Title: "Sensing matrix design criteria for adaptive compressed sensing"
Speaker: Valerio Cambareri (U. Bologna, Italy)
Location: "Shannon" Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor
Date / Time (duration): Wednesday 23/04/2014, 10h00 (~ 45')
Abstract: The quest for optimal sensing operators is crucial in the design of efficient architectures that perform compressive sampling of analog signals. While exact theoretical results exist for general sensing operators that guarantee the recovery of sparse signals, such guarantees are strict and often neglected in practical implementations. Moreover, natural signals are only approximately sparse, but often exhibit correlation properties that suggest a further possible optimization of the sensing operator under some signal-domain priors.
In this talk, we review two criteria that leverage on such properties for the design of adaptive sensing strategies. The first criterion aims at designing correlated subgaussian random matrices so that the measurements' energy is maximized under simple signal-domain correlation priors; this approach is shown to be generally applicable and improving the recovery performances when the signal verifies such priors. The second criterion aims at designing sensing matrices from sets of deterministic sensing vectors so that, under similar signal-domain priors, the information carried by the measurements is optimized in a maximum-entropy rationale; this optimization strategy allows a reduction of the number of measurements required to provide accurate recovery of the signal. These two approaches are shown to be widely applicable to compressive sampling of natural signals with an emphasis on physically realizable sampling architectures that can be designed from them.
Site for supporting code (on Maximum Entropy Sensing Design):
Biography: Valerio Cambareri received the B.S. and M.S. degrees (cum laude) in Electronic Engineering from the University of Bologna, Italy, in 2008 and 2011 respectively. He is currently a Ph.D. Student in Electronics, Computer Science and Telecommunications at DEI - University of Bologna, and a member of the Statistical Signal Processing Group at ARCES - University of Bologna. His research interests include compressed sensing, statistical signal processing and hyperspectral imaging.
Last updated May 04, 2014, at 10:36 PM