Radar Image Reconstruction from Raw ADC Data using Parametric Variational Autoencoder with Domain Adaptation

TL;DR

This paper introduces a novel parametric variational autoencoder framework for direct radar target detection from raw ADC data, employing domain adaptation to enhance generalization from simulated to real data, outperforming traditional methods.

Contribution

It proposes a parametric variational autoencoder with domain adaptation for radar target detection directly from raw ADC data, improving generalization and detection accuracy.

Findings

Superior detection and localization performance compared to traditional methods.

Effective domain adaptation from simulated to real radar data.

Outperforms state-of-the-art deep learning architectures like U-Net.

Abstract

This paper presents a parametric variational autoencoder-based human target detection and localization framework working directly with the raw analog-to-digital converter data from the frequency modulated continous wave radar. We propose a parametrically constrained variational autoencoder, with residual and skip connections, capable of generating the clustered and localized target detections on the range-angle image. Furthermore, to circumvent the problem of training the proposed neural network on all possible scenarios using real radar data, we propose domain adaptation strategies whereby we first train the neural network using ray tracing based model data and then adapt the network to work on real sensor data. This strategy ensures better generalization and scalability of the proposed neural network even though it is trained with limited radar data. We demonstrate the superior detection and localization performance of our proposed solution compared to the conventional signal processing pipeline and earlier state-of-art deep U-Net architecture with range-doppler images as inputs