1: \begin{abstract}

2: We address covariance estimation in the sense of minimum

3: mean-squared error (MMSE) for Gaussian samples. Specifically, we

4: consider shrinkage methods which are suitable for high dimensional

5: problems with a small number of samples (large $p$ small $n$).

6: First, we improve on the Ledoit-Wolf (LW) method by conditioning

7: on a sufficient statistic. By the Rao-Blackwell theorem, this

8: yields a new estimator called RBLW, whose mean-squared error

9: dominates that of LW for Gaussian variables. Second, to further

10: reduce the estimation error, we propose an iterative approach

11: which approximates the clairvoyant shrinkage estimator.

12: Convergence of this iterative method is established and a closed

13: form expression for the limit is determined, which is referred to

14: as the oracle approximating shrinkage (OAS) estimator. Both RBLW

15: and OAS estimators have simple expressions and are easily

16: implemented. Although the two methods are developed from different

17: persepctives, their structure is identical up to specified

18: constants. The RBLW estimator provably dominates the LW method.

19: Numerical simulations demonstrate that the OAS approach can

20: perform even better than RBLW, especially when $n$ is much less

21: than $p$. We also demonstrate the performance of these techniques

22: in the context of adaptive beamforming.

23: \end{abstract}

24: