Enabling Low-power Radiometers with Machine Learning Calibration

Abstract

Extracting the maximum value from data obtained by smart sensors will require identifying and leveraging trends and mutual information in raw data, calibration measurements, and telemetry data. Use of machine learning algorithms enables the extraction of information content which is not exploited by current approaches. To reduce power draw of radiometric sensors, we propose an approach using machine learning to produce calibrated radiometer measurements prior to reaching steady state. By enabling calibration during instrument power cycling, or between instrument turn-on and reaching equilibrium, the average power draw of the sensor can be reduced, while reducing gaps in data acquisition and increasing capabilities of existing sensor platforms.