PaperlayerView exact revision on arXiv

Source-linked AI summary

A Unified Framework for Characterizing General MIMO Channels

Zeyan Zhuang, Anzheng Tang, Xin Zhang, Dongfang Xu, Shenghui Song

arXiv:2610.08932v1cs.ITmath.IT

TL;DR

Modern MIMO architectures bring complex channel correlations that existing specialized analyses do not generally cover. The paper develops a covariance-based framework with deterministic mutual-information approximations, and reports accurate results for two representative systems with an explicit convergence rate.

  • Problem

    Increasingly diverse MIMO architectures produce complex correlations, while analyses based on specific channel structures do not directly address general MIMO channels.

    Source: §I · p. 1

  • Method

    The paper controls the spectral norm of a generally correlated Gaussian channel and derives deterministic resolvent and EMI approximations through matrix-valued self-consistent equations.

    Source: B. Contributions · p. 3

  • Results

    For downlink distributed and uplink heterogeneous MIMO, the EMI approximations achieve O(N^-1/2) convergence, and simulations validate their accuracy.

    Source: B. Contributions · p. 3

  • Takeaways & Limitations

    The covariance-based framework recovers several structured channel results as special cases and supports analysis of a broad range of MIMO architectures.

    Source: B. Contributions · p. 3, §VI · p. 12

Abstract

from arXiv · show

Modern MIMO systems are evolving toward higher-dimensional and more flexible architectures, such as distributed and holographic MIMO, offering substantial benefits while giving rise to increasingly complex channel correlation structures. This growing architectural diversity makes it difficult to characterize the fundamental limits of different MIMO systems on a case-by-case basis, motivating a unified analytical framework applicable across a broad range of architectures. This paper addresses this research gap by investigating a generally correlated Rayleigh channel whose vectorized form follows a general Gaussian distribution with an arbitrary covariance matrix. For that purpose, we first prove that the spectral norm of the channel matrix is bounded in the asymptotic regime where the numbers of transmit and receive antennas grow proportionally. We then derive a deterministic approximation for the ergodic mutual information, with an explicit convergence rate governed by the structure of the channel correlation. The approximation is characterized by a pair of matrix-valued self-consistent equations, for which we establish the existence and uniqueness of the solution and propose an iterative numerical algorithm. Furthermore, we develop a framework based on positive linear maps to analyze the stability of these equations, which can facilitate the spectral analysis of broader classes of random matrices. The proposed characterization unifies many existing results for structured MIMO channels as special cases while remaining applicable to a broad class of general MIMO architectures that are difficult to evaluate using existing analytical frameworks. To demonstrate its utility, we apply the developed theory to two representative systems, namely downlink distributed MIMO and uplink heterogeneous MIMO. Numerical results confirm the accuracy of the derived deterministic approximations.

I. INTRODUCTION

Modern MIMO architectures produce increasingly complex channel correlations, while many existing analyses rely on specialized channel structures. The paper presents a general covariance-based framework for characterizing such channels.

A. Related Work

  • Emerging architectures increase correlation diversity, while analyses built for separable, variance-profile, or block-structured channels do not directly cover general MIMO channels.The paper models Vec(H) as zero-mean complex Gaussian with covariance C, allowing the discussed correlation structures to be represented through choices of C.Source: §I · p. 1, A. Related Work · p. 2

B. Contributions

  • The general-channel analysis establishes spectral-norm control and deterministic resolvent, linear-spectral-statistic, and mutual-information approximations, recovering several structured models as special cases.The stated special cases include separable, column-independent, mixed-separable, and additive-separable channel models.Source: B. Contributions · p. 3
  • A pair of matrix-valued self-consistent equations characterizes the resolvent approximation; the paper proves existence and uniqueness and uses positive linear maps to control stability.The stability framework is central to deriving the explicit convergence rate, and an iterative algorithm computes the solution.Source: B. Contributions · p. 3
  • For downlink distributed and uplink heterogeneous MIMO, the EMI approximations have O(N^-1/2) convergence, and simulations validate their accuracy.Here N is the number of receive antennas.Source: B. Contributions · p. 3

II. SYSTEM MODEL

  • The system model assumes a generalized Rayleigh channel with arbitrary correlation, represented by a deterministic Hermitian non-negative covariance matrix for Vec(H).The model includes i.i.d., Kronecker, and Weichselberger channels as special cases; the receiver has perfect CSI while the transmitter has statistical CSI.Source: §II · pp. 3–4

III. MAIN RESULTS

The section establishes a spectral-norm bound for generally correlated Gaussian channels, then develops a deterministic resolvent approximation based on matrix-valued self-consistent equations. The approximation supports asymptotic spectral analysis and convergence results.

A. A Bound for the Spectral Norm

  • Under Assumptions 1 and 2, the spectral norm of a generally correlated Gaussian channel is bounded with a constant independent of N and M.The paper also establishes a moment bound for every positive p, with its constant independent of N and M.Source: A. A Bound for the Spectral Norm · p. 4

B. Deterministic Approximation of the Resolvent

  • A pair of matrix-valued self-consistent equations has a unique solution that yields the deterministic approximation of the trace-form resolvent.The solution can be computed with an iterative algorithm; the stated general per-iteration complexity is O(N^4).Source: B. Deterministic Approximation of the… · p. 5
  • The approximation error rate depends on the correlation structure through E_N, which can remain nonvanishing in the worst case but is small for many correlated models.The convergence-rate parameter satisfies 1/N ≲ E_N ≲ 1.Source: B. Deterministic Approximation of the… · p. 6
  • The resolvent approximation gives almost-sure convergence of the empirical spectral distribution to a deterministic limiting spectral distribution.For fixed z outside the nonnegative real axis, the normalized resolvent traces converge almost surely.Source: B. Deterministic Approximation of the… · p. 6
  • When the stated condition holds, no eigenvalues occur outside the limiting spectral support, supporting Gram-matrix invertibility for zero-forcing precoding.The passage applies this no-eigenvalue property to multi-user MIMO communication.Source: B. Deterministic Approximation of the… · p. 6

C. Convergence of the EMI

The section applies the resolvent approximation to ergodic mutual information (EMI) and uses it to motivate a deterministic optimization for transmit covariance. For many structured MIMO channels, the iterative procedure yields an asymptotically capacity-achieving covariance.

C. Convergence of the EMI

  • Theorem 3 provides a deterministic approximation for the ergodic mutual information under the same settings as Theorem 2.The mutual information is treated as a linear spectral statistic of HHH with f(λ) = log(1 + λ/σ^2).Source: C. Convergence of the EMI · p. 6
  • When E_N is small, the EMI maximization problem can be approximated by a deterministic optimization over the transmit covariance.The approximation addresses the computational challenge posed by the expectation in the objective.Source: C. Convergence of the EMI · p. 7
  • For many structured MIMO channels, the iterative procedure yields an asymptotically capacity-achieving transmit covariance.The stated procedure alternates covariance updates with computation of the self-consistent-equation solution.Source: C. Convergence of the EMI · p. 7

D. Recovery of Existing Results for Structured Channel Models

The section shows how the general characterization recovers deterministic approximations for structured channel models. It illustrates this with separable correlation and column-independent correlation.

D. Recovery of Existing Results for Structured Channel Models

  • Expressing a channel's correlation through C lets Theorems 2 and 3 recover its deterministic resolvent and EMI approximations and associated convergence rates.The section presents these models as special cases of the general framework.Source: D. Recovery of Existing Results for… · p. 7

1) Separable Correlated Model:

  • For the separable correlated model, the paper's error bound matches the bound in [13, Theorem 3, Proposition 5].In this setting, C is a Kronecker product of the transmit- and receive-side correlation matrices.Source: 1) Separable Correlated Model · p. 7

2) Column-Independent Correlation Model:

  • For the column-independent model, C is block diagonal, and the order of the error bound coincides with [39, Proposition 3].The model has applications in multi-user MIMO communication systems and multivariate statistics.Source: 2) Column-Independent Correlation Model · p. 7

3) Mixed-Separable Correlation Model:

The mixed-separable and additive-separable models recover O(1/N) convergence under their stated conditions. The applications section uses the framework for distributed MIMO, including an antenna-partition regime in which both a large antenna set and the number of sets can grow.

3) Mixed-Separable Correlation Model:

  • With fixed block counts, proportional block growth, and bounded correlation norms, the mixed-separable model has E_N ≍ 1/N and recovers the zero-mean results in.Its structural dimension is bounded by the product of the block counts.Source: 3) Mixed-Separable Correlation Model · p. 7

4) Additive-Separable Correlation Model:

IV. APPLICATIONS TO SPECIFIC MIMO SYSTEMS

  • Distributed and heterogeneous MIMO applications address emerging antenna deployments and hardware architectures whose complex spatial correlations conventional structured models do not cover.Source: §IV · p. 8

A. Downlink Distributed MIMO System

  • The downlink distributed model allows both the central antenna set and the number of single-antenna sets to grow, capturing highly imbalanced partitions, with its approximation characterized by positive equation solutions.In the presence of spatial correlation, such imbalance typically leads to performance degradation.Source: A. Downlink Distributed MIMO System · pp. 8–9

B. Uplink Heterogeneous MIMO System

The uplink model combines two distributed receive sets with different channel statistics, including a Weichselberger-type channel and a Kronecker channel. Its non-separable correlation structure has a slower convergence rate than the mixed-separable model.

B. Uplink Heterogeneous MIMO System

  • The two receive sets can have different scattering environments and channel statistics; one channel can follow the Weichselberger model while the other follows the Kronecker model.The Weichselberger case arises when the correlation matrices are unitary; the setting can represent a holographic array with mutual coupling alongside a uniform linear array.Source: B. Uplink Heterogeneous MIMO System · p. 9
  • The EMI approximation is characterized by positive solutions to a system of equations, with matrices T1, T2, and eT defined accordingly.Source: B. Uplink Heterogeneous MIMO System · p. 9
  • Non-separable correlations from Sr, St, and Σ yield O(N^-1/2) convergence, slower than the mixed-separable model's O(N^-1) rate.Source: B. Uplink Heterogeneous MIMO System · p. 10

V. NUMERICAL RESULTS

The simulations compare the theoretical spectral and EMI approximations with numerical results, then apply the framework to downlink distributed and uplink heterogeneous MIMO. The conclusion identifies broader uses of the covariance-based framework and future extensions.

V. NUMERICAL RESULTS

  • The numerical results use a spatial-correlation model commonly used for uniform linear arrays, parameterized by antenna spacing, mean angle, and RMS angular spread.The mean angle and RMS angular spread are measured in degrees.Source: §V · p. 10

A. Validation of the ESD Approximation

  • The empirical spectral distribution of HᴴH is close to its limiting spectral distribution for dimensions N = 180 and M = 250.The comparison validates Theorem 2 under the stated column-pair covariance model.Source: A. Validation of the ESD Approximation · p. 10

B. EMI Approximation for Downlink Distributed MIMO

C. EMI Approximation for Uplink Heterogeneous MIMO

  • In the uplink heterogeneous system, the deterministic approximation closely matches Monte Carlo results; allocating more antennas to BS2 and increasing user antennas improve EMI.The system has a total antenna count held fixed for the BS2 allocation comparison, and the simulations average 500 independent trials.Source: C. EMI Approximation for Uplink… · p. 11, Figure 3 · p. 11

VI. CONCLUSION

  • The general Gaussian-channel analysis applies to downlink distributed and uplink heterogeneous MIMO, with numerical results demonstrating approximation accuracy.The conclusion notes that different antenna deployments and propagation environments can be incorporated through covariance matrix C.Source: §VI · pp. 11–12

APPENDIX A MATHEMATICAL TOOLS

The appendix supplies tools for bounding band-matrix commutators and analyzing positive linear maps, then uses Gaussian concentration and a net argument to bound the channel matrix norm.

A. Results on Band Matrices

  • For a band matrix A with bandwidth W, the commutator map B_I(A) has norm at most W times the norm of A.The proof uses a trigonometric-polynomial Bernstein inequality.Source: A. Results on Band Matrices · p. 12

B. Results on Positive Linear Maps

  • A positive linear map with X = T(X) + Y for positive definite X and Y has spectral radius below one, supporting control of the inverse of Id_N − T.The cited lemma gives a bound on powers of T and states that (Id_N − T)⁻¹ is positive.Source: B. Results on Positive Linear Maps · pp. 12–13
  • A positive map whose Choi matrix is Hermitian non-negative admits a representation as a sum of matrix actions V_k X V_kᴴ.This is the representation established by Choi’s theorem in the appendix.Source: B. Results on Positive Linear Maps · p. 13

APPENDIX B PROOF OF THEOREM 1

  • Under proportional dimensional growth and bounded covariance, the Gaussian channel matrix has bounded spectral norm and uniformly bounded moments of every finite order.The proof combines a probabilistic tail bound with an epsilon-net argument.Source: Appendix B · pp. 13–14

APPENDIX C PROOF OF PROPOSITION 1

This appendix develops the matrix-valued Stieltjes-transform properties used to establish the self-consistent equations’ solution. It proves uniqueness and constructs an iterative solution by convergence of successive iterates.

APPENDIX C PROOF OF PROPOSITION 1

  • The class S_n(R+) consists of matrix-valued functions represented through a non-negative matrix-valued measure supported on the non-negative reals.The appendix uses this class to characterize the solution of the equations.Source: Appendix C · pp. 14–15

A. Uniqueness

  • Two solutions of the self-consistent equations in the matrix-valued Stieltjes-transform classes must coincide.The proof first establishes equality on a domain in the upper half-plane, then uses holomorphicity to extend uniqueness.Source: A. Uniqueness · p. 15

B. Existence

  • Starting from −1/z identity matrices, the iterative scheme produces Cauchy sequences whose limits satisfy the self-consistent equations.The iterates remain in the Stieltjes-transform classes, and their limits satisfy the required norm bounds.Source: B. Existence · p. 16

APPENDIX D PROOF OF THEOREM 2

  • The proof of the resolvent approximation has a stochastic step using Gaussian tools and a deterministic step showing the expected-resolvent intermediate converges to the fixed-point solution.The intermediate object is more analytically tractable than the resolvent itself.Source: Appendix D · p. 16

A. Stochastic Step

The stochastic-step analysis relates the expected resolvent to an intermediate approximation, then bounds error terms using resolvent moments and covariance structure.

A. Stochastic Step

  • The stochastic-step derivation expresses the approximation error through covariance terms and uses trace-form resolvent moment bounds to control them.The appendix introduces a central-moment control lemma under Assumptions 1 and 2.Source: A. Stochastic Step · p. 17
  • The error in the initial bound becomes negligible when the span of the correlation matrices is low-rank, but can remain O(1) when its dimension is about N^2.The proof therefore develops a refined bound for the latter regime.Source: A. Stochastic Step · p. 18
  • The proof combines bounds for three error terms to obtain a bound for the resolvent approximation when the test matrix has uniformly bounded spectral norm.The result is stated for z outside the non-negative real axis.Source: A. Stochastic Step · p. 20

B. Deterministic Step

The deterministic step controls the difference between Tα and T through linear-map bounds and resolvent identities. These estimates establish invertibility of the auxiliary map and complete the proof of Theorem 2.

B. Deterministic Step

  • The proof controls the linear functional associated with Tα−T using propositions that bound the map and its error term.The section introduces four propositions to establish this connection.Source: B. Deterministic Step · pp. 20–21
  • The linear maps L and eL admit representations as sums indexed by matrices Vk, with the number of terms bounded by NM.The representation is established using a Hermitian non-negative matrix and Lemma 4.Source: B. Deterministic Step · p. 20
  • The proof relates Tα−T to the error term through the resolvent identity, then expresses L(∆) using the inverse of IdM−FI,α.The resulting bound is used to complete the proof of Theorem 2.Source: B. Deterministic Step · pp. 21–22
  • Invertibility of IdM−FI,α is established using a Neumann-series argument after bounding the relevant spectral radius.The analysis introduces an auxiliary map F involving only T and its adjoint to handle the coupled dependence on T and Tα.Source: B. Deterministic Step · pp. 21–22

APPENDIX E PROOF OF LEMMA 2

The appendices bound resolvent entries by perturbing the linear map and use Poincaré–Nash inequalities to establish variance control. The variance argument is part of a broader proof that extends the bound to higher central moments.

APPENDIX E PROOF OF LEMMA 2

  • Finite bandwidth of L limits the effect of a small perturbation, allowing G to be constrained through its perturbed version.The proof uses a diagonal weight matrix Dα and bounds the perturbation for z outside the non-positive real axis.Source: Appendix E · pp. 22–23

APPENDIX F PROOF OF LEMMA 6

  • The proof of Lemma 6 first bounds variances of resolvent linear functionals, then extends the bound to arbitrary integer-order central moments by induction.The section focuses on proving (80), omitting the analogous proof of (81).Source: Appendix F · p. 23

A. Variance Case

  • By the Poincar´e-Nash inequality, the right-hand side of (135) is written using two vectors, x = Vec(QHAHQHH) and x = Vec(QAQH).Source: A. Variance Case · p. 23

Tr HHQAQQHAHQHH + Tr HHQHAHQHQAQH

The proof develops variance and higher-moment bounds for resolvent functionals, establishes a lower bound used in Proposition 6, and connects the resolvent to mutual information. It then shows that the deterministic approximation is close to the expected mutual information.

Tr HHQAQQHAHQHH + Tr HHQHAHQHQAQH

B. General Central Moment Case

  • The proof extends the central-moment bound to all integer orders by induction, obtaining E|a Ẋ|2k = O_k(1).It treats the variance and squared-mean terms separately and uses the induction hypothesis, a tail bound, and Hölder’s inequality.Source: B. General Central Moment Case · p. 24

APPENDIX G PROOF OF PROPOSITION 6

  • The proof of Proposition 6 uses a matrix-valued measure identity and tightness to obtain a uniform lower bound on its mass over [0, η].It then bounds |xᴴT(z)x| by considering the cases ℜz≥0 and ℜz<0.Source: Appendix G · pp. 24–25

APPENDIX H PROOF OF THEOREM 3

  • The resolvent is connected to mutual information, and the proof shows that I is a good approximation of E[I].The argument derives the connection using a log-determinant derivative and then differentiates the deterministic approximation.Source: Appendix H · p. 25
Loading 2610.08932v1…