Near-Field MIMO Channel Acquisition: Geometry-Aided Feedback and Transmission Design

TL;DR

This paper presents a geometry-aided feedback and transmission scheme for near-field MIMO systems that efficiently acquires CSI with minimal pilot overhead by leveraging known antenna geometries and angular parameters.

Contribution

It introduces a compact NF LoS MIMO channel parameterization using two AoDs and a relative rotation angle, reducing the need for fine-grained distance grids and enabling efficient CSI feedback.

Findings

As few as four pilots are sufficient for accurate CSI estimation in dominant LoS conditions.

A two-stage precoding method improves robustness against noise and NLoS components.

The proposed scheme achieves high data rates with fewer OTA iterations, nearing perfect CSI performance.

Abstract

Near-field (NF) line-of-sight (LoS) MIMO systems enable efficient channel state information (CSI) acquisition and precoding by exploiting known antenna geometries at both the base station (BS) and user equipment (UE). This paper introduces a compact parameterization of the NF LoS MIMO channel using two angles of departure (AoDs) and a BS-UE relative rotation angle. The inclusion of the second AoD removes the need for fine-grained distance grids imposed by conventional NF channel parametrization. To address the user-specific uplink pilot overhead in multiuser NF CSI acquisition, we propose a scheme that uses a fixed, UE-independent set of downlink pilots transmitted from a carefully selected subset of BS antennas. In dominant LoS conditions, as few as four pilots suffice, with Cram\'er-Rao bound (CRB) analysis confirming that increased antenna spacing improves estimation accuracy. Each UE estimates and quantizes its angular parameters and feeds them back to the BS for geometry-based CSI reconstruction, eliminating the need for full channel feedback. To enhance robustness against noise, quantization errors, and non-line-of-sight (NLoS) components, we introduce a two-stage precoding method. The initial precoding is computed from estimated LoS CSI and refined through bidirectional over-the-air (OTA) training. Furthermore, a two-step stream allocation strategy reduces pilot and computational overhead. Simulations demonstrate that the proposed approach achieves high data rates with significantly fewer OTA iterations, approaching the performance of perfect CSI.