Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detecting radar targets swarms in range profiles with a partially complex-valued neural network

Martin Bauw

TL;DR

The paper addresses radar target detection in range profiles under clutter, proximity effects, and waveform distortions by replacing conventional pulse compression and CA-CFAR with a partially complex-valued neural network that processes the full received signal $\mathbf{x}$ to output a detection profile. The proposed RX-scale processing uses a hybrid complex-real neural network (first complex layer with modReLU, followed by real layers and a sigmoid) to map raw IQ data to detection probabilities, functioning analogously to an autoencoder over the range axis. Evaluations on synthetically distorted LFM chirp data show that, with an enriched training set, the neural network can outperform MF+CA-CFAR across multiple metrics, though performance degrades for low-energy echoes and certain distortions, highlighting the need for diverse training data. The approach offers a scalable, potentially embedded-friendly alternative to traditional pulse compression, and future work aims to extend to waveform-scale processing and broader distortion robustness.

Abstract

Correctly detecting radar targets is usually challenged by clutter and waveform distortion. An additional difficulty stems from the relative proximity of several targets, the latter being perceived as a single target in the worst case, or influencing each other's detection thresholds. The negative impact of targets proximity notably depends on the range resolution defined by the radar parameters and the adaptive threshold adopted. This paper addresses the matter of targets detection in radar range profiles containing multiple targets with varying proximity and distorted echoes. Inspired by recent contributions in the radar and signal processing literature, this work proposes partially complex-valued neural networks as an adaptive range profile processing. Simulated datasets are generated and experiments are conducted to compare a common pulse compression approach with a simple neural network partially defined by complex-valued parameters. Whereas the pulse compression processes one pulse length at a time, the neural network put forward is a generative architecture going through the entire received signal in one go to generate a complete detection profile.

Detecting radar targets swarms in range profiles with a partially complex-valued neural network

TL;DR

The paper addresses radar target detection in range profiles under clutter, proximity effects, and waveform distortions by replacing conventional pulse compression and CA-CFAR with a partially complex-valued neural network that processes the full received signal to output a detection profile. The proposed RX-scale processing uses a hybrid complex-real neural network (first complex layer with modReLU, followed by real layers and a sigmoid) to map raw IQ data to detection probabilities, functioning analogously to an autoencoder over the range axis. Evaluations on synthetically distorted LFM chirp data show that, with an enriched training set, the neural network can outperform MF+CA-CFAR across multiple metrics, though performance degrades for low-energy echoes and certain distortions, highlighting the need for diverse training data. The approach offers a scalable, potentially embedded-friendly alternative to traditional pulse compression, and future work aims to extend to waveform-scale processing and broader distortion robustness.

Abstract

Correctly detecting radar targets is usually challenged by clutter and waveform distortion. An additional difficulty stems from the relative proximity of several targets, the latter being perceived as a single target in the worst case, or influencing each other's detection thresholds. The negative impact of targets proximity notably depends on the range resolution defined by the radar parameters and the adaptive threshold adopted. This paper addresses the matter of targets detection in radar range profiles containing multiple targets with varying proximity and distorted echoes. Inspired by recent contributions in the radar and signal processing literature, this work proposes partially complex-valued neural networks as an adaptive range profile processing. Simulated datasets are generated and experiments are conducted to compare a common pulse compression approach with a simple neural network partially defined by complex-valued parameters. Whereas the pulse compression processes one pulse length at a time, the neural network put forward is a generative architecture going through the entire received signal in one go to generate a complete detection profile.
Paper Structure (5 sections, 7 equations, 3 figures, 2 tables)

This paper contains 5 sections, 7 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Datasets generation
  3. Range profile processing
  4. Experiments
  5. Conclusion

Figures (3)

  • Figure 1: Matched filter range profile processing and RX-scale range profile processing: the MF implies applying the same filter sequentially over shifted slices of the range profile, whereas the proposed RX-scale range profile processing considers the entire range profile all at once using the neural network $\Phi$.
  • Figure 2: RX-scale range profile processing using a hybrid real-complex-valued autoencoder. The proposed RX-scale range profile processing considers the entire range profile all at once using the neural network $\Phi$, and produces a detection profile immediately comparable to an arbitrary threshold.
  • Figure 3: Example range profile processing for a test range profile. From top to bottom: the first figure is the matched filtered range profile with the corresponding CA-CFAR threshold; the second figure is the neural network processed range profile with the corresponding arbitrary threshold set to $0.5$.