The Second-Order Coding Rate of the MIMO Quasi-Static Rayleigh Fading Channel

Jakob Hoydis, Romain Couillet, Pablo PiantanidaIEEE Trans. Information Theory 2015 · OpenReview

TL;DR

This paper provides a theoretical framework for understanding the error probability in MIMO wireless communication systems operating with finite block lengths. By leveraging advanced random matrix theory, the authors derive tight upper and lower bounds on the optimal error probability for rates close to the channel's capacity, offering crucial insights for designing more reliable wireless systems.

What Problem Does This Paper Address?

In real-world wireless communication, transmissions are limited by finite block lengths due to practical constraints like delay and complexity. This poses a challenge for determining the fundamental performance limits, especially in complex scenarios like MIMO (Multiple-Input Multiple-Output) systems with quasi-static fading channels. Unlike simpler scenarios, these channels are non-ergodic, meaning their statistical properties don't hold over long periods, making it difficult to achieve rates arbitrarily close to the theoretical capacity. Existing methods for analyzing error probability in these finite-block-length, non-ergodic channels are often intractable. This paper aims to bridge this gap by providing a rigorous analysis of the second-order coding rate, which describes performance when the coding rate is a small perturbation of the asymptotic capacity, specifically within $O(1/\sqrt{nK})$ of it.

Methods

The core of this paper lies in its application of information-spectrum theory and random matrix theory (RMT). The authors tackle the problem by analyzing the "information density," a key quantity in information theory, in a regime where the number of antennas ($N$, $K$) and the block length ($n$) all grow large simultaneously.

Key techniques include:

  • Central Limit Theorem (CLT) for Information Density: The authors derive a novel CLT for the information density in the MIMO quasi-static fading channel. This is a significant contribution as it allows for the statistical analysis of the information density in this complex regime.
  • Statistical Bounds: They utilize recent RMT techniques to derive statistical bounds on the information density. This is crucial for handling the high-dimensional random matrices involved in MIMO channels.
  • Feinstein's and Verdú-Han's Lemmas: These foundational lemmas from information theory are adapted and used to establish lower and upper bounds on the optimal error probability.
  • Gaussian Approximations: The analysis relies heavily on Gaussian approximations for the distributions of random matrices, a common and powerful tool in RMT.

The novelty lies in the specific application of these RMT tools to the MIMO quasi-static fading channel at finite block lengths and the derivation of a CLT for the information density in this setting.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and bounds, not empirical evaluation on specific datasets.

Key Insights and Ideas

  • Bridging the Gap Between Theory and Practice: The paper provides a theoretical foundation for understanding error performance in practical, finite-block-length communication systems, which is crucial for designing efficient wireless technologies.
  • The Power of Random Matrix Theory: The work highlights the indispensable role of RMT in analyzing the complex statistical behavior of MIMO channels, especially when dealing with large numbers of antennas and long block lengths.
  • Second-Order Coding Rate is Key: By focusing on the second-order coding rate, the authors offer a more refined understanding of performance beyond the asymptotic capacity, revealing how error probability behaves when operating close to the theoretical limit.
  • Tight Bounds for Practical Design: The derived closed-form upper and lower bounds on error probability are not just theoretical curiosities; they offer concrete performance envelopes that engineers can use for system design and optimization.

Key Results

The paper's main contribution is the derivation of closed-form upper and lower bounds on the optimal average error probability for the MIMO quasi-static Rayleigh fading channel. These bounds are characterized for the second-order coding rate, meaning rates within $O(1/\sqrt{nK})$ of the asymptotic capacity.

  • The bounds are expressed in terms of the channel parameters: the ratio of antennas ($c=N/K$), the block length to antenna ratio ($\beta=n/K$), and the signal-to-noise ratio (SNR).
  • While the bounds are not always equal (unlike in some simpler AWGN scenarios), the gap is shown to be tight, especially for practical SNR values and when the coding rate is below capacity ($r<0$).
  • The paper also provides insights into the high-SNR and low-SNR regimes, showing how the bounds behave under different signal conditions.
  • Numerical comparisons with LDPC codes suggest good agreement between the theoretical bounds and practical performance.

Limitations and Future Work

  • Gaussianity Assumption: The analysis relies on the Gaussianity of the channel matrix entries. The paper notes that deviations from this assumption (e.g., non-zero fourth-order cumulants) can lead to a "bias term" that modifies the relationship between the bounds and the achievable error probability.
  • Tightness of Bounds for $r>0$: For coding rates above capacity ($r>0$), the lower bound is noted to be loose, suggesting that further work might be needed to tighten it.
  • Future Work: The authors suggest that their approach can be extended to other scenarios, such as block-fading regimes with multiple coherence blocks or practical scenarios involving linear receive filters. The analysis of the asymptotic statistics of the mutual information density is highlighted as a promising avenue for future research.

Summary updated

Abstract

The second-order coding rate of the multiple-input multiple-output (MIMO) quasi-static Rayleigh fading channel is studied. We tackle this problem via an information-spectrum approach and statistical bounds based on recent random matrix theory techniques. We derive a central limit theorem (CLT) to analyze the information density in the regime where the block length n and the number of transmit and receive antennas K and N, respectively, grow simultaneously large. This result leads to the characterization of closed-form upper and lower bounds on the optimal average error probability when the coding rate is within O(1/√(nK)) of the asymptotic capacity.
BibTeX
@article{hoydis2015second,
  author = {Jakob Hoydis and Romain Couillet and Pablo Piantanida},
  title = {{The Second-Order Coding Rate of the MIMO Quasi-Static Rayleigh Fading Channel}},
  journal = {IEEE Trans. Information Theory},
  year = {2015},
  url = {https://openreview.net/forum?id=9DEnLZlvvdn},
}
Figure