A comparison of nonlinear filtering methods for blackbody radiation applications in photonics

Kalman filtering (KF) is a widely used filtering technique in highly predictable temporal-mechanical systems where system noise can be modelled with a gaussian function. Improving the signal quality during acquisition is conventionally accomplished by increasing integration time in acquisition. However, this increases the signal acquisition time in photonic systems. In high noise applications, acquisition time is low, and this post-process filtering technique can be applied to increase signal quality. This work explores the comparison of the KF, and nonlinear filtering methods to a simulated blackbody radiation signal where gaussian noise is added to mimic electrical interference. Three filters are selected for comparison on the ability to improve the root mean square error (RMSE) of a simulated measured signal with respect to a simulated actual signal. The filters that are compared in this work are the Extended Kalman Filter (EKF), the Unscented Kalman (UKF), and the Extended Sliding Innovation Filter (ESIF). The filters use a calibration temperature that the filter model uses to determine expected values. To compare the filters, the RMSE is evaluated when error is introduced to the simulation by changing the actual temperature to values equal, below, and above the calibration temperature. Two additional scenarios were considered to test filter robustness. The first scenario uses changes in model temperature occurring as a function of wavelength (i.e., temperature change mid-scan). The second scenario introduces impurities with different emission values. The ESIF demonstrated favorable performance over the other considered filters, showing promise in optical applications.