Analysis of Nonstationary Radiometer Gain Using Ensemble Detection

Abstract

Radiometer gain is generally a nonstationary random process, even though it is assumed to be strictly or weakly stationary. Since the radiometer gain signal cannot be observed independently, analysis of its nonstationary properties would be challenging. However, using the time series of postgain voltages to form an ensemble set, the radiometer gain may be characterized via radiometer calibration. In this article, the ensemble detection algorithm is presented by which the unknown radiometer gain can be analytically characterized when it is following a Gaussian model (as a strictly stationary process) or a 1st order autoregressive, AR(1) model (as a weakly stationary process). In addition, in a particular radiometer calibration scheme, the nonstationary gain can also be represented as either Gaussian or AR(1) process, and parameters of such an equivalent gain model can be retrieved. However, unlike stationary or weakly stationary gain, retrieved parameters of the Gaussian and AR(1) processes, which describe the nonstationary gain, highly depend on the calibration setup and timings. Index Terms-Autoregressive AR(1) model, ensemble detection, nonstationary radiometer gain, radiometer calibration. I. INTRODUCTION R ADIOMETERS are widely used to measure geophysical parameters to examine variations in the earth and planetary systems. These measurements typically need to be obtained over large temporal or spatial scales. However, enhanced radiometric accuracy and sensitivity are required in microwave radiometry, as an accurate radiometer facilitates the possibility of high-resolution contrasts in variations of physical parameters from the rest of the measured noise. For instance, the retrieval of geophysical parameters, such as precipitable water vapor, ocean surface salinity, wind measurements, and liquid and ice water paths require enhanced accuracy and finer resolution [1]-[5]. Thus, given the increasing importance of radiometer calibration in deriving greater geophysical information from Manuscript

Reducing Instrument Power Using Machine Learning Calibration

“…John Bradburn (1) , Mustafa Aksoy (1) , and Paul E. Racette (2) (1) University at Albany SUNY, Albany, NY, USA; e-mail: jbradburn@albany.edu; maksoy@albany.edu…”

“…Data collected from laboratory receivers have been analyzed to learn the receiver gain characteristics and properties of the transient effects at the receiver turn on. These data are used for tuning a synthetic radiometer model [1,2] that is used to generate a training dataset for the CNN. The CNN model is evaluated on laboratory data collected within this training distribution.…”