Dr. Zhi Tian from Michigan Technological University made a report on “Compressive Covariance Sensing” at F804 of New Main Building on October 31st, Friday.

Abstract:
Compressive signal sampling is one of the recent important advances in signal processing and statistical learning, with impact to various applications including data sciences, communications, sensor networks, and medical imaging. It requires information-bearing signals to be sparse over known domains, either naturally or by design. In this talk, I will introduce the fresh notion of compressive covariance sensing, and advocate its exciting implications for (cyclo) stationary processes characterized by second-order statistical descriptors. Such descriptors include (periodic) covariances or frequency, cyclic, angular and Doppler spectra, which already effect signal compression even for non-sparse signals. Using this key observation, I will demonstrate how the attribute of sparsity can be bypassed, or leveraged more effectively, when recovering the second-order statistical information of interest. I will also delineate the minimal sampling rates for recovering certain useful statistics of non-sparse random signals, along with the compressive sampler designs for approaching such rates. I will illustrate the usefulness of compressive covariance sensing using several engineering applications that rely on frequency or angular spectrum sensing, such as wireless cognitive radio and statistical array processing.