Radiometer calibration using machine learning

Area of Science:

• Radio astronomy
• Cosmology
• Instrumental calibration

Background:

• Radiometers are key radio astronomy instruments, measuring electromagnetic radiation intensity.
• Impedance mismatches in receivers cause signal distortions, challenging traditional calibration like Dicke switching.
• Detecting the faint, high-redshift 21-cm signal is a major cosmological challenge.

Purpose of the Study:

• Introduce and test a novel machine learning (ML) calibration framework for radiometric experiments.
• Achieve the precision necessary for detecting the challenging sky-averaged 21-cm signal.
• Provide an alternative to traditional calibration methods for complex radio astronomical systems.

Main Methods:

• Utilize neural networks trained on known signal sources to model instrumental effects.
• Develop a machine learning-based calibration framework for radiometers.
• Test the framework's performance in achieving required radiometric precision.

Main Results:

• The ML calibration framework demonstrates capability for high-precision radiometric measurements.
• Successfully models and corrects for complex instrumental effects.
• Achieves precision suitable for detecting the faint 21-cm signal.

Conclusions:

• Machine learning offers a powerful approach to calibrate complex radiometric systems.
• The developed ML framework meets the precision requirements for 21-cm signal detection experiments.
• This work advances observational cosmology by enabling new detection capabilities.