CNN-Based RFI Detection for SMAP Radiometer

Radio-frequency interference (RFI) poses a significant challenge for the Soil Moisture Active Passive (SMAP) mission’s L-band radiometer, which measures natural microwave emissions from the Earth’s surface for soil moisture and freeze/thaw state retrievals. While SMAP employs onboard and ground-based algorithms for RFI detection, these methods can struggle with weak or non-stationary interference patterns.

This project introduces a convolutional neural network (CNN) approach for automated RFI detection, trained on time–frequency (TF) representations of SMAP radiometer data. The CNN learns to classify TF bins as contaminated or clean, enabling finer-grained detection than traditional threshold-based algorithms.

Methodology

• Input Representation:
  SMAP’s Level 1B radiometer data is transformed into spectrograms capturing power variations across frequency and time.

• CNN Architecture:
  A multi-layer convolutional model extracts spatial–spectral features from the TF input to identify RFI patterns.

• Training Data:
  The network is trained on labeled datasets containing both RFI-contaminated and clean TF samples, derived from known SMAP overpasses in RFI-heavy and RFI-free regions.

• Output:
  Binary classification masks marking contaminated TF bins, enabling selective removal before geophysical retrieval.

Evaluation

• Performance:
  Achieves higher detection accuracy, precision, and recall than SMAP’s operational kurtosis and threshold methods.

• Robustness:
  Effective against varying interference types, including continuous wave, pulsed, and frequency-hopping signals.

• Impact on Science Data:
  Improves brightness temperature retrieval accuracy, especially in moderate RFI environments where legacy methods underperform.

Impact

By integrating deep learning into SMAP’s RFI mitigation pipeline, this work demonstrates the potential for spaceborne AI-enhanced signal cleaning. The framework could be adapted to other passive sensing missions operating in congested spectrum bands.

Read the full paper: IEEE Xplore – CNN-Based RFI Detection for SMAP Radiometer