Coherent Multiple-Antenna Block-Fading Channels at Finite Blocklength

Austin Collins, Yury Polyanskiy

Abstract

In modern wireless communications, such as 5G and autonomous systems, data is often transmitted in short bursts with strict latency requirements. Traditional information theory focuses on "Capacity" ($C$), which defines the maximum data rate over infinite time. However, at finite blocklengths ($n$), systems cannot reach this limit. This paper characterizes the fundamental performance limits of coherent Multiple-Input Multiple-Output (MIMO) block-fading channels by deriving the "channel dispersion" ($V$). The results demonstrate that while capacity is symmetric regarding transmit and receive antennas, dispersion is not, and that classic space-time codes like Alamouti’s scheme are optimal for minimizing coding delay.

Introduction

Information theory has long been dominated by the Shannon capacity, which provides the ultimate "speed limit" for data transmission. While this metric is sufficient for long-form data transfer, it fails to capture the realities of low-latency communication. In systems where the blocklength $n$ is small, the achievable rate $R$ is better approximated by the second-order expansion:

$$\text{Rate} \approx C - \sqrt{\frac{V}{n}}Q^{-1}(\epsilon)$$

where $V$ represents the channel dispersion and $\epsilon$ is the error probability. While dispersion is well-understood for simple additive white Gaussian noise (AWGN) channels, calculating it for block-fading MIMO channels—where signal strength fluctuates randomly—is mathematically complex due to the multiplicative nature of the noise. This research provides a precise framework for estimating the coding delay required to approach theoretical capacity in these dynamic environments.

Theoretical Framework and Methodology

The authors employ the Normal Approximation framework to characterize the fundamental limits of the MIMO channel. The primary technical challenge lies in deriving the variance of the information density, which defines the channel dispersion. To achieve this, the study utilizes several advanced mathematical techniques:

  • Information Density Analysis: The information density is decomposed into terms dependent on the channel's eigenvalues and the chosen input distribution.
  • Singular Value Decomposition (SVD): To navigate the complex geometry of MIMO channels, the authors use SVD and properties of isotropic (direction-independent) fading.
  • Hurwitz-Radon Theorem: This deep mathematical result is used to link information-theoretic optimality with the existence of "orthogonal designs" (space-time codes).
  • Hypothesis Testing Bounds: The results are validated using the $\kappa\beta$ achievability bound and a meta-converse, ensuring the approximation remains tight across various scenarios.

Numerical evaluations were conducted using the Spectre project, an open-source software toolbox designed for computing finite-blocklength bounds.

Key Insights: The Asymmetry of Latency

One of the most significant findings of this research is that reciprocity fails for latency. In traditional MIMO theory, a $16 \times 100$ antenna configuration and a $100 \times 16$ configuration share the same capacity. However, this paper proves that their dispersion—and thus their required coding delay—is vastly different. For instance, at 20 dB SNR, a $16 \times 100$ system can reach capacity with a coding delay that is only 60% of what a $100 \times 16$ system requires. This suggests that for latency-sensitive applications, the physical distribution of antennas (transmit vs. receive) is a critical design factor.

Furthermore, the paper provides a new justification for the Alamouti space-time code. Originally designed to simplify demodulation and reduce error rates, the authors demonstrate that for Multiple-Input Single-Output (MISO) channels, these orthogonal designs are actually dispersion-optimal. They represent the most efficient way to maximize data rates at short blocklengths.

Results and Validation

The research culminates in closed-form dispersion formulas (Theorems 2 and 3) for both high-rank and rank-1 MIMO fading channels. Key results include:

  1. Quantifying the Delay Penalty: The authors show that using standard i.i.d. Gaussian codebooks—which are optimal for infinite time—can result in a 40% longer delay to achieve 90% of the channel capacity compared to dispersion-optimal orthogonal designs.
  2. Rank Dependency: The behavior of the channel changes fundamentally based on the rank of the channel matrix. In high-rank scenarios, the choice of input distribution is less critical. In rank-1 scenarios (like MISO), structured inputs are essential to minimize delay.
  3. Model Accuracy: Numerical simulations confirm that the derived Normal Approximation closely tracks computationally expensive achievability bounds, making it a reliable tool for practical system design.

Conclusion and Future Work

While this paper provides a robust framework for understanding finite-blocklength MIMO performance, certain limitations remain. The current model assumes perfect Channel State Information at the Receiver (CSIR). In scenarios with extremely fast fading, the time required to learn the channel may become a significant factor not captured here. Additionally, the "converse" proof requires specific constraints on peak power growth relative to blocklength. Future research may focus on refining the $o(\sqrt{n})$ remainder terms and exploring non-coherent scenarios where the receiver does not have prior knowledge of the channel state.

Summary updated

Abstract

In this paper we consider a channel model that is often used to describe the mobile wireless scenario: multiple-antenna additive white Gaussian noise channels subject to random (fading) gain with full channel state information at the receiver. Dynamics of the fading process are approximated by a piecewise-constant process (frequency non-selective isotropic block fading). This work addresses the finite blocklength fundamental limits of this channel model. Specifically, we give a formula for the channel dispersion -- a quantity governing the delay required to achieve capacity. Multiplicative nature of the fading disturbance leads to a number of interesting technical difficulties that required us to enhance traditional methods for finding channel dispersion. Alas, one difficulty remains: the converse (impossibility) part of our result holds under an extra constraint on the growth of the peak-power with blocklength. Our results demonstrate, for example, that while capacities of $n_t\times n_r$ and $n_r \times n_t$ antenna configurations coincide (under fixed received power), the coding delay can be quite sensitive to this switch. For example, at the received SNR of $20$ dB the $16\times 100$ system achieves capacity with codes of length (delay) which is only $60\%$ of the length required for the $100\times 16$ system. Another interesting implication is that for the MISO channel, the dispersion-optimal coding schemes require employing orthogonal designs such as Alamouti's scheme -- a surprising observation considering the fact that Alamouti's scheme was designed for reducing demodulation errors, not improving coding rate. Finding these dispersion-optimal coding schemes naturally gives a criteria for producing orthogonal design-like inputs in dimensions where orthogonal designs do not exist.
BibTeX
@article{collins2017coherent,
  author = {Austin Collins and Yury Polyanskiy},
  title = {{Coherent Multiple-Antenna Block-Fading Channels at Finite Blocklength}},
  journal = {IEEE Trans. Information Theory},
  year = {2017},
  url = {https://openreview.net/forum?id=2pmY7uajFYC},
}
Figure