Source Enumeration via RMT Estimator Based on Linear Shrinkage Estimation of Noise Eigenvalues Using Relatively Few Samples

TL;DR

This paper introduces a novel RMT estimator that uses linear shrinkage to more accurately estimate noise eigenvalues, effectively addressing bias and noise variance estimation issues in signal enumeration.

Contribution

It proposes the LS-RMT estimator that incorporates bias correction and noise variance estimation, improving signal detection accuracy over traditional RMT methods.

Findings

LS-RMT estimator reduces under-estimation of signals.

It outperforms existing estimators in simulations.

Addresses bias and noise variance estimation issues.

Abstract

Estimating the number of signals embedded in noise is a fundamental problem in array signal processing. The classic RMT estimator based on random matrix theory (RMT) tends to under-estimate the number of signals as it does not consider the non-negligible bias term among eigenvalues for finite sample size. Moreover, the RMT estimator suffers from uncertainty in noise variance estimation problem. In order to overcome these problems, we firstly derive a more accurate expression for the distribution of the sample eigenvalues and the bias term among eigenvalues by utilizing the linear shrinkage (LS) estimate of noise sample eigenvalues. Then, we analyze the effect of the bias term among eigenvalues on the estimation performance of the RMT estimator, and derive the increased under-estimation probability of the RMT estimator incurred by this bias term. Based on these results, we propose a novel RMT estimator based on LS estimate of noise eigenvalues (termed as LS-RMT estimator) by incorporating the bias term into the decision criterion of the RMT estimator. As the LS-RMT estimator incorporates this bias term among eigenvalues into the decision criterion of the RMT estimator, it can detect signal eigenvalues immersed in this bias term. Therefore, the LS-RMT estimator can overcome the higher under-estimation probability of the RMT estimator incurred by the bias term among eigenvalues, and also avoids the uncertainty in the noise variance estimation suffered by the RMT estimator as the noise variance is estimated under the assumption that the eigenvalue being tested is arising from noise. Finally, extensive simulation results are presented to show that the proposed LS-RMT estimator outperforms the existing estimators.