Abstract:
The problem of recovering samples lost from time series or sensor data is important in signal
processing. When the underlying signal is known to be bandlimited, and the sample rate is higher
than the Nyquist rate, the samples are dependent. In this case a missing sample or samples can
be recovered from the remaining samples. In the absence of noise, the accuracy of the sample
estimates depends on the degree of oversampling and the total number of good samples available.
In previous work, researchers often assumed that large numbers of high quality (high signal-to-noise
ratio) samples were available. This assumption may not be valid in practice. In practice the number
of samples is finite and the signal is corrupted by noise. The truncation and the noise will result
in errors in the sample estimates. This thesis investigates a least squares solution to the problem,
and uses the data from SwellEx-96 experiment to evaluate several approaches, including the least
squares approach.
