Hop-Aware CSI Fusion for Frequency-Asynchronous SRS in TDD Massive MIMO

Abstract

Wideband time-division duplexing (TDD) massive multiple-input multiple-output (MIMO) precoding requires channel state information (CSI) that is both frequency-selective and temporally fresh. In practical fifth-generation New Radio networks, uplink sounding reference signal (SRS) resources are often sparsely allocated and configured with frequency hopping to reduce pilot overhead and support multi-user multiplexing. As a result, SRS measurements over different frequency segments are obtained at different sounding occasions and stitched at the next-generation NodeB (gNB), leading to frequency-asynchronous SRS observations with frequency-dependent CSI aging. This letter proposes a hop-aware CSI fusion framework for frequency-asynchronous SRS in TDD massive MIMO systems. The proposed method reconstructs physical resource block-resolution CSI by combining quantized subband CSI feedback with stitched SRS observations at the gNB. To explicitly model frequency asynchrony and observation reliability, the decoder incorporates per-physical resource block measurement age and uplink signal-to-noise ratio (SNR) as side information, and further introduces precoding resource group (PRG)-level residual gating to adaptively regulate SRS-driven refinement under stale or unreliable frequency segments. In addition, a compact subband feedback representation augmented with intra-subband variation features is designed to enhance frequency selectivity under limited feedback payload. Simulation results show that the proposed framework consistently improves PRG-level squared generalized cosine similarity across different SRS periodicities, feedback payload sizes, and user mobilities.