An Enhanced Polar-Domain Dictionary Design for Elevated BSs in Near-Field U-MIMO

TL;DR

This work tackles the challenge of near-field channel representation in sub-THz U-MIMO by departing from the common ground-plane assumption for grid design. It introduces a height-aware polar-domain dictionary design that uses level curves and NMSE-based optimization to construct a grid that accurately captures near-field channels when the base station is elevated above the ground plane. By integrating this grid with a P-SOMP channel estimator in a hybrid uplink system, the approach yields superior channel estimation accuracy and spectral efficiency compared with existing correlation-based designs, including at large dictionary sizes that enable angular oversampling. The methodology provides a practical path to improved near-field wireless performance in elevated BS deployments, with clear quantitative gains demonstrated in high-frequency simulations.

Abstract

Near-field U-MIMO communications require carefully optimized sampling grids in both angular and distance domains. However, most existing grid design methods neglect the influence of base station height, assuming instead that the base station is positioned at ground level - a simplification that rarely reflects real-world deployments. To overcome this limitation, we propose a generalized grid design framework that accommodates arbitrary base station locations. Unlike conventional correlation-based approaches, our method optimizes the grid based on the minimization of the optimal normalized mean squared error, leading to more accurate channel representation. We evaluate the performance of a hybrid U-MIMO system operating at sub-THz frequencies, considering the P-SOMP algorithm for channel estimation. Analytical and numerical results show that the proposed design enhances both channel estimation accuracy and spectral efficiency compared to existing alternatives.

Figures (8)

• Figure 1: Normalized correlation $\mu$ and grid size $Q$ as a function of the distance sampling factor $\beta$ with the design in cui2022channel.
• Figure 2: Level curves $\Gamma = g$ with seven different values of $g$ and BS height $b = \{0,15\}$ m.
• Figure 3: Example of the proposed grid construction with grid size $Q = 135$, number of level curves $N_\Gamma = 9$ and distance sampling factor $\beta = 0.57$.
• Figure 4: Proposed design with BS height $b = 15\,$m.
• Figure 5: Polar-domain grids obtained using the design in cui2022channel with $\beta = \{2.5,1.56\}$ and $Q = \{501,1298\}$, respectively.