Recently there has been a surge of interest in sampling theory in signal processing
community. New efficient sampling techniques have been developed that allow
sampling and perfectly reconstructing some classes of non-bandlimited signals at
sub-Nyquist rates. Depending on the setup used and reconstruction method involved,
these schemes go under different names such as compressed sensing (CS),
compressive sampling or sampling signals with finite rate of innovation (FRI).
In this thesis we focus on the theory of sampling non-bandlimited signals
with parametric structure or specifically signals with finite rate of innovation. Most
of the theory on sampling FRI signals is based on a single acquisition device with
one-dimensional (1-D) signals. In this thesis, we extend these results to the case of
2-D signals and multichannel acquisition systems. The essential issue in multichannel
systems is that while each channel receives the input signal, it may introduce
different unknown delays, gains or affine transformations which need to be estimated
from the samples together with the signal itself. We pose both the calibration of
the channels and the signal reconstruction stage as a parametric estimation problem
and demonstrate that a simultaneous exact synchronization of the channels and reconstruction
of the FRI signal is possible. Furthermore, because in practice perfect
noise-free channels do not exist, we consider the case of noisy measurements and
show that by considering Cramer-Rao bounds as well as numerical simulations, the
multichannel systems are more resilient to noise than the single-channel ones.
Finally, we consider the problem of system identification based on the multichannel and finite rate of innovation sampling techniques. First, by employing our
multichannel sampling setup, we propose a novel algorithm for system identification
problem with known input signal, that is for the case when both the input signal and
the samples are known. Then we consider the problem of blind system identification
and propose a novel algorithm for simultaneously estimating the input FRI signal
and also the unknown system using an iterative algorithm.