Novel Technique for Compton edge Calibration of Neutral Particle
Detectors using Kernel Density Estimation
Abstract
Simple neutral particle detectors consisting of organic scintillators,
sensitive photodetectors, and readout systems have been used to great
effect in space- and Earth-based science for observing gamma rays and
particle radiation from solar weather, astrophysical phenomena, and even
lightning storms on Earth. The calibration of simple gamma ray detectors
was improved through kernel density estimation (KDE) in a user-friendly
Jupyter notebook running a Python script. When calibrating organic
particle detectors, a gamma ray source is applied and the distribution
of raw voltages produced by the detector are represented in a histogram
including a distinct Compton edge. The location of the edge in the
histogram can be associated with the energy of the gamma ray source to
convert the histogram of voltage measurements into a representation of
the gamma ray energies interacting with the detector. The process of
estimating the location of the edge on a histogram of real measurements
with imperfect resolution is often done by hand or with a half gaussian
fit, but we propose an alternative technique where KDE of the raw
voltages replaces the histogram. Using KDE, Gaussian kernels are built
up to a representation of the underlying energy distribution which can
be easily numerically differentiated. The minimum of the modified
distribution identifies the location of the Compton edge, regardless of
the resolution of the detector. To demonstrate the technique on a model
of a Compton edge, we calculated the theoretical Klein-Nashina Compton
distribution of Cesium-137 gamma ray interactions, created a model of
the edge with a piecewise 2nd order polynomial, and convolved the
piecewise function with a Gaussian to approximate the experimental
response. The response was numerically differentiated and the location
of the minimum was consistent with the energy of the Compton edge, as
the inflection point of the approximated Gaussian distribution occurs at
the Compton edge. We also demonstrate the technique with data from a
MEGAlib simulation of a simple block particle detector. For projects
with repeated adjustment and testing of particle detectors, this method
should improve consistency and accuracy of calibration while reducing
analysis time.