Performance of Quantized Massive MIMO With Fronthaul Rate Constraint Over Quasi-Static Channels

Yasaman Ettefagh, Sina Rezaei Aghdam, Giuseppe Durisi, Sven Jacobsson, Mikael Coldrey, Christoph StuderIEEE Access 2023 · DOI · OpenReview

Introduction

Massive Multiple-Input Multiple-Output (MIMO) technology is a cornerstone of modern wireless communication, enabling base stations (BS) to serve multiple users (UEs) simultaneously using large antenna arrays. However, the practical implementation of these systems faces a significant hurdle: the "fronthaul" bottleneck. This occurs when the high-volume data generated by hundreds of antennas at the Remote Radio Head (RRH) must be transmitted to a centralized Baseband Unit (BBU) for processing.

To manage this data load, engineers often employ low-resolution Analog-to-Digital Converters (ADCs). While this reduces the fronthaul rate, it introduces non-linear quantization distortion. This paper provides a rigorous information-theoretic framework to evaluate system performance under these constraints, specifically focusing on quasi-static channels where traditional ergodic models fall short.

System Architecture and the Fronthaul Bottleneck

The study considers a system where multiple UEs communicate with a BS. The BS architecture is split into an RRH and a BBU, connected by a rate-constrained fronthaul link.

System model of a massive MIMO base station with RRH-BBU functional split.

As shown in the system model, the RRH performs radio frequency (RF) processing and quantization via ADCs before sending the data over the fronthaul. The BBU then performs channel estimation and linear combining. The core challenge is balancing the number of antennas ($B$) against the quantization resolution ($Q$) to maximize performance without exceeding the fixed fronthaul capacity.

Methodology: A Three-Pillar Framework

To accurately model performance in quasi-static environments with finite-length packets, the authors developed a framework based on three key mathematical concepts:

  1. Finite-Blocklength Information Theory: Unlike traditional models that assume infinite data streams, this work utilizes the Random-Coding Union (RCUs) bound. This allows for the characterization of error probabilities for short packets, which are common in machine-type communications.
  2. Bussgang’s Theorem: This theorem is used to "linearize" the non-linear distortion inherent in quantization. By treating quantization noise as a statistically equivalent linear term, the system can be analyzed using standard signal-to-noise-and-distortion (SINR) and linear processing techniques.
  3. Mismatch Decoding: Recognizing that perfect channel state information (CSI) is unattainable, the framework incorporates a scaled nearest-neighbor decoder. This accounts for the performance degradation caused by imperfect CSI derived from noisy, quantized pilot signals.

A significant contribution of this work is the derivation of a Generalized Mutual Information (GMI)-based outage bound. This approximation is computationally efficient and remains highly accurate compared to the more complex RCUs bounds.

Key Insights and Performance Trade-offs

The research reveals several critical insights into the design of quantized massive MIMO systems:

  • The Impact of Imperfect CSI: A common pitfall in MIMO research is assuming perfect CSI. Under such assumptions, higher-resolution converters with fewer antennas appear optimal. However, when accounting for the difficulty of estimating channels through low-resolution hardware, the optimal design shifts toward using more antennas with lower resolution (typically 1–4 bits).
  • Uplink-Limited Design: The system's performance is often constrained by the uplink. Consequently, design choices optimized for the uplink—where the BS receives data from UEs—tend to define the overall system efficiency.
  • Pilot Overhead: The framework explicitly models the spectral efficiency loss associated with pilot signals. It highlights a critical balance: spending too much time on channel estimation leaves insufficient resources for data transmission.
  • Optimal Configurations: Under a fixed fronthaul constraint (e.g., 512 bit/s/Hz), the optimal configuration depends on transmit power. At low power, a massive array of 256 antennas with 1-bit resolution is superior. At high power, the system benefits from higher resolution (e.g., 64 antennas with 4-bit converters).

Conclusion

This paper demonstrates that the optimal design of massive MIMO systems under fronthaul constraints is highly sensitive to channel estimation quality and packet length. By providing a reliable GMI-based outage approximation, the authors offer a practical tool for engineers to navigate the complex trade-offs between antenna count and converter resolution. Future work may expand this framework to include non-linear processing techniques and multi-cell interference scenarios.

Summary updated

Abstract

We provide a rigorous framework for characterizing and numerically evaluating the error probability achievable in the uplink and downlink of a fully digital quantized multiuser multiple-input multiple-output (MIMO) system. We assume that the system operates over a quasi-static channel that does not change across the finite-length transmitted codewords, and only imperfect channel state information (CSI) is available at the base station (BS) and at the user equipments. The need for the novel framework developed in this paper stems from the fact that, for the quasi-static scenario, commonly used signal-to-interference-and-distortion-ratio expressions that depend on the variance of the channel estimation error are not relatable to any rigorous information-theoretic achievable-rate bound. We use our framework to investigate how the performance of a fully digital massive MIMO system subject to a fronthaul rate constraint, which imposes a limit on the number of samples per second produced by the analog-to-digital and digital-to-analog converters (ADCs and DACs), depends on the number of BS antennas and on the precision of the ADCs and DACs. In particular, we characterize, for a given fronthaul constraint, the trade-off between the number of antennas and the resolution of the data converters, and discuss how this trade-off is influenced by the accuracy of the available CSI. Our framework captures explicitly the cost, in terms of spectral efficiency, of pilot transmission—an overhead that the outage capacity, the classic asymptotic metric used in this scenario, cannot capture. We present extensive numerical results that validate the accuracy of the proposed framework and allow us to characterize, for a given fronthaul constraint, the optimal number of antennas and the optimal resolution of the converters as a function of the transmitted power and of the available CSI.
BibTeX
@article{ettefagh2023performance,
  author = {Yasaman Ettefagh and Sina Rezaei Aghdam and Giuseppe Durisi and Sven Jacobsson and Mikael Coldrey and Christoph Studer},
  title = {{Performance of Quantized Massive MIMO With Fronthaul Rate Constraint Over Quasi-Static Channels}},
  journal = {IEEE Access},
  year = {2023},
  doi = {10.1109/access.2023.3282369},
}
Figure