On MMSE Properties and I-MMSE Implications in Parallel MIMO Gaussian Channels

Ronit Bustin, Miquel Payaró, Daniel P. Palomar, Shlomo Shamai

Abstract

This paper extends the "single crossing point" property of the scalar MMSE function, derived by Guo, Shamai and Verdú, to the parallel degraded MIMO scenario. It is shown that the matrix $Q(t)$, which is the difference between the MMSE assuming a Gaussian input and the MMSE assuming an arbitrary input, has, at most, a single crossing point for each of its eigenvalues. Together with the I-MMSE relationship, a fundamental connection between Information Theory and Estimation Theory, this new property is employed to derive results in Information Theory. As a simple application of this property, we provide an alternative converse proof for the broadcast channel (BC) capacity region under covariance constraint in this specific setting.

Introduction

The intersection of information theory and estimation theory has yielded some of the most profound insights in modern communication. Central to this intersection is the I-MMSE relationship, which establishes that the derivative of mutual information with respect to the signal-to-noise ratio (SNR) is equal to the Minimum Mean-Square Error (MMSE).

In scalar Gaussian channels, researchers previously identified a "single crossing point" property: as the channel improves, the MMSE of a Gaussian input and the MMSE of any other input distribution will intersect at most once. This property has served as a "silver bullet" for proving the capacity of various multi-user channels. However, extending this to Multiple-Input Multiple-Output (MIMO) systems is mathematically complex because the MMSE becomes a matrix rather than a scalar. This paper addresses this challenge for parallel MIMO Gaussian channels.

The Challenge of MIMO Systems

In moving from scalar to MIMO systems, two primary complications arise:

  1. Matrix-Valued Error: The MMSE is no longer a single number; it is a covariance matrix.
  2. Multi-dimensional SNR: In MIMO, "improving the channel" is not a simple scalar increment. Channel gains can change independently across different antennas, requiring a more sophisticated way to define the evolution of the system.

The authors focus on parallel MIMO channels, where the channel matrix is diagonal. While this simplifies the interaction between antennas, the inputs remain coupled through the joint distribution of the transmitted vector $X$, providing a tractable middle ground between scalar channels and general MIMO.

Methodology

To bridge the gap between scalar and matrix-valued MMSE, the authors employ several advanced mathematical techniques:

  • Path Parameterization: To handle the multi-dimensional nature of MIMO, they define a monotonic "path" between two channel states. By parameterizing the channel matrix as a function of a single scalar variable $t$, they can analyze how the system evolves as the channel improves along that path.
  • Matrix Differential Calculus: They utilize the Jacobian operator and chain rules for matrices to find the derivative of the MMSE matrix along the chosen path.
  • Eigenvalue Analysis: The core of the analysis focuses on the matrix $Q(t)$, defined as the difference between the MMSE of a Gaussian input and an arbitrary input. By applying Sylvester’s Law of Inertia, they show that the number of negative eigenvalues of this matrix cannot increase as the channel improves.

Key Insights: The "No-Return" Property

The "aha moment" of the research is the discovery that the single crossing property applies to each eigenvalue of the matrix $Q(t)$. The authors demonstrate a "no-return" property: once an eigenvalue of the difference matrix becomes non-negative, it stays non-negative as the channel quality improves.

This implies that for each eigenvalue, there is at most one point where it crosses the horizontal axis. This mathematical behavior ensures that the Gaussian input maintains a specific relationship with any other input distribution throughout the evolution of the channel, reinforcing the idea that Gaussian inputs are often the "worst-case" for estimation but the "best-case" for information transmission.

Information Theoretic Implications

The primary application of this new property is providing an alternative "converse" proof (a proof that a certain rate is impossible to exceed) for the Parallel Degraded MIMO Gaussian Broadcast Channel.

By integrating the eigenvalue properties of $Q(t)$ back into the domain of mutual information using the I-MMSE relationship, the authors provide a streamlined and elegant proof for the capacity region of these channels under covariance constraints. This provides a rigorous mathematical framework for intuitions that were previously difficult to prove in multi-dimensional settings.

Limitations and Future Work

While this work successfully extends the single crossing point property, it is currently limited to:

  • Parallel Channels: The proof relies on the channel matrices being diagonal.
  • Degraded Scenarios: The application to broadcast channels is limited to the case where one user's channel is strictly better than the other's.

The authors suggest that the next major hurdle is extending these results to general, non-diagonal MIMO matrices. This would require handling the non-commutative nature of general channel matrices, likely requiring even more advanced tools from matrix analysis.

Summary updated

Abstract

This paper extends the "single crossing point" property of the scalar MMSE function, derived by Guo, Shamai and Verdú (first presented in ISIT 2008), to the parallel degraded MIMO scenario. It is shown that the matrix Q(t), which is the difference between the MMSE assuming a Gaussian input and the MMSE assuming an arbitrary input, has, at most, a single crossing point for each of its eigenvalues. Together with the I-MMSE relationship, a fundamental connection between Information Theory and Estimation Theory, this new property is employed to derive results in Information Theory. As a simple application of this property we provide an alternative converse proof for the broadcast channel (BC) capacity region under covariance constraint in this specific setting.
BibTeX
@misc{bustin2010mmse,
  author = {Ronit Bustin and Miquel Payaró and Daniel P. Palomar and Shlomo Shamai},
  title = {{On MMSE Properties and I-MMSE Implications in Parallel MIMO Gaussian Channels}},
  year = {2010},
  eprint = {1004.4490},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1004.4490},
}
Figure