Implications of 1/f noise in uncooled thermal imaging

Abstract

The presence of 1/f noise in staring imagers is problematic in several ways. It is well-known that in bolometers it sets a limit on the degree to which one can improve NETD by increasing the detector bias, because 1/f noise and responsivity both increase linearly with bias. Perhaps more importantly, at very low frequencies 1/f noise is manifest as drift and appears as spatial noise, which impairs target recognition far more than does an equivalent measure of temporal noise. This results from the fact that the eye filters temporal noise to a far greater extent than it does spatial noise. The perceived spatial noise resulting from 1/f noise leads to a degradation of performance with time after the insertion of a calibration shutter, even if the temperature is perfectly stable. Averaging many samples is not as effective for reduction of 1/f noise as it is for white noise, because the low frequency components are coherent over multiple samples. Additional implications of 1/f noise include uncertainty in measurement of signal and both temporal and spatial noise, as well as inherent errors in nonuniformity correction.