From TimeFrequency to Angular Domain: A Transformative Paradigm for Massive MIMO Systems

Abstract

[EN] Wireless systems have traditionally organized transmission, estimation, and feedback on a time-frequency grid. While effective at moderate array sizes, this paradigm becomes difficult to scale to ultra-large MIMO: the amount of channel state information (CSI) grows rapidly with antenna count, and feedback and processing overhead can become impractical in wideband operation. This paper argues that much of this burden can be reduced by working in a representation that matches propagation structure. By mapping channel information to the angular domain, the channel can be described by a small number of dominant propagation paths, which often remain limited even as arrays grow. Building on this idea, we compare conventional FFT-based angular projection with two adaptive low-rank alternatives based on partial singular value decomposition (SVD): an iterative Krylov-subspace method and a randomized SVD scheme. Unlike fixed-grid FFT compression, the SVD-based approaches estimate the dominant subspaces directly from channel observations, mitigating basis mismatch in near-field or irregular scattering conditions. We analyze complexity and protocol implications, and evaluate performance using ray-traced simulations at cmWave and mmWave frequencies. In the considered scenarios, partial SVD methods approach full-CSI capacity while substantially reducing feedback overhead, with the randomized variant offering the most favorable accuracy-latency trade-off.