End-to-End Through-Wall Human Localization Network Using Raw Radar ADC Data

The radar signal processing algorithm is one of the core processes in through-wall radar human localization technology. However, traditional algorithms often struggle to adaptively handle low signal-to-noise ratio (SNR) echo signals in challenging and dynamic through-wall application environments. These environments, characterized by complex and varying conditions such as multi-path reflections and signal attenuation through different wall materials, exacerbate the difficulties in accurately localizing human targets. In this paper, we introduce a novel end-to-end through-wall radar human localization network, which directly processes raw radar Analog-to-Digital Converter (ADC) signals without any preprocessing. By bypassing traditional signal processing stages, our approach leverages the raw data to capture more comprehensive information from the environment, allowing for more robust and adaptable localization. To achieve this, we replace the conventional radar signal processing flow with the proposed DFT-based adaptive feature extraction (DAFE) module. This module employs learnable parameterized 3D complex convolution layers, which are specifically designed to extract superior feature representations from ADC signals. The ability to learn implicit features from raw data enables our network to overcome the limitations of traditional preprocessing methods, which often fail to adequately capture the nuances of low-SNR conditions. We rigorously trained and validated our proposed method on extensive data collected in real-world through-wall scenarios. The experimental results confirm the effectiveness and superiority of our approach, demonstrating its potential to significantly improve human localization accuracy in challenging environments.