Speaker
Description
Radiometers are central to radio astronomy but suffer from instrumental effects such as impedance mismatches between the antenna and receiver. Traditional calibration schemes like Dicke switching rely on mechanical or thermal reference loads, making them complex and less reliable in space environments. We present a machine learning–based calibration framework that models and removes instrumental distortions using neural networks trained on known signals. This method eliminates the need for active switching, improving stability, reducing mass and power requirements, and enabling simpler, more robust radiometer designs. Applied to experiments targeting the global 21-cm signal, it achieves the precision required for cosmological measurements and next-generation space missions.
