[Paper Review] Roy's largest root under rank-one alternatives:The complex valued case and applications

Prathapasinghe Dharmawansa, Boaz Nadler+1 more|arXiv (Cornell University)|Nov 16, 2014

Random Matrices and Applications21 references3 citations

TL;DR

This paper extends Johnstone and Nadler's small noise perturbation approach to derive simple, accurate stochastic approximations for the distribution of Roy's largest root (RLR) in complex-valued Wishart matrices under rank-one alternatives. The key contribution is five new approximations for RLR in single- and double-matrix scenarios, enabling precise analysis of signal detection and MIMO system design at high SNR with finite dimensions.

ABSTRACT

The largest eigenvalue of a Wishart matrix, known as Roy's largest root (RLR), plays an important role in a variety of applications. Most works to date derived approximations to its distribution under various asymptotic regimes, such as degrees of freedom, dimension, or both tending to infinity. However, several applications involve finite and relative small parameters, for which the above approximations may be inaccurate. Recently, via a small noise perturbation approach with fixed dimension and degrees of freedom, Johnstone and Nadler derived simple yet accurate stochastic approximations to the distribution of Roy's largest root in the real valued case, under a rank-one alternative. In this paper, we extend their results to the complex valued case. Furthermore, we analyze the behavior of the leading eigenvector by developing new stochastic approximations. Specifically, we derive simple stochastic approximations to the distribution of the largest eigenvalue under five common complex single-matrix and double-matrix scenarios. We then apply these results to investigate several problems in signal detection and communications. In particular, we analyze the performance of RLR detector in cognitive radio spectrum sensing and constant modulus signal detection in the high signal-to-noise ratio (SNR) regime. Moreover, we address the problem of determining the optimal transmit-receive antenna configuration (here optimality is in the sense of outage minimization) for rank-one multiple-input multiple-output Rician Fading channels at high SNR.

Motivation & Objective

• To develop accurate stochastic approximations for Roy’s largest root (RLR) in complex-valued Wishart matrices under finite-dimensional and finite-degree-of-freedom settings.
• To extend Johnstone and Nadler’s small noise perturbation framework from the real to the complex-valued case.
• To analyze the leading eigenvector’s fluctuation and its overlap with the population eigenvector under rank-one alternatives.
• To apply the new approximations to signal detection problems, including cognitive radio spectrum sensing and constant-modulus signal detection.
• To determine optimal transmit-receive antenna configurations in rank-one Rician fading MIMO channels for outage minimization.

Proposed method

• Derive stochastic approximations for the distribution of the largest eigenvalue (RLR) in five common complex Wishart scenarios using small noise perturbation theory.
• Use matrix decomposition and conditional distribution techniques to decouple the largest eigenvalue from the rest of the spectrum.
• Apply results from random matrix theory and Wishart distribution properties to derive exact and approximate expressions for the RLR distribution under rank-one non-centrality.
• Leverage the Slepian model and conditional independence structures to derive the joint distribution of key statistics like $ S^{11} $ and $ S_{22} $.
• Use the inverse Wishart and conditional independence results to derive the distribution of $ S^{11} $ and its relation to $ S_{22} $, enabling the derivation of moments and asymptotic behavior.
• Apply the derived approximations to compute detection power and outage probability in MIMO systems, using analytical expressions for the RLR distribution under rank-one alternatives.

Experimental results

Research questions

• RQ1 How can the distribution of Roy’s largest root be accurately approximated in complex Wishart matrices when degrees of freedom and dimension are finite and relatively small?
• RQ2 What is the behavior of the leading eigenvector’s overlap with the true population eigenvector under a rank-one alternative in complex Wishart matrices?
• RQ3 How do the new stochastic approximations improve the performance analysis of RLR detectors in cognitive radio spectrum sensing?
• RQ4 What is the optimal transmit-receive antenna configuration in a rank-one Rician fading MIMO channel to minimize outage probability at high SNR?
• RQ5 Can the proposed approximations accurately predict the power of Roy’s largest root test in constant-modulus signal detection?

Key findings

• The paper derives five new stochastic approximations for the distribution of Roy’s largest root in complex Wishart matrices under rank-one alternatives, valid for finite dimensions and degrees of freedom.
• The approximations are shown to be highly accurate even in low-sample regimes, outperforming classical asymptotic Tracy-Widom approximations.
• The leading eigenvector’s overlap with the true signal direction is analytically characterized, enabling precise fluctuation analysis under rank-one perturbations.
• In cognitive radio spectrum sensing, the approximations provide simple, accurate expressions for detection power under known noise covariance.
• For MIMO systems with rank-one Rician fading, the analysis proves analytically that equal numbers of transmit and receive antennas minimize outage probability at high SNR.
• The results confirm and generalize prior simulation-based observations on optimal MIMO configuration, now with analytical justification.