Robust multi-dimensional model order estimation in the presence of brief sensor failures

Abstract

In this work, a robust version of the tensor-based model order selection technique R-D Exponential Fitting Test (EFT) is developed to deal with the model order estimation in the presence of brief sensor failures. Moreover, extensions of the previously proposed robust R-D Akaike's Information Criterion (AIC) and robust R-D Minimum Description Length (MDL) are proposed by constructing a real-valued measurement tensor where forward-backward averaging is incorporated. Then the unfoldings of the real-valued data tensor are used as the input of the robust covariance estimator instead of stacking the real and imaginary parts of the unfoldings of the original complex-valued measurement tensor. Such enhanced versions of robust R-D AIC, robust R-D MDL, and robust R-D EFT experience an improved performance and a reduced computational complexity. Numerical simulations are performed to demonstrate the promising performances of the proposed tensor-based robust model order selection techniques in the presence of brief sensor failures.