Dark Signal Modeling for UV CCD-Array Spectrometer Using Ghost Pixels

Abstract

Ultraviolet (UV) solar radiation causes multiple health detriments and materials degradation, these effects being strongly wavelength-dependent. Although coupled-charge device (CCD)-array spectrometers are an alternative to expensive and more labor-intensive scanning ones, they suffer from different sources of error, such as the dark signal and its Poisson noise associated, that must be previously characterized in depth to guarantee the quality of the measurements. This study proposes using the signal registered by the ghost pixels available in the instruments under study to estimate the dark signal. These ghost pixels are different from the known “blind pixels” since these ones register the dark signal, and the first ones were found to correspond to the output preamplifier offset. More than 7000 dark signal frames were measured at different temperatures and integration times for this study. The ghost-pixels signal was found to depend linearly on the temperature and be independent on the integration time, so it can be a proxy for temperature in dark signal correction models. To estimate the dark signal, several multivariate models dependent on the integration time and the ghost-pixels signal were tested. Finally, the best two, according to the Akaike information criterion (AIC), were selected. The chosen models perform notably well, with $R^{2}$ values over 0.95 and low rRMSE values, below 25%. One of the ghost-pixels models chosen compares well with temperature-based models, although the latter show lower relative root mean squared error (rRMSE) values. This study presents the ghost-pixels signal as a reliable and efficient alternative to temperature measurements for models aimed at estimating the dark signal.