This work presents the design, implementation, and experimental validation of a 2 × 2 circular optical reflective intelligent surface (C-ORIS) for dynamic indoor visible light communication (VLC) systems. The C-ORIS autonomously performs RSS-guided beam alignment for a slowly moving receiver present in a non-line-of-sight (NLoS) configuration relative to a fixed transmitter.A closed-loop RSS-guided control circuit dynamically adjusts each mirror’s tilt to maximize the received optical power and spectral efficiency (SE). The system tracks the receiver over ±60°, re-establishing the link in approximately 3 seconds for movements from 15° to 60°, with shorter times for smaller shifts. The experimental setup uses a blue light-emitting diode (LED)-array transmitter, a PDA36A2 photodiode (PD) receiver, a Universal Software Radio Peripheral (USRP), and microcontroller-driven control at both the receiver and C-ORIS for precise mirror alignment. SE is measured as a function of receiver distance and benchmarked against theoretical simulations, showing that the NLoS configuration at a 30° beam steering angle achieves the highest SE, surpassing even the LoS case.
The lognormal distribution is widely used to model shadowing and atmospheric turbulence. However, the analytical intractability of the lognormal distribution hinders closed-form performance analysis. In this paper, we develop a reciprocal beta (RB) approximation for lognormal shadowing by minimizing the Kullback–Leibler (KL) divergence. A safeguarded Newton-type algorithm is proposed to determine the KL-optimal RB parameters, whereas moment matching provides a closed-form reference. Comparisons with Gamma, inverse-Gamma, inverse-Gaussian, and product-form approximations show that the product-form mixture achieves the highest accuracy among the models compared. RB retains a compact two-parameter density and a lower per-output evaluation cost. It also provides tractable expressions for outage probability and average bit error rate (BER) of differential binary phase-shift keying (DBPSK).
Physical-Layer Security (PLS) is attractive for Internet of Things (IoT) and short-packet links where security features are often optional or disabled to minimize computational and power cost. While classical wiretap coding achieves information-theoretic security, it relies on the eavesdropper’s channel being degraded compared to the legitimate receiver, an assumption that cannot be guaranteed in realistic scenarios. In this work, we present a channel-independent computational-hardening mechanism embedded in the channel codes. We equip Non-Binary Polar Codes (NBPC) with two configurable components: three independently selectable kernel elements at every polarization branch and Finite Bi-Right-Quasigroup (FBRQ) arithmetic realized through configurable addition and multiplication tables. For q = 16 and Ns = 64, the raw logarithmic configuration count is approximately 3720 bits, compared with a conservative 61-bit upper bound for hidden I/R/F allocations. Simulations show no observed penalty of the configurable components on the code performance, while Eve’s Bit Error Rate (BER) under the evaluated configuration mismatch remains near 0.5 even at high SNR. An initial complexity analysis demonstrates significant area savings compared with higher-layer security schemes, highlighting the efficiency of the proposed scheme for IoT applications.
While direct sequence spread spectrum (DSSS) is foundational for low probability of detection (LPD) communications, its inherent cyclostationary signatures remain vulnerable to specialized feature detectors. This letter proposes and experimentally validates a novel LPD enhancement technique using a time-modulated reconfigurable intelligent surface (RIS). By inducing a time-varying phase shift at the RIS, a standard DSSS signal is transformed into an FM-DSSS waveform over-the-air. This mechanism provides a comprehensive enhancement to LPD by further broadening the spectrum to hinder energy detection, while substantially attenuating cyclostationary signatures to counter feature-based detection. Experimental results using a software-defined radio testbed confirm significant signature degradation for unauthorized detectors while maintaining reliable data recovery for the legitimate receiver.
Graph neural networks over air (AirGNNs) exhibit great potential for efficient distributed GNN training, which aggregates feature representations leveraging the over-the-air computation (AirComp) technique. Nevertheless, the issue of interference management within AirGNNs remains largely unaddressed. In this letter, we investigate transmission power control policies for devices to minimize the mean squared errors (MSEs) of the aggregated signals received at individual devices. Specifically, we formulate an online optimization problem to efficiently minimize long-term transmission error by optimizing the transmission power of transmitters and the denoising factor of each AirComp transceiver, subject to individual average power constraints at devices. To tackle this problem, we convert the long-term individual average transmit power constraint of each device into a queue stability constraint, thereby transforming the stochastic optimization problem into a single time-slot optimization formulation. The transformed problem is solved via convex optimization assisted by bisection search. Moreover, we theoretically prove that the defined queueing system is meanrate stable. Extensive experimental results on real-world datasets validate the effectiveness of our proposed methods.
