IRNet: Interference Recognition Networks for Automotive Radars via Autocorrelation Features

With the rapid increase of the number of autonomous driving vehicles, more and more radars working on a similar frequency band are equipped on vehicles, which incurs risks of automotive radar interference. Therefore, a precise interference recognition method is becoming a compelling task. In this article, a novel interference recognition network, called IRNet, is proposed to recognize eight types of interference signals that include mutual interference, unintended interference, and adverse intended interference. First, autocorrelation features of interference signals are obtained via calculating their autocorrelation functions (ACFs) and converted as two-dimensional feature images. Next, these images are employed as inputs of the IRNet that owns significant representation power and adaptive feature selection ability. Finally, the prediction types of interference will be output via the IRNet directly. The robustness of autocorrelation features is verified via simulations. The simulation results show that the proposed method can achieve 90.85% overall recognition accuracy when the jamming-to-noise ratio (JNR) is −14 dB and nearly 100% when JNR
$> -8$
dB. Compared with six state-of-the-art methods, the IRNet achieves much better recognition performance, especially under lower JNR conditions with relatively lower and acceptable computational complexity. Results on measured signals also verify the powerful generalization ability of the proposed method.