Max-Min SINR in Large-Scale Single-Cell MU-MIMO: Asymptotic Analysis and Low Complexity Transceivers

Houssem Sifaou, Abla Kammoun, Luca Sanguinetti, Mérouane Debbah, Mohamed-Slim Alouini

Abstract

This work focuses on the downlink and uplink of large-scale single-cell multiuser multiple-input multiple-output systems in which the base station (BS) endowed with M antennas communicates with K single-antenna user equipments (TIEs). Particularly, we aim at reducing the complexity of the linear precoder and receiver that maximize the minimum signal-to-interference-plus-noise ratio (SINR) subject to a given power constraint. To this end, we consider the asymptotic regime in which M and K grow large with a given ratio. Tools from random matrix theory (RMT) are then used to compute, in closed form, accurate approximations for the parameters of the optimal precoder and receiver, when imperfect channel state information (modeled by the generic Gauss- Markov formulation form) is available at the BS. The asymptotic analysis allows us to derive the asymptotically optimal linear precoder and receiver that are characterized by a lower complexity (due to the dependence on the large scale components of the channel) and, possibly, by a better resilience to imperfect channel state information. However, the implementation of both is still challenging as it requires fast inversions of large matrices in every coherence period. To overcome this issue, we apply the truncated polynomial expansion (TPE) technique to the precoding and receiving vector of each TIE and make use of RMT to determine the optimal weighting coefficients on a per-TIE basis that asymptotically solve the max-min SINR problem. Numerical results are used to validate the asymptotic analysis in the finite system regime and to show that the proposed TPE transceivers efficiently mimic the optimal ones, while requiring much lower computational complexity.

Introduction

In massive MIMO systems, a base station (BS) equipped with a large number of antennas ($M$) serves multiple single-antenna users ($K$). A primary objective for network operators is ensuring fairness among users, typically formulated as the "Max-Min SINR" problem. While optimal linear precoders and receivers exist to solve this, they face significant practical challenges. Specifically, they require computationally expensive matrix inversions and iterative solutions that must be updated within every coherence interval. Furthermore, these designs are often sensitive to imperfect Channel State Information (CSI), which is inevitable in real-world deployments.

Methodology: Bridging Theory and Practice

To address these challenges, this research leverages two key mathematical frameworks:

  1. Random Matrix Theory (RMT): By analyzing the system in the asymptotic regime where $M$ and $K$ grow large, the authors derive "deterministic equivalents." These closed-form expressions depend only on large-scale channel statistics (e.g., path loss), which change much more slowly than instantaneous fast fading. This allows for a significant reduction in the frequency of complex optimization tasks.
  2. User-Specific Truncated Polynomial Expansion (US-TPE): To eliminate the need for full matrix inversions, the authors replace the inverse of the channel covariance matrix with a truncated polynomial expansion. Unlike standard TPE, which uses uniform weights, US-TPE optimizes weights for each user individually. This provides the necessary degrees of freedom to maintain high performance while reducing complexity from $O(K^2 M)$ to $O(KM)$.

The study also incorporates a Gauss-Markov model to account for imperfect CSI, ensuring that the resulting "Asymptotically Optimal" designs are robust to estimation noise.

Performance and Robustness

The proposed methods were evaluated using Monte Carlo simulations with $M=128$ antennas and $K=32$ users. A key finding is the effectiveness of low-order polynomials.

Figure 2: Average per UE rate versus power constraint $P_{max}$ for different precoding schemes. US-TPE with $J=2$ or $J=3$ closely tracks the optimal OLP performance.

As shown in Figure 2, the US-TPE precoder with a truncation order as low as $J=2$ achieves performance nearly identical to the computationally intensive Optimal Linear Precoder (OLP), significantly outperforming simple Maximum Ratio Transmission (MRT). This demonstrates that a very short polynomial is sufficient to capture the benefits of full matrix inversion.

Furthermore, the asymptotic approach provides inherent resilience to CSI errors.

Figure 1: Average per UE rate versus CSI parameter $\eta$. The proposed A-OLP and US-TPE ($J=2$) maintain high rates even as channel estimation quality degrades.

Figure 1 illustrates how the average per-user rate varies with the CSI parameter $\eta$ (where higher $\eta$ indicates poorer CSI). The Asymptotic OLP (A-OLP) and the US-TPE ($J=2$) precoder show remarkable robustness, effectively managing the trade-off between complexity and performance under realistic channel conditions.

Conclusion

This work demonstrates that near-optimal fairness in massive MIMO can be achieved with significantly reduced computational overhead. By moving the heavy lifting of optimization to the large-scale fading timescale and using US-TPE to avoid matrix inversions, the proposed transceivers offer a practical path for high-performance, fair wireless networks. Future work may extend these findings to coordinated multi-cell environments and explore adaptive TPE orders for varying power budgets.

Summary updated

Abstract

This work focuses on the downlink and uplink of large-scale single-cell MU-MIMO systems in which the base station (BS) endowed with $M$ antennas communicates with $K$ single-antenna user equipments (UEs). Particularly, we aim at reducing the complexity of the linear precoder and receiver that maximize the minimum signal-to-interference-plus-noise ratio subject to a given power constraint. To this end, we consider the asymptotic regime in which $M$ and $K$ grow large with a given ratio. Tools from random matrix theory (RMT) are then used to compute, in closed form, accurate approximations for the parameters of the optimal precoder and receiver, when imperfect channel state information (modeled by the generic Gauss-Markov formulation form) is available at the BS. The asymptotic analysis allows us to derive the asymptotically optimal linear precoder and receiver that are characterized by a lower complexity (due to the dependence on the large scale components of the channel) and, possibly, by a better resilience to imperfect channel state information. However, the implementation of both is still challenging as it requires fast inversions of large matrices in every coherence period. To overcome this issue, we apply the truncated polynomial expansion (TPE) technique to the precoding and receiving vector of each UE and make use of RMT to determine the optimal weighting coefficients on a per-UE basis that asymptotically solve the max-min SINR problem. Numerical results are used to validate the asymptotic analysis in the finite system regime and to show that the proposed TPE transceivers efficiently mimic the optimal ones, while requiring much lower computational complexity.
BibTeX
@article{sifaou2016max,
  author = {Houssem Sifaou and Abla Kammoun and Luca Sanguinetti and Mérouane Debbah and Mohamed-Slim Alouini},
  title = {{Max-Min SINR in Large-Scale Single-Cell MU-MIMO: Asymptotic Analysis and Low Complexity Transceivers}},
  journal = {IEEE Transactions on Signal Processing},
  year = {2016},
  doi = {10.1109/tsp.2016.2645518},
}
Figure