AI-Enhanced RF Interference Rejection Using Transformer Models

A new study on arXiv (2604.22816) explores AI-enhanced interference rejection in radio frequency transmissions. Deep learning models trained on both the signal of interest and the signal mixture outperform traditional approaches that only consider the signal of interest. The goal is to detect, demodulate, and decode signals across varying signal-to-interference-plus-noise levels without detailed knowledge of the interfering signal or propagation conditions. The researchers used Autoregressive Transformer Decoder models, achieving orders of magnitude faster inference throughput than earlier WaveNet models. As a specific example, they investigated an analog FM 'Walkie Talkie' signal in the presence of an Orthogonal Frequency-Division Multiplexing interferer.