| dc.description.abstract | This thesis performs a quantitative study, derived from the Neyman-Pearson
framework, on the robustness of the matched filter detector corrupted by zero mean,
independent and identically distributed white Gaussian noise. The variance of the
noise is assumed to be imperfectly known, but some knowledge about a nominal
value is presumed. We utilized slope as a unit to quantify the robustness for different
signal strengths, nominals, and sample sizes. Following to this, a weighting method
is applied to the slope range of interest, the so called tolerable range, as to analyze
the likelihood of these extreme slopes to occur. A ratio of the first and last quarter
section of the tolerable range have been taken in order to obtain the likelihood ratio
for the low slopes to occur. We finalized our analysis by developing a method that
quantifies confidence as a measure of robustness. Both weighted and non-weighted
procedures were applied over the tolerable range, where the weighted procedure puts
greater emphasis on values near the nominal.
The quantitative analysis results show the detector to be non-robust and deliver
poor performance for low signal-to-noise ratios. For moderate signal strengths, the
detector performs rather well if the nominal and sample size are chosen wisely. The
detector has great performance and robustness for high signal-to-noise ratios. This
even remains true when only a few samples are taken or when the practitioner is
uncertain about the nominal chosen. | en |
