Detection of illegal narcotics using NQR

Abstract:
This masters thesis deals with mathematical signal detection in the field of NQR,
Nuclear Quadrapole Resonance, which is a non-invasive spectroscopic technique used
to identify specific substances. The thesis has two main focuses, where the first is
to propose novel methods for parameter estimation, especially for identification of
unknown signals. These extend the existing Capon and APES methods for spectral
estimation to take the data model used for NQR signals into account. The less
general of them, named ETCAPA, works reasonably well for strong measurements.
This algorithm is a continuation of the ETCAPES method also derived during this
master thesis. The more general one, ETCapon, works less well for multi peak signals.
To the benefit of the algorithms, the number of frequency components must
not be defined in advance, which is the case for parametric models like ETAML
and least squares based estimation. The main contribution of these algorithms is to
find suitable search regions and to define the number of frequency components in
the signal, making it possible to use parametric algorithms for better estimation. A
pure interference canceling algorithm is also proposed, that uses a secondary data
set to remove any deterministic sinusoidal signals from the primary data. Initial
simulations indicate that this may work efficiently for simple interference signals.
The second focus of this thesis addresses the issue of detecting the illegal narcotic
methamphetamine in various situations. Together with the Itozaki Lab of
Osaka University and Tokyo Customs Lab in Japan, experiments have been made
possible in order to classify methamphetamine by identifying the parameters in the
data model, specific for the substance, and to find reasonable experimental settings
from which good detection can be made. It has also been confirmed that for quite
weak signals the ETAML detector is far superior to the commonly used FFT-based
method.