Wireless access-control systems that rely on persistent identifiers, such as manufacturer-assigned MAC addresses, remain vulnerable to MAC address spoofing. This work presents a Cross-Layer Fingerprinting (CL-FP) approach that combines identity information from multiple protocol layers with device-specific physical-layer features to strengthen security verification. We illustrate CL-FP for detecting MAC spoofing, using Error Vector Magnitude (EVM) and the claimed MAC address. The MAC address identifies an enrollment record, and the EVM distribution verifies consistency with that record. In the controlled-distance studies, we compute hardware-related EVM by compensating for modeled noise and empirically estimated distance effects. Mutual information (MI) is used to assess feature relevance. We evaluate CL-FP through Monte Carlo simulations using a dataset of 16 bit-similar USRP X310 transmitters and one USRP B210 receiver. The EVM measurements include captures in a high signal-to-noise ratio (SNR) cabled setting where each transmitter is connected directly to the receiver, and over-the-air (OTA) captures at 2, 8, and 20 ft. In the cabled experiments, CL-FP achieves a true positive rate (TPR) of 97.5% and a true negative rate (TNR) of 95.2% using 500 EVM observations per device and a detection threshold of 6.17%. The threshold is calibrated using legitimate-device observations, without requiring labeled spoofing samples for training. When devices are enrolled at 2 ft and tested at 8 and 20 ft, TNR decreases to 68% and 34%, respectively, even after compensation. These results identify CL-FP cross-distance MAC-spoofing detection using a single enrollment distance as a limitation.
Physical layer security of discrete phase reconfigurable intelligent surface (RIS) enabled orthogonal time frequency space (OTFS) systems is analyzed under passive eavesdropping. We derive expressions for signal-to-noise ratios (SNRs) at the legitimate receiver (LRx) and at the eavesdropper (Eve) in terms of eigenvalues by exploiting the delay-Doppler (DD) domain characteristics of OTFS channel matrices. Thereafter, we develop new statistical models for SNRs based on gamma distribution. This involves modeling the uncompensated residual phase errors at RIS elements arising due to quantization. We then derive a new approximate analytical expression for secrecy outage probability (SOP) based on the statistical modeling to bring out the dependence of SOP on the number of levels used to quantize phases at RIS and the number of RIS elements. Asymptotic SOP analysis is presented to obtain secrecy diversity order and secrecy array gain. Generalization of the analytical framework to practical rectangular waveforms and fractional Doppler is also presented. We present numerical results to benchmark SOP performance obtained using quantized phase strongest DD cascaded channel response based RIS configuration against the worst case of random phase and the ideal case of coherent phase RIS configurations. Numerical results are also presented to corroborate our mathematical analysis and to elucidate the impact of number of quantization bits, number of RIS elements, target secrecy rate, presence of direct Tx-Eve channel, practical rectangular waveform and fractional Doppler on secrecy performance. Benchmarking against RIS-aided OFDM systems under high mobility and impact of channel estimation errors are also presented.
Computation-hungry, low-latency applications in mobile networks require efficient job allocation and migration in edge networks. Local renewable energy sources offer a promising pathway to power edge infrastructure sustainably, but their intermittent nature creates fundamental challenges in balancing service quality, operational costs, and environmental impact. We address this problem by optimizing two objectives: maximizing operator profit and balancing monetary gains with environmental sustainability through a parametrizable utility function. We propose both a heuristic and DRL-based algorithms that adapt to varying network conditions and intermittent energy availability. Evaluations using real renewable energy traces demonstrate near-optimal performance across all tested scenarios, outperforming state-of-the-art approaches. We publicly release our code to foster reproducibility and further research.
Accurate indoor localization is essential for the reliable operation of Automated Guided Vehicles (AGVs) in industrial environments. Visible Light Positioning (VLP) offers a cost-effective and robust alternative to conventional Radio Frequency (RF)-based technologies, yet its practical deployment requires both efficient positioning algorithms and a systematic approach to infrastructure design. This paper presents a VLP system designed for AGV localization. We propose a direct two-Light Emitting Diodes (2-LED) geometric positioning algorithm that determines the vehicle’s 2D pose (position and yaw) using as few as two LED beacons. The system’s performance was evaluated through static, dynamic, and industrial trials executed across operational spaces with transmitter deployment heights ranging from 2.2 m in laboratory setups to 4.2 m in industrial facilities. In static laboratory tests conducted within a physical evaluation area of 6.75 m2, the proposed algorithm achieved a mean positioning error of 2.97 cm, outperforming a standard solvePnP benchmark (4.47 cm). This performance advantage was maintained in dynamic industrial scenarios, where the proposed algorithm yielded a mean error of 3.81 cm along straight-line paths and 4.46 cm along zigzag trajectories. To support physical deployment, we model and optimize light placement to maximize effective coverage while accounting for vehicle dynamics and receiver orientation. The optimization methodology quantifies how lights layout and AGV yaw influence coverage bounds, enabling systematic infrastructure planning. The empirical findings indicate that combining an ultra-low latency 2-LED solver with a systematic layout methodology offers a precise and scalable approach for indoor AGV positioning under planar motion constraints.
Worst-case robust adaptive beamforming usually protects one nominal direction, whereas 5G and 6G channels can split the signal energy across a few clusters. When the cluster gains are independent and the whitened steering vectors are nearly orthogonal, as in the tested geometries, the max-SINR beamformer is close to the strongest-cluster beamformer, so a multi-cluster design gains little in expected SINR and is useful mainly for protecting every cluster. Multi-primary robust adaptive beamforming (MPR-RAB) takes its gain targets and phases from the max-SINR beamformer of the estimated model, which it reproduces at zero margin when the target and objective covariances coincide, and adds a worst-case margin on every cluster, so one constant trades expected SINR for per-cluster protection. The resulting second-order cone program, solved at every update by an active-set Karush-Kuhn-Tucker method with numerical convergence checks, yields a linearly constrained beamformer loaded by its margins. It runs on a reconstructed interference covariance and tracks cluster directions in azimuth and elevation without interferer side information. On an 8 × 8 planar array over nine scenarios and six conditions, with every practical baseline using the tracked directions and all tuned parameters selected on separate seeds, MPR-RAB is 0.33 dB above the max-SINR beamformer and 0.56 dB below the oracle in expected SINR, and its minimum-cluster SINR is 11.3 and 19.1 dB above those of the max-SINR and strongest-cluster beamformers (95% confidence intervals, 30 runs). On a 16-element linear array it is 0.28 and 0.09 dB above the max-SINR and strongest-cluster beamformers in expected SINR.
