Terahertz (THz) time domain spectroscopy is emerging as a powerful tool to characterise
samples both chemically and physically. In this work different methods of
estimating spectroscopic parameters of a sample, its thickness and the uncertainty
of these estimates is presented. A number of case studies are also examined including
paracetamol polymorphs and a method of creating a spectroscopic simulant of
Semtex-H is presented.
Approximation of the sample spectroscopic parameters, real refractive index and
absorption coeficient were formed by building up a simple model of the samples
interaction with THz radiation. Methods of correcting unwrapping error in the real
refractive index were developed, including a method to correct in the presence of
discontinuities in the refractive index itself. These approximations were then applied
to extract parameters of both lactose and paracetamol samples.
An algorithm to generate spectroscopic simulants was developed and applied to
Semtex-H. These simulants consisted of simple mixtures of inert compounds, which
were measured and found to have similar spectrum to the target sample.
Methods of fitting resonant models to the sample response were developed to
extract both the spectroscopic parameters and sample thickness. These were refined
by calibrating for the Gaussian beam profile of the THz radiation, which was shown
to increase the accuracy of the extracted thickness. The thickness and spectroscopic
parameters of a lactose sample were measured with temperature, and it was found
that the spectroscopic parameter change was underestimated when thickness was
assumed constant.
A resonant model for multilayered samples was then developed and used to
characterise IPA in a flowcell measurement. This was then combined with a method
of time segmentation of the sample response, to extract spectroscopic parameters
and sample thickness simultaneously. This was then applied to a two layer sample,
to extract the spectroscopic parameters of a silicon and a quartz layer from a single
measurement.
Finally, methods of propagating the uncertainty from the time domain to the
spectroscopic parameters were developed. These were based on a multivariate normal
statistical model of the measurements andwere compared to numerical bootstrap
and Monte–Carlo estimates. These were used to develop confidence intervals for
the extracted refractive index, absorption coefficient and thickness. These methods
were applied to both a lactose and quartz sample.
