# Optimization of MIMO Systems Capacity Using Large Random Matrix Methods

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. General Principles

#### 2.1. Optimization of the Average Mutual Information

#### 2.2. Large System Approximation of $I(\mathbf{Q})$

#### 2.2.1. The Case $\mathbf{Q}=\mathbf{I}$

**Background on the behaviour of the eigenvalue distribution of**$\mathbf{H}{\mathbf{H}}^{H}$ Let ${\mathbf{H}}_{r,t}$ be a $r\times t$ complex Gaussian random matrix representing the channel matrix of a MIMO system (we indicate that $\mathbf{H}$ is a $r\times t$ matrix by denoting $\mathbf{H}$ by ${\mathbf{H}}_{r,t}$), and consider its associated Gram matrix defined by ${\mathbf{H}}_{r,t}{\mathbf{H}}_{r,t}^{H}$. Under certain assumptions on the probability distribution of ${\mathbf{H}}_{r,t}$ (see the examples below), it appears that the empirical eigenvalue distribution ${\widehat{\mu}}_{r,t}$ of ${\mathbf{H}}_{r,t}{\mathbf{H}}_{r,t}^{H}$ defined as $\frac{1}{r}{\sum}_{i=1}^{r}\delta \left(\lambda -{\lambda}_{i}({\mathbf{H}}_{r,t}{\mathbf{H}}_{r,t}^{H})\right)$ tends to have a deterministic behaviour when the dimensions r and t converge to $+\infty $ in such a way that $\frac{t}{r}\to c$, where $0<c<+\infty $. In order to shorten the notations, $t\to +\infty $ stands in the following for r and t converge to $+\infty $ in such a way that $\frac{t}{r}\to c$. This is equivalent to saying that it exists deterministic probability measures ${\mu}_{r,t}$ (which depend on the dimensions r and t) for which

**Large system approximation of**I We finally consider the problem of approximating $I=\mathbb{E}\left(log\mathrm{det}({\mathbf{I}}_{r}+\frac{1}{{\sigma}^{2}}\mathbf{H}{\mathbf{H}}^{H})\right)$. We first note that

**Examples**

#### 2.2.2. The General Case $\mathbf{Q}\ne \mathbf{I}$

- for model (15), matrix $\tilde{\mathbf{C}}$ by ${\mathbf{Q}}^{1/2}\tilde{\mathbf{C}}{\mathbf{Q}}^{1/2}$,
- for model (27), matrices ${({\tilde{\mathbf{C}}}_{l})}_{l=1,\dots ,L}$ by ${({\mathbf{Q}}^{1/2}{\tilde{\mathbf{C}}}_{l}{\mathbf{Q}}^{1/2})}_{l=1,\dots ,L}$,
- for model (36), matrices $\mathbf{A}$ and $\tilde{\mathbf{C}}$ by $\mathbf{A}{\mathbf{Q}}^{1/2}$ and ${\mathbf{Q}}^{1/2}\tilde{\mathbf{C}}{\mathbf{Q}}^{1/2}$ respectively,
- for model (51), matrices ${({\tilde{\mathbf{C}}}_{l})}_{l=1,\dots ,L}$ by ${({\mathbf{Q}}_{l}^{1/2}{\tilde{\mathbf{C}}}_{l}{\mathbf{Q}}_{l}^{1/2})}_{l=1,\dots ,L}$ (where $\mathbf{Q}=\mathrm{Diag}({\mathbf{Q}}_{1},\dots ,{\mathbf{Q}}_{L})$)

## 3. Concavity of the Large System Approximation

## 4. Properties of $\overline{I}({\overline{\mathbf{Q}}}_{*})$

#### 4.1. Comparison between $I({\mathbf{Q}}_{*})$ and $I({\overline{\mathbf{Q}}}_{*})$

**Theorem**

**1**

- (1)
- ${sup}_{t}\parallel {\mathbf{Q}}_{*}\parallel <+\infty $
- (2)
- ${sup}_{t}\parallel {\overline{\mathbf{Q}}}_{*}\parallel <+\infty $
- (3)
- $I({\mathbf{Q}}_{*})-I({\overline{\mathbf{Q}}}_{*})=\mathcal{O}\left(\frac{1}{t}\right)$

#### 4.2. Structure of $I({\overline{\mathbf{Q}}}_{*})$

**Theorem**

**2**

**Proof**

**of**

**Theorem 2.**

**Definition**

**1**

**Proposition**

**1**

- φ is Gâteaux differentiable at $\mathbf{Q}$ in the direction $\mathbf{P}-\mathbf{Q}$ for each $\mathbf{P},\mathbf{Q}\in {\mathcal{C}}_{1}$,
- ${\mathbf{Q}}_{opt}$ is the unique maximizer of φ on ${\mathcal{C}}_{1}$ if and only if it verifies:$$\forall \mathbf{Q}\in {\mathcal{C}}_{1},\phantom{\rule{0.166667em}{0ex}}\langle {\varphi}^{\prime}({\mathbf{Q}}_{opt}),\mathbf{Q}-{\mathbf{Q}}_{opt}\rangle \le 0$$

## 5. Optimization Algorithms of $\overline{I}(\mathbf{Q})$

- Initialization: choose ${\delta}^{(0)}$ and ${\tilde{\delta}}^{(0)}$ as vectors with strictly positive components
- Evaluation of $({\delta}^{(k+1)},{\tilde{\delta}}^{(k+1)})$ from $({\delta}^{(k)},{\tilde{\delta}}^{(k)})$:$$\begin{array}{c}\hfill {\delta}^{(k+1)}=f(\mathbf{Q},{\delta}^{(k)},{\tilde{\delta}}^{(k)})\end{array}$$$$\begin{array}{c}\hfill {\tilde{\delta}}^{(k+1)}=\tilde{f}(\mathbf{Q},{\delta}^{(k)},{\tilde{\delta}}^{(k)})\end{array}$$

- Initialization: ${\mathbf{Q}}^{(0)}=\mathbf{I}$
- Evaluation of ${\mathbf{Q}}^{(m+1)}$ from ${\mathbf{Q}}^{(m)}$: ${\delta}^{(m+1)}$ and ${\tilde{\delta}}^{(m+1)}$ are defined as the solutions of the canonical Equation (73, 74) when $\mathbf{Q}={\mathbf{Q}}^{(m)}$, and ${\mathbf{Q}}^{(m+1)}$ coincides with the argument of the maximum w.r.t. $\mathbf{Q}$ of $log\mathrm{det}\left(\mathbf{I}+\mathbf{Q}\mathbf{G}({\delta}^{(m+1)},{\tilde{\delta}}^{(m+1)})\right)$.

- Initialization: choose ${\delta}^{(0)}$ and ${\tilde{\delta}}^{(0)}$ as vectors with strictly positive components
- Evaluation of ${\delta}^{(n+1)}$ and ${\tilde{\delta}}^{(n+1)}$ from ${\delta}^{(n)}$ and ${\tilde{\delta}}^{(n)}$: ${\mathbf{Q}}^{(n)}$ coincides with the argument of the maximum w.r.t. $\mathbf{Q}$ of $log\mathrm{det}\left(\mathbf{I}+\mathbf{Q}\mathbf{G}({\delta}^{(n)},{\tilde{\delta}}^{(n)})\right)$ and ${\delta}^{(n+1)}$ and ${\tilde{\delta}}^{(n+1)}$ are defined by:$$\begin{array}{c}\hfill {\delta}^{(n+1)}=f({\mathbf{Q}}^{(n)},{\delta}^{(n)},{\tilde{\delta}}^{(n)})\end{array}$$$$\begin{array}{c}\hfill {\tilde{\delta}}^{(n+1)}=\tilde{f}({\mathbf{Q}}^{(n)},{\delta}^{(n)},{\tilde{\delta}}^{(n)})\end{array}$$

$l=1$ | $l=2$ | $l=3$ | $l=4$ | $l=5$ | |
---|---|---|---|---|---|

mean departure angle | 4.44 | 0.20 | 1.74 | 0.29 | 0.61 |

departure angle spread | 0.09 | 0.08 | 0.04 | 0.11 | 0.01 |

mean arrival angle | 4.12 | 0.22 | 5.34 | 5.87 | 4.26 |

arrival angle spread | 0.09 | 0.08 | 0.05 | 0.08 | 0.03 |

$r=t=4$ | $r=t=8$ | |
---|---|---|

Vu–Paulraj | $503s$ | $1877s$ |

Frozen Water-filling | $17.3ms$ | $20.1ms$ |

Water-filling of [8] | $5.10ms$ | $7.13ms$ |

## 6. Concluding Remarks

## References

- Telatar, E. Capacity of multi-antenna gaussian channels. Eur. Trans. Telecommun.
**1999**, 10, 585–595. [Google Scholar] [CrossRef] - Vu, M.; Paulraj, A. Capacity Optimization for Rician Correlated MIMO Wireless Channels. In Proceedings of the Thirty-Ninth Asilomar Conference on the Signals, Systems and Computers, Pacific Grove, CA, USA, November 2005; pp. 133–138.
- Goldsmith, A.; Jafar, S.; Jindal, N.; Vishwanath, S. Capacity limits of MIMO channels. IEEE J. Select. Areas Commun.
**2003**, 21, 684–702. [Google Scholar] [CrossRef] - Jafar, S.; Goldsmith, A. Transmitter optimization and optimality of beamforming for multiple antenna systems. IEEE Trans. Wirel. Commun.
**2004**, 3, 1165–1175. [Google Scholar] [CrossRef] - Hosli, D.; Kim, Y.; Lapidoth, A. Monotonicity results for coherent MIMO Rician channels. IEEE Trans. Inform. Theory
**2005**, 51, 4334–4339. [Google Scholar] [CrossRef] - Moustakas, A.; Simon, S.; Sengupta, A. MIMO capacity through correlated channels in the presence of correlated interferers and noise: A (not so) large N analysis. IEEE Trans. Inf. Theory
**2003**, 49, 2545–2561. [Google Scholar] [CrossRef] - Moustakas, A.; Simon, S. On the outage capacity of correlated multiple-path MIMO channels. IEEE Trans. Inf. Theory
**2007**, 53, 3887. [Google Scholar] [CrossRef] - Taricco, G. Asymptotic mutual information statistics of separately correlated Rician fading MIMO channels. IEEE Trans. Inf. Theory
**2008**, 54, 3490–3504. [Google Scholar] [CrossRef] - Wen, C.; Jin, S.; Wong, K.; Chen, J.; Ting, P. On the Ergodic Capacity of Jointly-Correlated Rician Fading MIMO Channels. In Proceedings of IEEE International Conference on Acoustics, Speech and Signal, Prague, Czech Republic, May 2011; pp. 3224–3227.
- Tulino, A.; Verdú, S. Random Matrix Theory and Wireless Communications; Now Publishers Inc: Hanover, MA, USA, 2004. [Google Scholar]
- Dumont, J.; Hachem, W.; Lasaulce, S.; Loubaton, P.; Najim, J. On the capacity achieving covariance matrix for Rician MIMO channels: An asymptotic approach. IEEE Trans. Inf. Theory
**2010**, 56, 1048–1069. [Google Scholar] [CrossRef] - Dupuy, F.; Loubaton, P. On the capacity achieving covariance matrix for frequency selective MIMO channels using the asymptotic approach. Inf. Theory IEEE Trans.
**2011**, 57, 5737–5753. [Google Scholar] [CrossRef] [Green Version] - Couillet, R.; Debbah, M.; Silverstein, J. A deterministic equivalent for the analysis of correlated MIMO multiple access channels. IEEE Trans. Inf. Theory
**2011**, 57, 3493–3514. [Google Scholar] [CrossRef] - Wen, C.; Ting, P.; Chen, J. Asymptotic analysis of MIMO wireless systems with spatial correlation at the receiver. IEEE Trans. Commun.
**2006**, 54, 349–363. [Google Scholar] [CrossRef] - Taricco, G.; Riegler, E. On the ergodic capacity of correlated Rician fading MIMO channels with interference. IEEE Trans. Inf. Theory
**2011**, 57, 4123–4137. [Google Scholar] [CrossRef] - Wen, C.; Jin, S.; Wong, K. On the sum-rate of multiuser MIMO uplink channels with jointly-correlated Rician fading. IEEE Trans. Commun.
**2011**, 59, 2883–2895. [Google Scholar] [CrossRef] - Luenberger, D. Optimization by Vector Space Methods; John Wiley and Sons: New York, NY, USA, 1969. [Google Scholar]
- Marčenko, V.; Pastur, L. Distribution of eigenvalues for some sets of random matrices. Math. USSR Sb.
**1967**, 1, 457. [Google Scholar] [CrossRef] - Pastur, L.; Shcherbina, M. Eigenvalue Distribution of Large Random Matrices. In Mathematical Surveys and Monographs; American Mathematical Society: Providence, RI, USA, 2011. [Google Scholar]
- Bai, Z.; Silverstein, J. Spectral Analysis of Large Dimensional Random Matrices (Springer Series in Statistics), 2nd ed.; Springer: Heidelberg, Germnay, 2009. [Google Scholar]
- Girko, V. Theory of Stochastic Canonical Equations, Volume I; Kluwer Academic Publishers: Dordrecht, The Netherlands, 2001; Chapter 7. [Google Scholar]
- Taricco, G. On the Capacity of Separately-Correlated MIMO Rician Channels. In Proceedings of Global Telecommunications Conference, San Francisco, CA, USA, December 2006.
- Sayeed, A. Deconstructing multiantenna fading channels. IEEE Trans. Signal Process.
**2002**, 50, 2563–2579. [Google Scholar] [CrossRef] - Hachem, W.; Loubaton, P.; Najim, J. Deterministic equivalents for certain functionals of large random matrices. Ann. Appl. Probab.
**2007**, 17, 875–930. [Google Scholar] [CrossRef] - Borwein, J.; Lewis, A. Convex Analysis and Nonlinear Optimization : Theory and Examples; Springer: Verlag, Germany, 2000. [Google Scholar]
- Boyd, S.; Vandenberghe, L. Convex Optimization; Cambridge University Press: Cambridge, UK, 2004. [Google Scholar]
- Bolcskei, H.; Gesbert, D.; Paulraj, A. On the capacity of OFDM-based spatial multiplexing systems. IEEE Trans. Commun.
**2002**, 50, 225–234. [Google Scholar] [CrossRef] - Kumar, K.; Caire, G.; Moustakas, A. Asymptotic performance of linear receivers in MIMO fading channels. IEEE Trans. Inf. Theory
**2009**, 55, 4398–4418. [Google Scholar] [CrossRef] - Moustakas, A.; Kumar, K.; Caire, G. Performance of MMSE MIMO Receivers: A Large N analysis for Correlated Channels. In Proceedings of IEEE Vehicular Technology Conference, Barcelona, Spain, 2009.
- Artigue, C.; Loubaton, P. On the precoder design of flat fading MIMO systems equipped with MMSE receivers: A large system approach. IEEE Trans. Inf. Theory
**2011**, 57, 4138–4155. [Google Scholar] [CrossRef]

© 2012 by the authors licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license ( http://creativecommons.org/licenses/by/3.0/).

## Share and Cite

**MDPI and ACS Style**

Dupuy, F.; Loubaton, P.
Optimization of MIMO Systems Capacity Using Large Random Matrix Methods. *Entropy* **2012**, *14*, 2122-2142.
https://doi.org/10.3390/e14112122

**AMA Style**

Dupuy F, Loubaton P.
Optimization of MIMO Systems Capacity Using Large Random Matrix Methods. *Entropy*. 2012; 14(11):2122-2142.
https://doi.org/10.3390/e14112122

**Chicago/Turabian Style**

Dupuy, Florian, and Philippe Loubaton.
2012. "Optimization of MIMO Systems Capacity Using Large Random Matrix Methods" *Entropy* 14, no. 11: 2122-2142.
https://doi.org/10.3390/e14112122