RF-SHIELD: AI-Driven RF Sensemaking

Motivation

The RF environment is increasingly congested and increasingly difficult to interpret: many emitters, overlapping bands, short transmissions, and waveforms that were never catalogued. Analysts need systems that turn raw wideband RF observations into an understandable picture of what is present, what is unusual, and what deserves attention — with reasoning that a human can follow.

Approach

RF-SHIELD develops an AI-driven RF sensemaking framework built on interpretable learning architectures and software-defined radio experimentation:

• Wideband detection and classification — detecting emissions and recognizing waveform and modulation types directly from complex-valued RF data.
• Interpretable architectures — parameterized learnable filters (SincNet- and PLFNet-style) whose learned filters correspond to physically meaningful frequency responses, so classification decisions can be inspected rather than taken on trust.
• Anomaly and threat detection — identifying emissions that do not match known classes, and characterizing them for further analysis.
• SDR implementation — evaluation on recorded and live over-the-air data rather than synthetic signals alone.

Related work

• PLFNets: Interpretable Complex-Valued Parameterized Learnable Filters for Computationally Efficient RF Classification, IEEE Transactions on Radar Systems, 2024.
• Learnable Gaussian Filter-Based Automatic RF Waveform Modulation Recognition, IEEE Radar Conference, 2025.
• Automatic classification of radar and communication waveforms through interpretable deep learning, SPIE, 2025.