Compressive Sensing: From Algorithms to Circuits
Dr. Gianluca Setti
Professor of Engineering
Politecnico di Torino, Italy
Abstract
Many problems in modern information processing platforms can be modeled by the interaction of a stochastic process or algorithm (noise, interference, computation requests, …) with one or more electronic circuits. Under these conditions, any (near-)optimal system design procedure requires an understanding of the stochastic process or algorithm, a mapping of the algorithm to an implementation with appropriately tunable features, and finally the synergetic design of both the mapping and the algorithm to minimize the complexity of the underlying circuits or optimize the performance of the resulting overall system.
We will apply this procedure to two specific examples of algorithms:
1) Signal acquisition using Compressed Sensing (CS); and
2) Generation of tunable stochastic processes using nonlinear dynamical (chaotic) systems.
The first example focuses on CS, an acquisition technique which relies on the sparsity of the underlying signals to enable sampling below the classical Nyquist rate. We show that for signals whose energy concentrates in a specific spectral region, the CS acquisition sequences can be designed to maximize their capability to collect the energy of the samples (their “rakeness”) and increase by several dBs the average SNR achieved in signal reconstruction (ARSNR), or to reduce the energy per acquisition necessary to achieve the desired ARSNR. We also show how the use of rakeness can reduce the number of stages of a 0.18um CMOS A/D based on CS from 32 to 16 (64 to 24) for processing ECG (EMG) and that rakeness-derived sequences also eliminate the necessity for pre- or post-acquisition filtering stages used to suppress high frequency artifacts and 60-Hz power-line noise interference. Finally, we demonstrate through measurements on Samsung Exynos 5422 and TI TM4C1294NCPDT platforms that the use of rakeness during signal acquisition reduces the computational effort required for signal reconstruction.
In the second example, we show how nonlinear chaotic maps can be used as generators for stochastic processes with tunable statistical features easily embeddable and implementable in CMOS technology. This leads to multiple possible applications. The first one we will show is the implementation of a True RNG using a simple modification of a pipeline A/D converter. Contrary to other wide-spread commercial solutions (for instance by VIA and IdQuantique), the solution is easily implemented as a system-on- chip (SoC) and offers the opportunity for systematic design approaches. The second one is the reduction of the power spectrum peak for EMI in switching power converters or SATA interfaces. We will present experimental results showing that our method is able to reduce the power density spectrum peak for EMI by more than 7dB with respect to other state-of- the-art techniques.
Biography
Gianluca Setti received a Dr. Eng. degree in Electronic Engineering and a PhD degree in Electronic Engineering and Computer Science from the University of Bologna in 1992 and in 1997. He is currently a Professor with the School of Engineering at the University of Ferrara. He also held various visiting positions, most recently at the University of Washington, at IBM T. J. Watson Laboratories, and at EPFL (Lausanne). He is the recipient of numerous awards, including the 2004 IEEE Circuits and Systems (CAS) Society Darlington Award, the 2013 IEEE CASS Guillemin-Cauer Award and the 2013 IEEE CASS Meritorious Service Award. Dr. Setti has also served as Editor-in-Chief for both IEEE TCAS-I and IEEE TCAS-II and as the 2010 President of the CAS Society. In 2013-2014 he served as the Vice President for Publication Services and Products for the IEEE (the first one not from North America). He is also a Fellow of the IEEE. He has authored over 260 publications and has edited 3 books in the areas of nonlinear circuits, recurrent neural networks, implementation and application of chaotic circuits and systems, statistical signal processing, electromagnetic compatibility, biomedical circuits, and systems.
