Finite Rate of Innovation sampling techniques for embedded UWB devices.

This report studies the applicability of Finite Rate of Innovation (FRI) algorithms to UltraWide Band (UWB) communications, more precisely in the scope of Low Power Body Area Networks (LP-BAN ). Three main issues are studied and given proposed solutions. First, the classical FRI algorithm is modified to accomodate different symmetrical pulse shapes. Such a modification – necessary to get acceptable performances – is done by a simple equalization. Second, LP-BAN devices limitations such as drift, jitter and aggressive quantization are blended in the algorithm. It is done by adjusting the equalization template and development of a suited quantization algorithm. Third and last, the cost of FRI denoising procedure (Cadzow denoising) is greatly reduced to fit the requirements of a low power embedded device. It is centered on performing most of the computations in a low-dimension Krylov subspace of the matrix to be denoised. The particular structure of the projected matrix enables selective computation of the eigenpairs. The result is an algorithm able to resolve close paths within a reasonnable computational budget. Some issues remain on quantization.