In recent years, a growing number of researchers have attempted to overcome the constraints of size and scope in different medical studies to find out the overall treatment effects. As a widespread technique to combine results of multiple studies, commonly used meta-analytic approaches for continuous outcomes demand sample means and standard deviations of primary studies, which are absent sometimes, especially when the outcome is skewed. Instead, the median, the extrema, and/or the quartiles are reported. One feasible solution is to convert the preceding order statistics to demanded statistics to keep effect measures consistent. In this article, we propose new methods based on maximum likelihood estimation for known distributions with unknown parameters. For unknown underlying distributions, the Box–Cox transformation is applied to the reported order statistics so that the techniques for normal distribution can be utilized. Two approaches for estimating the power parameter in Box–Cox transformation are provided. Both simulation studies and real data analysis indicate that in most cases, the proposed methods outperform the existing methods in estimation accuracy.
1 November 2018 in IEEE Transactions on Signal Processing
Distributed estimation fusion is concerned with combination of local estimates from multiple distributed sensors to produce a fused result. In this paper, we characterize local estimates as posterior probability densities, and assume that they all belong to a parametric family. Our starting point is to consider this family as a Riemannian manifold by introducing the Fisher information metric. From the perspective of information geometry, the fused density is formulated as an informative barycenter in the space of probability densities and sought by minimizing the sum of its squared geodesic distances from the local posterior densities. Under Gaussian assumptions, a geodesic projection (GP) method and a Siegel distance (SD) method in the information-geometric framework are proposed to tackle the problem. The GP method gives a fusion result in accord with the covariance intersection estimate but under an information-geometric criterion, while the SD method appears to achieve a better approximation of the informative barycenter. Numerical examples are provided to demonstrate the performance of the proposed estimation fusion algorithms.
12 September 2019 in IEEE Transactions on Signal Processing
This paper deals with adaptive radar detection of targets in the presence of Gaussian disturbance sharing a block-diagonal covariance structure. The problem is formulated according to a very general signal model, which contains the point-like, range-spread, and subspace target (or targets) as special instances. Hence, a unified study on the resulting adaptive detection problem is handled with the use of the invariance theory. The obtained results, including an appropriate transformation group, a maximal invariant and an induced maximal invariant, are proven consistent with those existing in the literature for some simple scenarios. Meanwhile, since the widely-used generalized likelihood ratio detector does not admit a closed form expression, new invariant detectors and their CFAR versions are proposed in this general scenario. Finally, their detection performance is assessed and validated via numerical examples.
8 March 2017 in IEEE Transactions on Signal Processing
The invariance principle is adopted to develop an exhaustive study for adaptive detection of range-spread targets in Gaussian noise sharing a block-diagonal covariance structure. For this problem, the usual generalized likelihood ratio principle is intractable. In this paper, we first determine the largest group of affine transformations that does not alter the decision problem. Then a maximal invariant identified by this group is derived, which can characterize the totality of the invariant detectors and extends the existing results for the point-target case. A theoretical performance analysis of the maximal invariant is also given. Finally, we propose two classes of invariant detectors which are distinguished by whether a constant false alarm rate (CFAR) is maintained. Numerical experiments are provided for a comparison of the proposed detectors, where a tradeoff between the CFARness and the improved performance has been observed and studied.
13 March 2019 in IEEE Transactions on Signal Processing
This paper addresses the estimation of large-dimensional covariance matrices under both normal and nonnormal distributions. The shrinkage estimators are constructed by convexly combining the sample covariance matrix and a structured target matrix. The optimal oracle shrinkage intensity is obtained analytically for any prespecified target in a class of matrices which includes various structured matrices such as banding, thresholding, diagonal, and block diagonal matrices. After deriving the unbiased and consistent estimates of some quantities in the oracle intensity involving unknown population covariance matrix, two classes of available optimal intensities are proposed under normality and nonnormality, respectively, by plug-in technique. For the target matrix with unknown parameter such as bandwidth in banded target, an analytic estimate of unknown parameter is provided. Both the numerical simulations and applications to signal processing and discriminant analysis show the comparable performance of the proposed estimators for large-dimensional data.