Semantic communications (SEMCOM) offer a paradigm shift by transmitting the intended meaning of messages using neural networks and shared knowledge, rather than relying on raw data. This is essential for next-generation applications such as autonomous driving and holographic telepresence, where accurate semantic interpretation is critical. However, the open and knowledge-centric nature of SemCom makes it susceptible to adversarial attacks, especially during knowledge sharing and decoding. To enhance security, we introduce SECO, a robust learning framework designed to defend against both targeted and untargeted adversarial threats. SECO combines sensitivity-aware noise detection with Huber loss-based filtering to identify and suppress adversarial perturbations, while preserving the integrity of natural signals. Experimental results on speech datasets under Rayleigh and Rician fading channels show that SECO boosts speech-to-distortion ratios by up to 2.2× and 2.5×, respectively, and reduces semantic disruption by up to 17%. It also demonstrates resilience to natural noise and mitigates fixed-pattern attacks by exploiting unique voice characteristics.

Unmanned Aerial Vehicles (UAVs) have been widely adopted in various applications, including surveillance, search and rescue, cartography, and damage assessment in remote areas. This study aims to enhance the safe navigation of multiple UAVs in adverse environments by integrating computer vision with deep reinforcement learning (DRL) and a route planning algorithm known as REAP. REAP incorporates a real-time object detection module that identifies obstacle positions from surveillance drone footage or satellite imagery. Additionally, A* and tangent point search algorithms are employed to compute the shortest paths while maintaining connectivity and safe distances from threats, such as dynamic obstacles (e.g., birds). Notably, this study also introduces an optimized DRL approach to further refine flight paths and reduce the latency in search result retrieval. Evaluation results show that the proposed method can compute optimal paths in under 10 milliseconds-three times faster than state-of-the-art (SOTA) methods-while ensuring UAV safety from collisions with flying obstacles, tall structures, or bird swarms.

Quantum machine learning (QML) holds immense potential for revolutionizing computational intelligence, yet faces significant challenges in optimizing quantum neural network architectures for practical implementation. This paper introduces Q-Drop, an innovative quantum adaptation technique that enhances parameterized quantum circuits (PQCs) for training performance. The system includes two novel optimization strategies: statistical pruning and dynamic dropout (circuit flipflops), which significantly improve the learning performance and robustness of quantum orthogonal neural networks. Evaluation results across standard (MNIST, Fashion-MNIST) and medical imaging (Pneumonia-MNIST, Retina-MNIST) datasets demonstrate substantial performance gains. Notably, the approach achieves up to 94.3 % accuracy on medical image classification and 98.7 % accuracy on standard image datasets, outperforming existing quantum machine learning methods. By addressing critical challenges in quantum neural network training, this work highlights the potential of applying quantum machine learning to enhance performance for classical computer vision tasks and QML-driven scheduling optimization in quantum networks.

Monocular 3D object detection offers a cost-effective alternative to LiDAR and stereo cameras by determining 3D positions from a single image. DETR-based methods leverage transformers to integrate visual and depth representations globally, achieving state-of-the-art performance and competitive speeds without manual configurations like non-maximum suppression or anchor generation. However, averaging multiple-depth predictions hinders precise accuracy. This work enhances depth prediction through two innovations: (1) improving depth quality using a Shape-aware Foreground Depth Map (SFDM) and (2) Depth Adaptive Weight (DAW) helps the final depth prediction to benefit flexibly from each component’s contribution. Experiments on the KITTI benchmark demonstrate the proposed model’s state-of-the-art performance. The code is available at https://github.com/useracc687/saw-monodetr.

Currently, online map navigation applications are essential to our daily demands, such as route finding in a big city or booking a taxi trip. Further, the applications also play a pivotal role in finding access to the nearest medical aid in big cities with complex traffic networks. However, existing commercial map programs are unable to provide route suggestions in areas where a continuous network connection is assured, which is essential for distant emergency medical services. This study introduces a highly effective personalized shortest-path algorithm for finding the fastest path of an emergency vehicle (EV) while also ensuring stable network connections for potential remote robotic surgery. By using a customized A-star algorithm, the experimental findings demonstrate that this technique can outperform current search approaches with a 5% shorter path length per 5km. For rescue missions in a large city, the enhancement is remarkable and has the potential to save lives. Due to high compatibility with the existing search algorithms, this method can easily be included as a critical feature for existing commercial map apps.

The rapid advancement of wireless networks towards Artificial Intelligence (AI)-driven solutions attracts many vendors to build resilient and intelligent capabilities for Open Radio Access Networks (O-RAN). However, besides the benefits of achieving flexibility and intelligence, openness in native AI-driven O-RAN functions is also the target of severe AI-related security threats, e.g., adversarial attacks. This work addresses the security matter for the AI-powered solutions in the physical layer of O-RAN, specifically within the context of deep reinforcement learning (DRL)-based resource allocation. We introduce a new adversarial attack variant that manipulates the environment parameters and misleads the agent's observation during the inference phase. The attack can cause incorrect allocation decisions and significant degradation in the transmission data rate. Our evaluation results show that the attack degrades user data and packet delivery rates by up to 40% and 77.74%, respectively, particularly in ultra-low-latency services. We also found that the major weakness of DRL-driven radio resource allocation is the environment observation stage, where a group of compromised users or jammers can spoof noises and signal power to mislead environment interaction. In our context, the proposed policy infiltration attack is the most efficient approach to cause sustained network inefficiencies or reduced throughput for benign users.

WiFi and cellular networks have become essential components of our modern lives. On the other hand, these connectivity technologies also enable many radio localization techniques and even user monitoring that offer convenience and valuables, such as object tracking in augmented reality or patient tracking in emergency calls. This paper describes a novel approach for recreating a restricted-access architectural drawing that takes advantage of passive radio localization on multiple users. Initially, we produce data points in an empty building drawing from the data points provided by our built-in signal-to-image-based localization. Next, we propose a few-shot drawing reconstruction model, DRNet, for regenerating the raw structure based on collected data points from localization. Finally, we combine the time-series monitoring data to predict the functionality of the rooms. The assessment findings indicate that the precision of accessible area regeneration can reach a maximum of 83.4%. The methodology used in this study exhibits versatility in its applicability to a wide range of prospective applications. For instance, law enforcement authorities possess the capacity to effectively ascertain the ingress/egress pathways or accessible locations for potential intrusion during a narcotics enforcement operation. Another scenario involves the disclosure of frequently visited locations by criminals or elucidating the purposes served by certain areas inside a restricted-access facility.

Connected and automated vehicles already are the main force in realizing the vision of intelligent transportation in smart cities. However, enabling broadband connectivity for vehicles brings up new threats of spreading fake information. By broadcasting false sharing data, an attack vehicle has the ability to cause nearby vehicles to get confused or possibly collide in catastrophic accidents. The present study presents a resilient fusion-based multimodal abnormal detection technique, referred to as FMAD. FMAD facilitates a fusion model based on Dempster- Shafer's theory to strengthen confidence in the final detection assessment of detection results from multiple vehicles. FMAD can determine whether a vehicle is spreading false maneuver information with up to 96.18 percent accuracy of confidence. Meanwhile, our method outperforms all existing approaches in terms of the reliability of the detection decision.

Autonomous cars are already the driving force behind achieving the goal of intelligent mobility in smart cities. However, providing internet access for automobiles introduces additional risks of distributing false information. By broadcasting fake sharing data, an attack vehicle has the potential to mislead surrounding cars and trigger catastrophic accidents. This work introduces a novel incremental learning model for misbehavior detection in vehicular communications. Unlike previous methods which mostly relied on the pre-trained model, in this work, we develop accumulative learning capability for the misbehavior detection engines at vehicles or distributed roadside units. The model can assist in obtaining new knowledge from continuous learning. The simulation results indicate that the proposed model outperforms the state-of-the-art studies by up to 12% in terms of detection accuracy and 45% faster in continuous processing, particularly in the single detector mode.

Vehicular networks are expected to be the key technologies in the age of intelligent transportation and connected intelligence. However, by broadcasting false maneuver information in vehicular networks (emergency brake, merging/changing lane), an attacker can cause many vehicles to be disoriented or even crash in severe accidents. This work introduces a robust misbehavior detection scheme, namely TRIMO, by exploiting multimodal learning from various independent data sources (e.g., camera, joint radar and communications in the sixth-generation (6G) mobile networks). TRIMO can determine whether a car is lying about its sharing data with up to 92.7 percent accuracy by examining the consistency of data from numerous sources.

Unmanned Aerial Vehicles (UAVs) have a wide range of capacities, including surveillance, rescue missions, geographi-cal mapping, and military uses. This work focuses on combining computer vision with route planning methods, namely DragFly, to ensure the safety of a UAV throughout its whole flying mission to a designated target. We use an advanced object detection algorithm to precisely identify the locations of obstacles and create an accurate geographical map. Two path-planning algorithms are introduced to improve flying safety. We enhance the A* algorithm to include searching in hazardous areas, allowing the UAV to stay a safe distance away from dangers such as anti-drones and preserve communication. Furthermore, our tangent point approach may provide extremely fast searching performance in small to medium map sizes. The evaluation results demonstrate that our approach can locate the shortest path in just 31.66 milliseconds without compromising safety.

Network intrusion detection systems (NIDSs) are vital for identifying security attacks and predicting early invasion attempts, which is essential for protecting the Internet. Recently, deep learning (DL) has made significant achievements in enhancing intrusion detection accuracy. Nevertheless, the practical implementation of high-complexity DL models is limited by the constrained computational capabilities of the Internet of Things (IoT) devices, e.g., home routers and IoT gateways. This article introduces a novel NIDS approach explicitly tailored for IoT networks, leveraging a lightweight DL model. During the data preprocessing phase, we use a spatially enriched data conversion technique to decrease the dimensionality of high-dimensional raw traffic variables. This helps to offset the problem of increased model complexity. Furthermore, when spatial relationships often exist in the data, we can simplify the learning architecture by utilizing state-of-the-art vision transformer techniques in the computer vision field that can substantially reduce model complexity. The experimental results indicate that the proposed method achieves outstanding accuracy up to 99.57% with high-volume traffic input. Moreover, the proposed method reaches substantial reductions in learnable parameters by 55.35% and 82.07%, along with a remarkable decrease in floating point operations (FLOPs) by 93.56% and 99.28% compared to existing studies. The outstanding achievement highlights the proposed method’s ability to balance model complexity and accuracy performance, making it extremely appropriate for deployment on IoT gateways with limited resources.

For critical applications such as emergency medical rescue missions or telehealth in remote areas, stable network connectivity is vital for patient state monitoring and proper temporary care. Unexpected connection interruption or network lag can cause trouble for skilled doctors in remote care centers to predict the progress of a patient’s condition. Network quality is variable in many areas because of signal power degradation (zones without purple coverage) in rural areas with many building obstacles. As a result, many current emergency services still rely on on-site first aid efforts. The idea of unmanned aerial vehicles (UAVs) serving as aerial relaying stations to provide connectivity for ground users has received much attention over the years. However, controlling UAVs via cellular networks is still a challenging issue. In this work, we consider the mission of dispatching UAVbased relaying stations as a path-planning scheme, where the UAVs go to planned locations and serve the EVs with a certain connectivity requirement. The core novelty of this work is a novel searching scheme that can suggest a deployment plan for the swarm of UAVs at the time of the EVs’ departure. The search is also robust for path planning with real-time applications or dynamic environments.

Wireless networks have become a vital factor in our modern life. However, besides providing connectivity for many civil applications, wireless networks are the subjects of radio localization and user tracking. This work introduces a novel technique that exploits passive radio localization on multiple users for regenerating a restricted-access building drawing. Initially, from the data points generated by our built-in signal-to-image-based localization, we generate data points in an empty building drawing. Then by estimating the absolute distance among nearby data points, we group data points by a rectangle which represents the accessible area by human walk. Finally, we connect the center points of the generated rectangles to illustrate the accessible areas. The evaluation results show that the overall accessible area regeneration accuracy can reach up to 80%. Our technique can apply to many potential applications. For example, the police can safely determine the entrance/exit routes or walkable places to intrude in the drug raid mission. Another case is to expose the common visit places of the criminals or reveal the functions of an area in the restricted-access building.

The Denial-of-Service (DoS) attacks have rapidly increased over the years, particularly from the Internet of Things (IoT) devices such as connected IP cameras/vehicles and IP door entries. As a result of a lack of strong security implementation, these low-cost IoT devices can become zombie bots by brute force (using a password dictionary) and malware injection. Thousands of such zombies have been a powerful tool to initiate DoS attacks that can consume any target network's bandwidth. Recently, Deep Learning (DL) based methods have achieved admirable performance to mitigate DoS attacks significantly. However, besides the high latency of processing huge volumes of incoming traffic, current D L-based methods are often designed based on crafting a model carefully, which costs the developers significant time and effort. This paper introduces a novel automated deep-learning (autoDL) scheme to automatically generate an efficient DoS defense model. The system can find the best suitable detection model and a detailed configuration that is lightweight enough to deploy at IoT gateways/wireless routers/programmable switches/edge servers near the attack sources. To our knowledge, this is the first attempt to develop such an autoDL platform for DoS filter generation. The evaluation results show that the defense system generated by autoDL can ease 99.7% of malicious traffic before they go out to the Internet with only 0.593 ms for request reaction, a promising performance compared to the literature.

Radio positioning is critical for many indoor applications, such as behavioral monitoring and autonomous robots. Mobile users, however, can also be exposed to surveillance risks due to this capability. This work presents a Spatial-Temporal Angle-Delay Analysis Scheme (STADAS) for massive MIMO wireless networks that can help the attacker to track a user without the need to enter buildings. First, we transform the channel state information (e.g., angle of arrival, time of arrival) from massive MIMO transmission gained over time into living Angle-Delay profiles (ADPs) with fixed objects (building walls, furniture) and a moving object (the mobile user). Second, a generative adversarial network learning model is used to remove distorted data points from Angle-Delay video frames. The processed ADPs are trained with a Deep Convolutional Neural Network (DCNN)-based model on estimating the user's location. Evaluations on an empirical dataset indicate that radio positioning capabilities in emerging wireless communication technologies such as mmWave MIMO can pose severe privacy and surveillance threats.

Directional antenna technologies are crucial to enhance high-speed data transmission in emerging wireless communications such as mmWave. These technologies can enable high-accuracy radio positioning by exploiting spatial-temporal signal processing in wideband beamforming space. However, the radio positioning technique potentially poses surveillance risks to mobile users, particularly being tracked illegally. This work presents a novel scheme to track a user in a building based on passively received signals. The scheme includes a Deep Convolutional Neural Network (DCNN) localization module to train on the accumulated channel impulse responses (CIR) and corresponding Angle-Delay profiles. The user's estimated locations from the DCNN localization are then refined with an Unscented Kalman filter (UKF) data fusion module to eliminate outlier data points. Simulations indicate that the proposed scheme can accurately regenerate the user trajectory, even without the attacker's physical intrusion into the building. This poses a new concern of surveillance risks in directional wireless communications, given their expected popularity in 5G and beyond.

Resolving traffic bottlenecks caused by emergency situations on the freeways has been a challenge. It often takes much time for the vehicles voluntarily to line up and exit the congestion spot quickly. Unlike existing traffic scheduling schemes, which often rely on traffic signal controllers or are specified for known bottleneck areas, this paper introduces an efficient decentralized traffic coordination method, namely ETRACO, for resolving temporary bottlenecks on the multi-lane freeways. Initially, based on the Vehicle-to-Vehicle (V2V) warning notifications about the congestion, the vehicles negotiate with neighbors to determine a suitable configuration (platoon leader, distance gap, velocity, platoon size) for forming platoons. After that, each platoon leader commands the platoon members to change lane under the condition that there is a safe space on the lane next to the current lane so that the platoon can move safely to that lane. The experimental results demonstrate that our approach can reduce up to 22% delay for the last few vehicles driving through the congestion area during the congestion period. Furthermore, the proposed approach also effectively reduce congestion time for new incoming vehicles, and maintain the fairness for vehicles to leave the congestion area.

Dispatching emergency vehicles (EVs) to fatal accidents or fires as fast as possible is vital to save lives; however, minimizing an EV’s travel time to the rescue spot is still an open challenge. This work presents a path planning and traffic clear-out scheduling scheme to lessen the EV’s travel time in the vision of the Internet of Vehicles (IoV). Initially, the system searches a list of candidate paths to the rescue spot with the estimated time of arrival (ETA) at the EVs’ maximum speed, regardless of the traffic conditions. After that, a vehicle clear-out process evaluates the delay time of clearing out the traffic obstacles on each path to identify the fastest driving path. Finally, the system estimates and issues the signal preemption schedules for the junctions of the selected route to coordinate the traffic flows and let the EV pass through smoothly. From the macro perspective, this work seeks tocontrol the dynamic traffic proactively to reserve a lane for the EV– a feasible approach in the future of connected vehicles. This controllable model apparently contrasts with the conventional techniques of finding the least-cost paths with the uncertainty of traffic state prediction or traffic light preemption alone at the intersections. The simulation shows that our approach can outperform the state-of-the-art solutions in terms of the EV’s travel time reduction, particularly if the congestion or heavy load road segments appear on the selected route but far from the departure location of the EV.

In the next decades, the popularity of Internet of Vehicles (IoV) technologies and autonomous driving promises to fundamentally change the way of handling the traffic flow on the streets. Traffic separation can be entirely carried out from a remote traffic control center without the police. This work introduces a sequential coordination algorithm, namely SCA, to form and sort platoons of vehicles to quickly exit traffic bottleneck areas caused by temporary situations, such as vehicular accidents and a slow tractor. By exploiting maneuver information from IoV data sharing, SCA schedules the vehicles in a queue by their arrival and lane priority and then instructs them to safely drive through in order. The experimental results demonstrate our approach can reduce up to 32% waiting time for the vehicles to exit accident spots.

Vehicle platooning is a promising technology to bring up significant benefits of improved fuel economy and fewer traffic collisions. However, many security attacks such as beacon message falsification have been exposed, creating grave concerns about maintaining a vehicle platoon stably. This work introduces a robust positioning-based verification scheme, namely PVS, to enhance reliability of vehicle platoon control in vehicular networks. By exploiting geographic and maneuver information from 5G radio-based positioning, PVS can detect whether a vehicle is honest in reporting its location for platoon joining preparation or collision avoidance, with up to 96% accuracy.

Charging of a large number of electric vehicles (EV) at the same time raises several technical problems or can have significant impacts on power systems like high peak power consumption. During daytime, when EVs are connected at the different charging stations at the same time, they can cause a congestion problem in the network. The paper proposes charging strategies to reduce the peak load. These strategies are based on the interruption (on/off) or the modulation of EV charging power. By dividing the daytime in many intervals, a binary linear programming combined with the bisection scheme is used to manage the charging plan of the vehicles. The first application is used for reducing peak load in daytime. The second application is proposed to limit the charging power during a time defined by DSO (or TSO). The performance of the proposed strategies is validated by simulations for a charging station of 50 electric vehicles with a fast calculation time. The results obtained show that peak power can be reduced by more than 50 per cent.