Out-of-Distribution Radar Detection in Compound Clutter and Thermal Noise through Variational Autoencoders

TL;DR

This work tackles radar target detection in environments with heterogeneous clutter, including compound Gaussian noise and additive thermal noise. It introduces a variational autoencoder (VAE) trained on noise-only data to perform out-of-distribution detection, using reconstruction error as the detection statistic and a calibrated threshold to control the false alarm rate. Compared to classical detectors such as MF, NMF, and adaptive variants, the VAE detector shows superior robustness in challenging noise conditions, including compound clutter with thermal noise and Doppler diversity. The approach enables target detection without labeled target data and demonstrates strong practical potential for reliable CFAR-like performance in real-world radar scenarios.

Abstract

This paper presents a novel approach to radar target detection using Variational AutoEncoders (VAEs). Known for their ability to learn complex distributions and identify out-ofdistribution samples, the proposed VAE architecture effectively distinguishes radar targets from various noise types, including correlated Gaussian and compound Gaussian clutter, often combined with additive white Gaussian thermal noise. Simulation results demonstrate that the proposed VAE outperforms classical adaptive detectors such as the Matched Filter and the Normalized Matched Filter, especially in challenging noise conditions, highlighting its robustness and adaptability in radar applications.

Figures (4)

• Figure 1: VAE network architecture
• Figure 2: $\mathcal{L}_{\text{rec}}$ histogram across SNRs for cCGN + AWGN scenario.
• Figure 3: Detection performance under different noise for Doppler bin $d=0$ ($P_{fa}= 10^{-2}$, $\rho=0.5$, $\mu=1$, $m=16$, $K=32$).
• Figure 4: $P_d$-SNR-Doppler bin map comparing VAE to AMF and ANMF, under different noise scenarii.