PhD Thesis

< back to PhD thesis

« Non uniformly sampled signal processing : algorithm and architecture ».

Author: F. Aeschlimann
Advisor: M. Renaudin
Co-advisor: L. Fesquet
These de Doctorat Institut National Polytechnique de Grenoble - INPG
Speciality: signal, image, parole, télécoms
Defense: February 06 2006
ISBN: ISBN: 2-84813-078-4
Pages: 218


This PhD thesis deals with the development of new design approaches in order to reduce significantly the power consumption of Systems on Chips (SoC) and Communicating Objects used in digital signal processing. The goal is to obtain systems only driven by the events contained in the useful signal. In this context, a new kind of signal processing chain is proposed, combining an asynchronous design (no global clock) and a non-uniform sampling scheme called level-crossing sampling. As an analog-to-digital converter dedicated to this task has already been studied, this work is focussed on the sampled signal processing based on amplitude-time couples. A preliminary study shows that any operation has to use the temporal information. Then, Finite and Infinite Impulse Response Filters (FIR and IIR) are defined in the case of non-uniform sampled signals. Architectures are also proposed and compared to those commonly used showing that the computational complexity is increased. A criterion about the choice of the technology to favour, specifying the global computational load over a finite time, has proved that the asynchronous approach can compensate the complexity cost by decreasing the number of processed points. Thus the processing of low-active signals by an asynchronous chain, combining asynchronous design and non-uniform lead to a reduction of its average activity and so of the power consumption of the integrated circuits making this technology very attractive for Soc area.

pdf pdf

Other localisation