The Wi-Fi-enabled ambient IoT exhibits high asymmetry between uplink and downlink speeds. The uplink can reach up to 1 Mbps, while the downlink throughput is around 100 Kbps. In this paper, we present Wook, a novel high throughput downlink system to empower Commercial Off-The-Shelf (COTS) Wi-Fi devices to transmit high-speed OOK messages. The key innovation underpinning Wook is its ability to achieve sub-symbol level modulation, allowing a single OFDM symbol to carry multiple OOK bits. This is done by profoundly exploring the Wi-Fi PHY layer and identifying optimal input payload to achieve fine-grained Wi-Fi waveform manipulation. We fabricate a PCB prototype and employ the COTS Wi-Fi router to implement the entire system. Experimental results show that, with a simulated IC power consumption 88.2μW, Wook achieves a data rate of up to 1.1 Mbps, an 8.9X improvement over state-of-the-art systems. Moreover, even at a communication distance of 95 m, Wook maintains a throughput of 82.9 Kbps. Thanks to its native time-division multiplexing capability, Wook also supports up to five concurrent connections.
Uncrewed aerial vehicles (UAVs) are increasingly deployed as mobile edge computing agents for mission-critical post-disaster sensing. However, denoising and feature extraction from acoustic data for potential life-sign detection in earthquake ruins are challenging due to stochastic air-ground links, limited onboard processing capabilities, and constrained UAV energy budgets. Existing trajectory design and online control methods typically assume deterministic links or incur high online complexity. To address these challenges, we investigate the joint optimization of trajectory design and onboard service control, formulating it as a mixed-integer nonlinear fractional programming problem to maximize computational throughput efficiency (CTE). We propose a spatio-temporal decoupling framework termed Bisimulation-Guided Trajectory and Service Control (BiG-TSC). An offline-trained Bisimulation Quotienting for Neural Combinatorial Optimization model infers reference trajectories from the acoustic sensor node (ASN) topology at mission time. Task-cost-aware allocation then partitions the inferred order using expected retrieval and processing costs to minimize the longest predicted route duration under per-UAV budgets. The online Lyapunov controller makes stay-or-skip decisions using an efficiency-aware drift-plus-penalty rule based on estimated success probabilities and virtual backlogs, avoiding excessive retries at persistently blocked ASNs. Theoretical analysis bounds the local service-overhead gap and virtual backlog under the stated recurrent-epoch assumptions. Simulations with measured processing profiles show higher CTE than Always-Stay and Fixed-Retry under the default scale settings. Task-cost-aware allocation reduces mean mission completion time by up to 9.6% relative to equal-count allocation in the tested fleet settings.
Unmanned aerial vehicle (UAV) anti-jamming design is critical for ensuring secure and reliable communications in complex electromagnetic environments. While deep reinforcement learning (DRL) enables adaptive solutions, its black-box nature limits policy interpretability and effective human oversight, hindering trustworthy human–AI collaboration. This paper formulates the UAV–jammer confrontation as a zero-sum Markov game and proposes the Symbolic Hybrid Intelligence Enhanced Learning Defense (SHIELD) framework, which integrates DA-TD3-based adversarial policy learning with symbolic artificial intelligence. The framework jointly optimizes beamforming and three-dimensional trajectory decisions under power, flight-region, mobility, and collision-avoidance constraints while providing interpretable decision support. SHIELD employs a symbolic–neural synergistic architecture, in which the Symbolic Mapping Mechanism (SMM) translates continuous state–action pairs into first-order logic predicates, the Strategy Experience Graph (SEG) organizes symbolic experience into an interpretable decision graph, and the Intent-driven Action Guidance (IAG) exploits accumulated experience for policy guidance while enabling human-on-the-loop supervision through declarative constraints without requiring per-step human intervention. Simulation results show that SHIELD achieves a higher system sum rate and faster convergence than baseline methods. IAG experiments further demonstrate that human intent can be incorporated online to eliminate undesirable states and improve the overall reward without retraining the neural policy.
Online Virtual Reality (VR) experiences, in which the content is rendered by a remote server and streamed to the user's Head Mounted Display (HMD), are increasingly common in both entertainment and work training settings. However, traditional password-based authentication is often unwieldy for VR devices, where interaction is based on the physical actions of the user, in addition to being prone to known security issues. In this work, we propose a behavioral authentication pipeline based on users' movement patterns and the resulting network traffic, presenting a complete threat model and showing that appropriate design choices make such a pipeline robust against most attacks that come short of fully compromising the rendering server. We analyze the proposed scheme's authentication performance over a real dataset, showing false positive rates below 1% for 20 s of gameplay and discuss the explainability of the outcomes.
In practical applications, the transmit power of the target node and the absorption coefficient of the water can be unknown, which results in challenges for accurate underwater localization. This paper proposes an underwater acoustic localization technique with unknown transmit power and absorption coefficient. A first-order Taylor expansion and a ratio-based approach are applied to the received signal strength measurements to derive a distance-based formulation that eliminates the influence of transmit power and environmental absorption. Motivated by the bias–variance tradeoff, a minimization problem is developed to jointly reduce the squared bias and variance of the location estimate, thereby reducing the mean squared localization error compared with conventional unbiased estimation. The minimization problem is solved through semidefinite programming to obtain an estimate of the estimator's bias. The proposed technique first obtains an initial location estimate using linear least squares, and then employs the bias estimate obtained from the minimization problem to update the location estimate. The theoretical properties of the minimization problem are analyzed by establishing the optimality conditions and the bias sensitivity. Extensive simulations are conducted to evaluate the performance of the proposed technique, demonstrating superior localization accuracy in comparison with existing methods.
