Efficient Data Fusion Using Random Matrix Theory

Abstract

This letter addresses multisensor data fusion under the Gaussian noise. Under the Gauss-Markov model assumptions, data fusion based on maximum likelihood estimation (MLE) is the minimum variance unbiased estimator. Nonetheless, we propose a linear fusion algorithm based on the random matrix theory, which yields a biased estimator. The proposed estimator has a lower mean squared error (MSE) than the MLE fusion method when the dimensionality of signal is larger than the number of sensors, which is the typical use case in modern fusion systems. The fusion coefficients are directly solved in the proposed method without iteration, and this method can be considered as an approximate implementation of the linear minimum MSE (LMMSE) estimator. Numerical simulations demonstrate the performance gain of the proposed fusion method.