Refined Bayesian Optimization for Efficient Beam Alignment in Intelligent Indoor Wireless Environments

TL;DR

The paper tackles fast, reliable beam alignment in indoor mmWave/6G environments where exhaustive training is costly due to mobility and reflections. It introduces Refined Bayesian Optimization (R-BO), which combines a Gaussian Process surrogate with a Matérn kernel, Expected Improvement acquisition, and a localized refinement stage to exploit structured angular power distributions. Adaptive online hyperparameter re-optimization makes R-BO robust to non-stationary angular maps. Experimental results in a 60 GHz indoor lab show 97.7% alignment within 10° and <0.3 dB loss with 88% fewer probes than exhaustive search, outperforming ROMP under realistic conditions.

Abstract

Future intelligent indoor wireless environments require fast and reliable beam alignment to sustain high-throughput links under mobility and blockage. Exhaustive beam training achieves optimal performance but is prohibitively costly. In indoor settings, dense scatterers and transceiver hardware imperfections introduce multipath and sidelobe leakage, producing measurable power across multiple angles and reducing the effectiveness of outdoor-oriented alignment algorithms. This paper presents a Refined Bayesian Optimization (R-BO) framework that exploits the inherent structure of mmWave transceiver patterns, where received power gradually increases as the transmit and receive beams converge toward the optimum. R-BO integrates a Gaussian Process (GP) surrogate with a Matern kernel and an Expected Improvement (EI) acquisition function, followed by a localized refinement around the predicted optimum. The GP hyperparameters are re-optimized online to adapt to irregular variations in the measured angular power field caused by reflections and sidelobe leakage. Experiments across 43 receiver positions in an indoor laboratory demonstrate 97.7% beam-alignment accuracy within 10 degrees, less than 0.3 dB average loss, and an 88% reduction in probing overhead compared to exhaustive search. These results establish R-BO as an efficient and adaptive beam-alignment solution for real-time intelligent indoor wireless environments.

Figures (9)

• Figure 1: Measured AoA–AoD relative power map for an indoor environment under exhaustive search. The mainlobe and multiple sidelobes illustrate the non-convex angular power landscape. Measurements were obtained using 16 TX and 16 RX antennas operating at 60 GHz in an indoor laboratory environment at location ID RX23.
• Figure 2: Posterior predictive uncertainty ($\sigma$) map after the Bayesian optimization phase (before refinement). Warmer colors denote regions of higher model uncertainty, while cooler areas indicate well-explored angular regions. Black circles mark the probed beam pairs.
• Figure 3: GP posterior mean (predicted power map) for RX23 after R-BO refinement. The red star marks the final beam; the dashed box shows the $\pm10^{\circ}$ local refinement region.
• Figure 4: Room geometry and experimental setup in the Laboratory. The TX was fixed near the center (grid index 53), while the receiver was moved sequentially along marked grid positions to capture spatially distributed beam-sweep data.
• Figure 5: Average relative power loss versus Bayesian optimization iterations ($n_{\text{init}}{=}15$) compared to exhaustive search. The loss decreases sharply within the first 30 iterations and stabilizes below 0.3 dB after about 50 iterations, indicating convergence.