Approximate Gram-Matrix Interpolation for Wideband Massive MU-MIMO

Zequn Li, Charles Jeon, Christoph Studer

Abstract: Wideband massive multi-user multiple-input multiple-output (MU-MIMO) systems face significant computational challenges due to the requirement of per-subcarrier Gram matrix calculations. This paper explores low-complexity interpolation techniques—specifically 0th and 1st-order schemes—that exploit the "channel hardening" effect. Our results demonstrate that these approximate methods can achieve near-optimal performance with substantially reduced complexity, offering greater robustness to imperfect channel-state information (CSI) than traditional exact interpolation methods.

Introduction

Massive MU-MIMO is a cornerstone technology for 5G and beyond, utilizing large antenna arrays at base stations (BS) to serve multiple users simultaneously. In wideband systems employing Orthogonal Frequency-Division Multiplexing (OFDM), the BS must perform data detection and precoding for thousands of subcarriers. A primary bottleneck in high-performance algorithms is the computation of the Gram matrix ($G_\omega = H_\omega^H H_\omega$) for every subcarrier $\omega$. For a typical system with 128 antennas and 16 users, this computation can consume more than half of the total processing power, making real-time implementation difficult.

Proposed Interpolation Methods

To address this complexity, we propose moving away from exhaustive per-subcarrier computations toward approximate interpolation. The core idea is to compute the Gram matrix exactly for only a small fraction of "base-point" subcarriers and estimate the rest using simple schemes:

  1. 0th Order Interpolation (Nearest Neighbor): The Gram matrix for any given subcarrier is approximated by the value of the nearest computed base point.
  2. 1st Order Interpolation (Linear): An entry-wise linear interpolation is performed between the two nearest computed base points.
Visualization of 0th and 1st-order interpolation of the Gram matrix entries across subcarriers.

The Phenomenon of Channel Hardening

The success of these simple approximations is mathematically supported by "channel hardening." As the number of base station antennas ($B$) increases, the channel matrices become more predictable and the Gram matrix entries vary more smoothly across the frequency domain. This allows linear approximations to capture the underlying channel dynamics with high fidelity, a property that does not hold for traditional small-scale MIMO systems.

Performance and Robustness

We evaluated these methods using both theoretical i.i.d. Rayleigh fading models and the realistic, geometry-based QuaDRiGa channel model. A critical insight from our analysis is that approximate interpolation is often superior to "exact" DFT-based reconstruction in practical settings. Exact methods rely on matrix inversions that become ill-conditioned under noise, leading to significant performance degradation when channel-state information (CSI) is imperfect.

Bit-error rate (BER) performance comparison under various channel conditions and CSI accuracy.

Key Results

  • Complexity Reduction: 1st-order interpolation can achieve the same Signal-to-Noise Ratio (SNR) performance as brute-force computation (within 1 dB) while requiring only 45% of the computational complexity.
  • Robustness to Imperfect CSI: In scenarios with estimation errors, the proposed 1st-order approximation maintains a low error floor (BER $< 10^{-4}$), whereas exact interpolation methods hit a high error floor near $10^{-1}$.
  • Analytical Accuracy: Derived closed-form expressions for the Mean-Squared Error (MSE) of these approximations match simulation results perfectly, providing a reliable framework for system design.

Conclusion

Approximate Gram-matrix interpolation offers a practical and robust solution for wideband massive MU-MIMO. By leveraging channel hardening, these schemes significantly reduce the computational burden of base station processing without sacrificing performance in realistic, imperfect channel conditions.

Summary updated

BibTeX
@article{li2020approximate,
  author = {Zequn Li and Charles Jeon and Christoph Studer},
  title = {{Approximate Gram-Matrix Interpolation for Wideband Massive MU-MIMO}},
  journal = {IEEE Trans. Vehicular Technology},
  year = {2020},
  doi = {10.1109/tvt.2020.2983025},
}
Figure