLSTM, LSTM, CNN–BLSTM, Random Forest, and Support Vector Machine, targeting behaviors such as standing, walking, grazing, lying, flehmen, and mating. The results indicate that both deep learning and classical machine learning approaches achieve high classification performance, with Random Forest obtaining an overall accuracy of 0.82, while deep sequential models effectively capture temporal patterns and behavioral transitions. Furthermore, a scalable cloud architecture is introduced to automate data ingestion, preprocessing, inference, storage in InfluxDB, and visualization through an interactive web application. The proposed framework supports continuous monitoring and offers practical tools for precision livestock management. Full article

Underwater Wireless Sensor Networks (UWSNs) play a vital role in ocean monitoring and exploration. However, harsh underwater conditions and frequent topology changes caused by node and sink mobility pose significant challenges for reliable routing. Conventional routing protocols that depend on global route reconstruction and static paths generate excessive control overhead and degrade performance in large-scale underwater environments. In this paper, we propose an energy-efficient virtual cell-based mobile-sink adaptive routing (VC-MAR) protocol for UWSNs. The sensing field is logically partitioned into a three-dimensional grid of virtual cells, where a cell-gateway is elected in each cell to construct a low-overhead routing backbone. To support sink mobility, VC-MAR introduces a localized route-adjustment mechanism that updates only the affected backbone segments rather than reconstructing the entire routing structure. By confining routing updates to neighboring cells influenced by sink movement, the proposed protocol significantly reduces control packet exchanges while ensuring stable and reliable data delivery. Simulation results show that the proposed VC-MAR improves the packet delivery ratio by up to 20% and reduces routing control overhead by about 34% compared with traditional grid-based routing methods. These results confirm the suitability of VC-MAR for dynamic and realistic underwater sensing scenarios. Full article

The dynamic nature of Vehicular Ad-hoc Networks (VANETs) necessitates robust authentication mechanisms to prevent adversaries from compromising vehicle privacy. To address privacy concerns, many existing approaches employ pseudonyms in place of real vehicle identities. However, the use of a single pseudonym is insufficient, as vehicle trajectories can still enable tracking. Consequently, vehicles must frequently change pseudonyms, typically selecting them from a pre-assigned pool, to ensure unlinkability and preserve privacy. In most existing schemes, a central authority issues certificates corresponding to each pseudonym, which vehicles present for authentication. While effective, this approach incurs significant computation, storage, and communication overhead, particularly in managing certificate revocation lists (CRLs), since each vehicle may possess a large number of pseudonyms. To address these challenges, we propose a Low-overhead and Efficient Privacy-preserving Authentication (LEPA) scheme for VANETs, leveraging Merkle Hash Trees (MHTs) and Cuckoo Filters (CFs) to efficiently manage pseudonym sets and revocation. We analyze the security of the proposed scheme against various attacks and demonstrate, through performance evaluation, that LEPA significantly reduces authentication and revocation overhead while maintaining strong privacy and security guarantees. Full article

This paper provides a comprehensive overview of the progress in fiber-optic gyroscope technology, covering 260 key studies of the last ten years. A critical comparative analysis of fiber-optic gyroscope with alternative inertial sensors (Micro-Electro-Mechanical Systems, Hemispherical Resonator Gyroscope, Ring Laser Gyroscope) has been carried out. Confirming the unique advantages of fiber-optic gyroscope for autonomous navigation. Fundamental limitations of accuracy are considered in detail: temperature drifts, polarization noise, and Rayleigh backscattering. Modern hardware methods for suppressing these errors, including the use of photonic crystal and hollow fibers (Air-Core/Hollow-Core), are also considered in this work. The central place in the review is occupied by the analysis of the technological paradigm shift from bulky discrete circuits to hybrid integrated photonics (Indium Phosphide, Silicon Nitride, Lithium Niobate) and hybrid architectures to reduce weight and size characteristics. The role of artificial intelligence (Deep Learning, Long Short-Term Memory) methods in nonlinear drift compensation and calibration is discussed. The usage of the Brillouin effect and optomechanics promising areas are outlined, necessary to create a new generation of navigation systems operating in the absence of Global Navigation Satellite Systems signals. Full article

A discrete Hartley transform (DHT)-based orthogonal frequency division multiplexing (OFDM) scheme is investigated for intensity modulation/direct detection (IM/DD) visible light communication (VLC) systems, where transmitted signals are required to be real-valued and non-negative. To address this constraint, a practical unipolar transmission framework with corresponding bipolar reconstruction is developed. By exploiting the real-valued and self-inverse properties of the DHT, the proposed scheme removes the need for Hermitian symmetry and enables full utilization of available subcarriers. Under equal-bandwidth conditions, this results in an approximately 50% reduction in computational complexity compared with conventional DCO-OFDM and ACO-OFDM schemes. Theoretical analysis and numerical results further show that the proposed approach achieves comparable bit error rate (BER) performance while exhibiting improved spectral confinement, as reflected by reduced out-of-band sidelobes under identical filtering conditions. In addition, it maintains spectral efficiency equivalent to DCO-OFDM under the same bandwidth constraint. These advantages are achieved at the cost of restricting subcarrier modulation to real-valued constellations, which may reduce flexibility in frequency-selective channels. Overall, these findings support DHT-OFDM as a low-complexity, spectrally confined multicarrier waveform for IM/DD VLC systems, particularly in scenarios where efficient spectrum utilization and reduced adjacent-channel interference are required. Full article

The use of millimeter-wave spectrum in fifth-generation (5G) systems is increasing the need for accurate prediction of received power and coverage in real deployment scenarios. In this context, ray tracing (RT) is a promising approach for site-specific analysis, although its reliability depends on how accurately different tools reproduce measurements in complex urban environments. This work presents a comparative assessment at 27 GHz of three RT tools: in-house Exact tool based on Vertical Plane Launching (VPL), Matlab 5G and open-source Sionna RT based on Shooting and Bouncing Rays (SBR). The comparison relies on a large outdoor walk-test campaign, including about 14,725 measurement points collected in a real urban area around a 27 GHz mMIMO base station, using real operator-provided antenna radiation patterns. Measured and simulated power levels are compared using statistical metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and a planning-oriented coverage-rate metric. The results show a reasonable agreement between simulations and measurements, with RMSE and MAE values around 10–12 dB, highlighting tool-specific behaviors related to boundary effects, interaction modeling, and high-power overestimation. This work confirms that RT is a flexible support for 5G preliminary network design, reducing the need for extensive drive tests. Full article

Digital substations (DSs) play a critical role in modern Energy and Power Electrical Systems (EPESs), enabling intelligent control, monitoring, and automation. With increased reliance on communication and sensing technologies, DSs are vulnerable to cyberattacks such as False Data Injection (FDI). An adversary may falsify transformer temperature readings, misleading protection mechanisms and resulting in incorrect disconnection actions. These false disconnections may disrupt power delivery, cause economic losses, and reduce equipment lifespan. To address these challenges, we propose a reinforcement learning-based approach for cyber protection of smart grids against false temperature data injection attacks. Specifically, this work designs a Long Short-Term Memory Deep Deterministic Policy Gradient (LSTM-DDPG) deep reinforcement learning algorithm that learns to detect normal patterns and responds to suspicious thermal patterns by dynamically adjusting disconnection decisions. The agents process sequential state features to differentiate between legitimate overload conditions and sudden anomalies caused by FDI attacks. We implement the proposed approach on the IEEE 30-bus distribution network using the Pandapower simulator. The experimental results indicate that the LSTM-DDPG controller outperforms conventional DDPG and DQN baselines, achieving a recall of 0.897, F1 of 0.945, precision of 1.00 and accuracy of 0.981 with a confidence interval of 95%. In addition, grid stability reaches up to 0.9815, 1.0, 1.0, 0.9926 with respect to the voltage stability score, transformer stability value, disconnection stability, and stability index, respectively. The proposed method led to fewer false disconnections, providing improved robustness against sensor manipulations. Full article

The integration of service function chains (SFCs) and unmanned aerial vehicles (UAVs) lays a crucial technological foundation for achieving efficient, reliable, and adaptive future airborne service networks. Service functions (SFs) in the SFC will be deployed on UAVs; this type of SFC is called unmanned aerial vehicle-based service function chains (USFCs). However, due to the combined effects of open hardware and software architectures, exposed communication links, and complex mission environments, UAVs have become ideal targets for attackers. Once a vulnerability is successfully injected into a UAV, data from the SFs running on it will be stolen, seriously threatening the dependability of the USFC. Therefore, it is necessary to conduct a quantitative evaluation of the USFC dependability to provide insights for further improving its dependability. This paper develops a USFC dependability evaluation model based on a semi-Markov process (SMP) to capture the dynamic interaction between attacker behavior and USFC system recovery behavior. The dependability of the USFC is comprehensively evaluated from two perspectives: availability and security. Extensive numerical analysis experiments are conducted, and the results not only demonstrate the changing trends of various dependability metrics under different parameters but also show parameter combinations for synergistic optimization among metrics. Full article

Network Intrusion Detection Systems (NIDS) face increasing challenges from sophisticated cyber threats, particularly zero-day attacks that evade signature-based methods. While supervised learning is effective for known attack classification, it struggles with novel threats, whereas anomaly-based approaches suffer from high false positive rates and unstable thresholds. To address these limitations, this paper proposes a decision-level adaptive intrusion-detection framework combining hierarchical CNN-based closed-set classification with autoencoder-based zero-day detection in a cascade architecture. The framework enables deployment-time adaptation by dynamically adjusting class-specific confidence thresholds and fusion parameters without model retraining. Experiments on the CSE-CIC-IDS2018 dataset demonstrate strong closed-set performance, achieving 98.98% accuracy and a macro-F1-score of 0.9342, with improved recall for minority attack classes under adaptive thresholding. Under a zero-day evaluation protocol in which Web_Attacks and Infiltration are excluded from training and validation, the proposed approach achieves an F1-score of 0.9319 while maintaining a low false positive rate of 0.0019. The framework is further evaluated on the Simulated University Network Environment (SUNE) dataset representing campus network traffic, achieving 96.18% closed-set accuracy and 97.54% accuracy in the integrated cascade setting. These results demonstrate that the proposed framework effectively balances minority attack detection, zero-day identification, and false-alarm control in dynamic and resource-constrained network environments. Full article

Antivirus (AV) solutions remain a core defence mechanism against malicious software. However, many of these engines struggle to detect metamorphic malware, which continually alters its internal form in unpredictable ways. To address this limitation, we present an adversarially oriented approach that automatically generates novel malicious variants of existing malware that evade detection by a substantial proportion of AV systems, thereby providing material for strengthening defensive techniques. In this work, an Evolutionary Algorithm (EA) is used to evolve undetectable variants, guided by three fitness criteria: the evasiveness of the produced samples, and their behavioural and structural similarity to the original malware. The proposed method is assessed across three malware families to evaluate the effectiveness of the EA-generated variants. Results indicate that the EA produces diverse mutant variants capable of evading up to 94% of AV detectors for a given malware family, significantly surpassing the evasion rate of the original malware. Furthermore, we evaluated whether the mutants produced by the EA could enhance the training of machine learning models. In this context, a pretrained Natural Language Processing (NLP) transformer was employed within a transfer learning framework to improve the classification of metamorphic malware. When the evolved variants were incorporated into the training data, the approach achieved classification accuracies of up to 93%. These results highlight the value of using diverse EA-generated samples to strengthen malware classifiers, thereby improving the robustness of security systems against evolving threats. Full article

The rapid development of the Internet of Things (IoT) is transforming various economic and industrial sectors by embedding interconnected devices within their operational processes. However, security and privacy risks associated with these interconnected devices pose significant barriers to widespread adoption, particularly in light of potential quantum threats. To mitigate these challenges, it is imperative to employ post-quantum cryptographic schemes. However, essential constraints on IoT edge nodes complicate the effective implementation of such schemes. Among the most promising approaches in post-quantum cryptography are lattice-based schemes, which rely heavily on polynomial multiplication operations at their core. Improving the implementation of polynomial multiplication will significantly enhance the performance of these schemes. Therefore, this paper proposes an efficent low-complexity serial systolic array optimized for polynomial multiplication, particularly tailored for the Binary Ring Learning With Errors (BRLWE) scheme. Designed for cryptographic processors targeting capable IoT edge nodes, the proposed architecture demonstrates remarkable performance improvements, achieving a maximum operating frequency of 280 MHz for a field size of 256, while requiring only 8232 lookup tables (LUTs) and 2616 flip-flops (FFs). These results reflect a 16.8% reduction in LUT usage and a 19% reduction in FFs compared to the nearest competing designs, all while maintaining high throughput and low area utilization. This work significantly advances the establishment of secure and efficient infrastructure for IoT systems, bolstering their resilience against post-quantum attacks and supporting the growth of a robust digital economy. Furthermore, it aligns with sustainable development goals 8 and 9 by fostering trust and facilitating the adoption of cutting-edge IoT technologies, ultimately promoting more resilient and innovative economic activities. Full article

Fifth-generation (5G) mobile networks must simultaneously satisfy stringent latency targets, high user density, and energy-aware operation across heterogeneous services. Cloud Radio Access Networks (C-RAN) provide architectural flexibility through centralized baseband processing, but they also introduce new control challenges related to fronthaul constraints, dynamic traffic variations, and joint radio–compute coordination with Mobile Edge Computing (MEC). This paper proposes a unified AI-driven optimization framework for adaptive 5G C-RAN management, where the controller dynamically tunes key system decisions—including functional split selection, TDD downlink ratio, user–RU association, fronthaul load management, and MEC offloading proportion. To enable fair benchmarking under identical simulation settings, a static baseline policy is compared against five adaptive control strategies: Deep Q-Network (DQN), Proximal Policy Optimization (PPO), Deep Deterministic Policy Gradient (DDPG), Multi-Objective Reinforcement Learning (MORL), and a Deterministic Service-Level Agreement (SLA)-aware controller Penalty-Constrained Hierarchical Action Controller (PCHAC). Performance evaluation across techno-economic and service KPIs shows that intelligent control significantly improves operational profit, tail-latency behavior, and energy efficiency while enhancing SLA compliance compared with non-adaptive operation. The results highlight the practicality of multi-objective and constraint-aware learning for next-generation C-RAN orchestration under scaling traffic demand. Full article

Dynamic Spectrum Access (DSA) has emerged as a viable solution to address spectrum scarcity in shared-spectrum networks. In response, the FCC established the Citizens Broadband Radio Service (CBRS) to manage and facilitate shared use of the federal and non-federal spectrum in a three-tiered access and authorization framework. However, due to the open nature of spectrum access and the usually limited coverage of the monitoring infrastructure, enforcing access rights in a shared-spectrum network becomes a daunting challenge. In this paper, we stipulate the use of crowdsourcing as a viable approach to engaging volunteers in spectrum monitoring in order to enforce spectrum access rights robustly and reliably. The success of this approach, however, hinges strongly on ensuring that spectrum access enforcement is carried out by reliable and trustworthy volunteers within the monitored area. To this end, a hybrid spectrum monitoring framework is proposed, which relies on opportunistically recruiting volunteers to augment the otherwise limited infrastructure of trusted devices. Although a volunteer’s participation has the potential to enhance monitoring significantly, their mobility may become problematic in ensuring reliable coverage of the monitored spectrum area. To ensure continued monitoring, inspite of volunteer mobility, deep learning-based models are used to predict the likelihood that a volunteer will be available within the monitoring area. Three models, namely LSTM, GRU, and Transformer, are explored to assess their feasibility and viability to predict a volunteer’s availability likelihood over an extended time interval, in a given spectrum monitoring area. Recurrent Neural Networks (RNNs) such as GRU and LSTM are effective for tasks involving sequential data, where both spatial and temporal patterns matter, which is the focus of volunteer availability prediction in spectrum monitoring. Transformers, on the other hand, excel at handling long range dependencies and contextual understanding. Furthermore, their parallel processing capabilities allows faster training and inference compared to RNN-based models like GRU and LSTM. A simulation-based study is developed to assess the performance of these models, and carry out a comparative analysis of their ability to predict volunteers’ availability to monitor the spectrum reliably. To this end, a real-world trace dataset of volunteers’ location, collected over five years, is used. The simulation results show that the three models achieve high prediction accuracy of volunteers’ availability, ranging from 0.82 to 0.92. The results also show that a GRU-based model outperforms LSTM and Transformer-based models, in terms of accuracy, Root Mean Square Error (RMSE), geodesic distance, and execution time. Full article

Underwater Acoustic Sensor Networks (UASNs) are pivotal for environmental monitoring, surveillance, and marine data collection. However, their open and largely unattended operational settings, constrained communication capabilities, limited energy resources, and susceptibility to insider attacks make it difficult to achieve safe, secure, and efficient collaborative learning. Federated learning (FL) offers a privacy-preserving method for decentralized model training but is inherently vulnerable to Byzantine threats and malicious participants. This paper proposes trust-aware FL for underwater sensor networks (TAFL-UWSN), a trust-aware FL framework designed to improve security, reliability, and energy efficiency in UASNs by incorporating trust evaluation directly into the FL process. The goal is to mitigate the impact of adversarial nodes while maintaining model performance in low-resource underwater environments. TAFL-UWSN integrates continuous trust scoring based on packet forwarding reliability, sensing consistency, and model deviation. Trust scores are used to weight or filter model updates both at the node level and the edge layer, where Autonomous Underwater Vehicles (AUVs) act as mobile aggregators. A trust-aware federated averaging algorithm is implemented, and extensive simulations are conducted in a custom Python-based environment, comparing TAFL-UWSN to standard FedAvg and Byzantine-resilient FL approaches under various attack conditions. TAFL-UWSN achieved a model accuracy exceeding 92% with up to 30% malicious nodes while maintaining a false positive rate below 5.5%. Communication overhead was reduced by 28%, and energy usage per node dropped by 33% compared to baseline methods. The TAFL-UWSN framework demonstrates that integrating trust into FL enables secure, efficient, and resilient underwater intelligence, validating its potential for broader application in distributed, resource-constrained environments. Full article