Spectrometer Calibration by Expectation Maximization Method

Lulin Yuan

DOI: https://doi.org/10.48550/arXiv.physics/0512132

2005-12-15

Abstract:Expectation Maximization (EM) algorithm is a parameter estimation method from incomplete observations. In this paper, an implementation of this method to the calibration of HKS spectrometer at Jefferson Lab is described. We show that the application of EM method is able to calibrate the spectrometer properly in the presence of high background noise, while the traditional nonlinear Least Square method fail. The preliminary results of HKS spectrometer calibration is presented.

Data Analysis, Statistics and Probability,Instrumentation and Detectors

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the spectrometer calibration problem under high background noise. Specifically, when researchers at Hampton University were using the (e,e'K) reaction to perform high - resolution hypernuclear spectroscopy measurements in the HKS experiment at Jefferson Lab, they encountered the problem that the traditional nonlinear least - squares method (NLS) could not effectively handle high background noise during the calibration process.

### Core of the problem

1. **High background noise**: The HKS spectrometer detects electrons and K - mesons at very forward angles in actual experiments to increase the hypernuclear yield. This results in high accidental background noise generated by bremsstrahlung photons.
2. **Low signal - to - noise ratio**: In the actual data, the signal - to - noise ratio is almost 1:1, causing the traditional NLS method to produce incorrect results during calibration. As shown in Figure 2, the accidental background under the 12ΛB ground - state peak forms an artificial "bump".

### Solution

To overcome these problems, the author introduced the Expectation - Maximization (EM) algorithm for spectrometer calibration. The EM algorithm is a statistical method for estimating parameters from incompletely observed data and can better handle observed data diluted by a large amount of noise. Through the EM algorithm, the author was able to correctly calibrate the spectrometer in the presence of high background noise.

### Main steps

1. **Define hidden variables**: Introduce a hidden variable \( S_i \), which indicates whether the \( i \) - th event is a true signal (\( S_i=1 \)) or noise (\( S_i = 0 \)), although these variables cannot be directly observed.
2. **Energy function**: Define an energy function \( E(\{S_i\}, \{\pi\}) \) and introduce a truncation parameter \( \lambda \):
\[
E(\{S_i\}, \{\pi\})=\sum_{i}[S_i\Delta M_i^2+\lambda(S_i - 1)^2]
\]
3. **EM algorithm**: Divide the EM algorithm into the expectation step and the maximization step. In the expectation step, calculate the expected value of the energy function with respect to the hidden variables; in the maximization step, minimize the expected - value function \( Q(\{\pi\}|\{\pi'\}) \):
\[
Q(\{\pi\}|\{\pi'\})=\sum_{\{S_i\}}E(\{S_i\}, \{\pi\})P(\{S_i\}|\{\pi'\})
\]
4. **Iteratively re - weighted least - squares method**: Finally, the EM algorithm can be regarded as an iteratively re - weighted least - squares process, where the weights are no longer constants but depend on \( \Delta M_i \) or the parameter set \( \{\pi\} \).

Through these steps, the EM algorithm can effectively handle high background noise and provide more accurate spectrometer calibration results. Preliminary results show that the EM algorithm performs well in calibrating the HKS spectrometer and is superior to the traditional NLS method.