Beamspace Dimensionality Reduction for Massive MU-MIMO: Geometric Insights and Information-Theoretic Limits

TL;DR

The paper analyzes beamspace dimensionality reduction for scalable massive MU-MIMO, revealing why fixed-size per-user beamspace windows with LMMSE can maintain high performance as the number of antennas and users grows. It establishes a geometric framework showing that a small beamspace window can capture most of a desired path's energy while interference concentrates into a few dominant eigenmodes when users are spatially separated, enabling effective linear suppression with modest noise amplification. A lower bound on SINR and the SIR margin translate into concrete design guidelines for user loading, guard intervals, and power control, and the framework extends to wideband MIMO-OFDM with per-subcarrier beamspace LMMSE, validated against measured $28$ GHz data with information-theoretic benchmarks. The work also discusses the impact of zeropadding FFTs and provides practical implications for scalable DSP implementations in mmWave and higher-frequency regimes.

Abstract

Beamspace dimensionality reduction, a classical tool in array processing, has been shown in recent work to significantly reduce computational complexity and training overhead for adaptive reception in massive multiuser (MU) MIMO. For sparse multipath propagation and uniformly spaced antenna arrays, beamspace transformation, or application of a spatial FFT, concentrates the energy of each user into a small number of spatial frequency bins. Empirical evaluations demonstrate the efficacy of linear Minimum Mean Squared Error (LMMSE) detection performed in parallel using a beamspace window of small, fixed size for each user, even as the number of antennas and users scale up, while being robust to moderate variations in the relative powers of the users. In this paper, we develop a fundamental geometric understanding of this ``unreasonable effectiveness'' in a regime in which zero-forcing solutions do not exist. For simplified channel models, we show that, when we enforce a suitable separation in spatial frequency between users, the interference power falling into a desired user's beamspace window of size $W$ concentrates into a number of dominant eigenmodes smaller than $W$, with the desired user having relatively small projection onto these modes. Thus, linear suppression of dominant interference modes can be accomplished with small noise enhancement. We show that similar observations apply for MIMO-OFDM over wideband multipath channels synthesized from measured 28 GHz data. We propose, and evaluate via information-theoretic benchmarks, per-subcarrier reduced dimension beamspace LMMSE in this setting.

Figures (13)

• Figure 1: Uplink massive MU-MIMO system model with beamspace dimensionality reduction. We assume far field for each path reaching at the antenna array.
• Figure 2: Energy in the spatial frequency-frequency grid for a single user with 48 paths including the LOS, the BS has $N=128$ antenna elements. The center frequency is 28.5 GHz and the fractional bandwidth is 20$\%$. The channel vector at each frequency is constructed using path-loss, delay, and AoA parameters drawn from the 28.5 GHz channel dataset charbonnier2020calibration.
• Figure 3: Desired energy ratio with and without zeropadding in an interference-limited regime. (a) Fixed arrival angle ($\theta^\circ=15$), so the spatial-frequency offset to the DFT grid varies with $N$. (b) Fractional off-grid offset held constant ($\delta=0.3$) across $N$. (c) $E_{W,N}(\delta)$ and $E^{(\mathrm{zp2})}_{W,N}(\delta)$, for $N=128$.
• Figure 4: Fractional energy capture $\eta(W)$ in a noise-limited regime with and without zeropadding. (a) $\theta^\circ=15$ for all array sizes considered. (b) $\delta=0.3$ for all array sizes.
• Figure 5: Cosine similarities between the desired user's and the interferer's beamspace response, where the interferer's spatial frequency is swept in $[-\pi,\pi)$. For both figures, $N=128$, $W=5$. (a) The desired user's spatial frequency is at $\delta=0.1$ (b) at $\delta=0.5$, where $\delta$ is defined in the base $N$-point DFT grid.

Theorems & Definitions (9)

• Theorem 1: Lower bound on energy capture with optimal $W$-bin contiguous window placement
• Corollary 2: Energy capture with four-bin window
• Proposition 3: Energy capture with $2\times$ zeropadding
• Theorem 4: Lower bound on expected SINR via mean interference covariance matrix
• Proposition 5: Mean interference + noise covariance
• Theorem 6: Linear Scaling of $\mathrm{SIR}_{\mathrm{margin}}$
• proof
• proof
• proof