Driven by the development of wireless sensing, human activity recognition (HAR) based on WiFi channel state information (CSI) has attracted immense attention owing to its low-cost, device-free, and non-intrusive characteristics. Conventional approaches heavily rely on time-frequency analysis to generate spectrograms. However, fixed-window time-frequency transforms involve an inherent trade-off between temporal and frequency resolution, while residual noise and artifacts may obscure motion-related structures in the spectrograms. To tackle these challenges, we propose EWiAR, a framework designed for high-precision HAR. At its frontend, an Enhanced Spectrogram Analysis Network (ESAN) is introduced to mitigate resolution trade-offs. It leverages a multi-scale feature extraction (MSFE) bank to dynamically capture varying frequency resolutions, followed by a spectrogram refinement module (SRM) that actively purifies the representations from background interference. Rather than using standard classifiers, the backend employs a Time-Frequency Cross-Attention Transformer (TFAiT) to process these spectrograms. TFAiT orchestrates a dual-branch learning mechanism that concurrently models temporal motion dependencies via a bidirectional LSTM and extracts global spectral patterns using a vision transformer. These complementary features are fused through a dedicated Time-Frequency Cross-Attention mechanism. Comprehensive experiments across synthetic signals and three public datasets demonstrate that EWiAR achieves superior overall performance compared to eleven mainstream benchmarks. Specifically, it delivers accuracies of 98.6% on WiAR, 99.1% on UT-HAR and 98.9% on NTU-HAR, firmly validating its exceptional effectiveness for complex indoor environments.
Medium access control is witnessing renewed attention over the last decades, due to the Internet of Things and to the extensive development of advanced signal processing techniques, able to significantly expand the capacity of shared channel systems. This paper addresses a multiple access model encompassing Successive Interference Cancellation (SIC) at the receiver, aiming at establishing the stable achievable spectral efficiency. Both the transmission probability p and the target Signal-to-Interference-plus-Noise Ratio γ are adjusted dynamically by nodes, according to the overall number of backlogged nodes. Leveraging a Foster–Lyapunov approach and bounds on the number of decodable packets under a general SIC model, it is proved that the achievable stable spectral efficiency vanishes for fixed p and γ. It vanishes as well if p and γ scale with the number of backlogged nodes in the wrong way. It is proved that only the product p⋅γ matters and that a non-vanishing stable spectral efficiency is achieved, in case it scales inversely proportional to the number of backlogged nodes. Numerical evaluation demonstrates the trade-offs between stability, convergence to the asymptotic regime and mean access delay.
In vehicle-to-everything (V2X) communication, complex road conditions and tall buildings often obstruct the signals. This often disrupts the ground communication link, making it impossible to achieve full signal coverage over the entire area. While deploying aerial platforms provides a viable solution, a single-layer aerial architecture struggles to simultaneously accommodate wide-area coordination and localized high-density access. Facing these challenges, this paper considers a heterogeneous aerial–ground architecture comprising a high-altitude platform (HAP) for wide-area management and low-altitude platforms (LAPs) for flexible local access. This paper focuses on analyzing the transmission capacity of LAP-assisted local V2X communications. Specifically, we construct a three-layer heterogeneous network consisting of the HAP, LAPs and ground users. We also design a hybrid transmission strategy that can sense the scene. It can automatically switch between the two modes of direct ground communication and relay communication in the air, ultimately achieving the maximum utilization of the spectrum. To facilitate rapid mode switching under severe Doppler shifts caused by high-speed vehicle movement, we integrate a low-latency channel state information (CSI) tracking mechanism based on the least squares (LS) criterion, ensuring real-time responsiveness without the heavy computational overhead of complex estimators. Additionally, utilizing Palm theory and Laplace transform, we derive the LAP existence probability, expectation of horizontal relay link distance, and general form of successful transmission probability. Finally, simulation results demonstrate that the proposed scheme outperforms state-of-the-art approaches in transmission capacity, exhibiting favorable robustness particularly in high-density node and high-speed mobility scenarios.
Low-altitude UAV networks rely on onboard sensors for navigation and coordination, making them vulnerable to attacks such as GPS spoofing. Spoofed navigation data can propagate through neighbour updates and coordination messages, creating a cross-layer attack pathway, although communication links may remain unaltered. This paper presents a decentralized TinyML-guided integrity-aware communication scheduler that couples a physics-guided detector with source-level transmission control. Using 25 cross-sensor and temporal features, the detector achieves 95.71% accuracy, 99.85% attack recall, and a 96.16% F1-score on a flight-disjoint test set. The framework is evaluated in OMNeT++/Simu5G across 54 scenarios covering three trajectory classes and six UAV formations. For the Straight Matrix-49 scenario with seven compromised UAVs, an undetected attack increases compromised-source peer deliveries by 42.9% relative to normal operation. The proposed TinyML-guided framework reduces compromised-source packet generation and the corresponding peer-update traffic by 30% relative to the undetected-attack condition, restoring the compromised-source transmission rate to its normal operating level. The results further show that conventional Age of Information (AoI) can decrease during spoofing, indicating that freshness alone does not guarantee information integrity. These findings demonstrate the potential of the proposed framework to enhance the resilience of low-altitude multi-UAV systems operating in compromised environments.
