Biograph: Prof. Wu received the Ph.D. degree in information and communication engineering from the National Key Lab. for Radar Signal Processing at Xidian University in December 2003. From April 2004 to September 2006, he was a post-doctoral researcher at Institute of Acoustics, Chinese Academy of Sciences, Beijing, China.  During Oct. 2006–Feb. 2008, he worked as a senior research fellow at the City University of Hong Kong.  He was a Visiting Researcher in the Faculty of Engineering, Bar-Ilan University from April 2013 to March 2014, Israel. Currently, he is a full-time professor in Wuhan Institute of Technology.  At the same time, he is also a “Chutian Scholar Project in Hubei Province” distinguish professor at the same university. He has published over 100 journal and conference papers, and he is a senior member of the Chinese Institute of Electronic Engineers (CIE), an associate editor of Multidimensional Systems and Signal Processing (an international journal of Springer publisher). His research interests include signal detection, parameter estimation in array signal processing and source localization for wireless sensor networks, biomedicine signal analysis, etc.

Title: Multidimensional Frequency Estimation using Unitary PUMA Algorithm without Pairing Parameters 

Abstract: Multidimensional frequency estimation problem is an attractive research topic in recent decades, which has various applications in different fields such as sonar, wireless communications, radio astronomy observation, MIMO wireless channel detection, speech location, array signal processing, etc. In this talk, an algorithm of combining projection separation and unitary principal-singular-vector utilization for model analysis (PUMA) is reported to solve the problem of multidimensional sinusoidal frequency estimation, which is realized in terms of real-valued computations. Tensor representation is exploited to improve the parameter estimation accuracy. The frequency of $r$th dimension is first estimated from the autocorrelation of the $r$-mode signal tensor, using a subspace-based method where the subspace is obtained from a real-valued eigen-decomposition which is realized using a unitary matrix. The estimated frequencies are then utilized to construct the projection separation matrix to separate all frequencies in the remaining dimensions contained in the signal tensor. Unitary PUMA is then applied to estimate these separated frequencies in terms of real-valued computations while multidimensional frequency pairing is automatically achieved. Because of the real-valued computations, the presented algorithm is more computationally efficient than a number of state-of-the-art methods. Simulation results are included to validate the performance of the proposed algorithm in terms of effectiveness and accuracy.