On the Gaussian Cramér-Rao Bound for Blind Single-Input Multiple-Output System Identification: Fast and Asymptotic Computations

IEEE Access - Tập 8 - Trang 166503-166512 - 2020
Mohamed Nait-Meziane1, Karim Abed-Meraim1, Zhipeng Zhao2, Nguyen Linh Trung3
1PRISME Laboratory, University of Orléans, Orléans, France
2Mathematical and Algorithmic Sciences Lab, France Research Center, Huawei Technologies Company Ltd., Boulogne-Billancourt, France
3AVITECH, University of Engineering and Technology, Vietnam National University, Hanoi, Vietnam

Tóm tắt

The Cramér-Rao Bound (CRB) is a powerful tool to assess the performance limits of a parameter estimation problem for a given statistical model. In particular, the Gaussian CRB (i.e., the CRB obtained assuming the data are Gaussian) corresponds to the worst case; giving the largest CRB among a large class of data distributions. This makes it very useful in practice since optimizing under the Gaussian data assumption can be interpreted as a min-max optimization (i.e., minimizing the largest CRB). The Gaussian CRB is also the corresponding bound of Second-Order Statistics (SOS)-based estimation methods, which are frequently used in practice. Despite its practicality, computing this bound might be cumbersome in some cases, particularly in the case where the input is assumed deterministic and has a large number of samples. In this paper, we address this computational issue by proposing a fast computation for the deterministic Gaussian CRB of Single-Input Multiple Output (SIMO) blind system identification. More precisely, we exploit circulant matrix properties to reduce the cost from cubic to quadratic with respect to the sample size. Moreover, we derive a closed-form formula for the asymptotic (large sample size) Gaussian CRB and show how it can be computed using the residue theorem.

Từ khóa

#Blind system identification #Cramér-Rao Bound (CRB) #Single-Input Multiple-Output (SIMO) systems #fast and asymptotic computations

Tài liệu tham khảo

liu, 1994, A deterministic approach to blind identification of multi-channel FIR systems, Proc IEEE Int Conf Acoust Speech Signal Process (ICASSP), 4, 581 ahlfors, 1979, Complex Analysis An Introduction to the Theory of Analytic Functions of One Complex Variable 10.1109/5.622507 10.1007/BF01211654 10.1109/GLOCOM.2000.891295 10.1109/LSP.2004.836948 10.1109/SAM.2008.4606902 10.1049/ip-h-1.1983.0004 10.1007/BF01447872 10.1109/TSP.2004.836462 10.1109/TSP.2009.2023943 10.1016/j.mri.2016.10.014 stoica, 2001, Performance bounds for blind channel estimation, Signal Processing Advances in Wireless & Mobile Communications, 1, 41 kay, 1993, Fundamentals of Statistical Signal Processing 10.1109/78.348133 10.1109/MSP.2013.2238691 10.1109/ISSPA.2012.6310646 10.1109/78.558487 10.1109/TSP.2005.855100 10.1109/18.243446 10.1137/1031049 10.1080/14786444408520897 10.1109/78.552204 golub, 2012, Matrix Computations, 3 10.1111/j.2517-6161.1953.tb00131.x