Data-Driven Target Localization Using Adaptive Radar Processing and Convolutional Neural Networks

TL;DR

This work tackles target localization after adaptive radar detection in regimes where NAMF performance degrades near the breakdown threshold. It proposes a data-driven regression CNN that consumes NAMF heatmap tensors—generated via high-fidelity RFView simulations—to estimate target position (and velocity with Doppler). Empirical results show substantial localization and azimuth-velocity gains over traditional peak-based and local-search methods, including near the NAMF breakdown region; robustness to mismatches is addressed via coordinated transformation and few-shot learning. The findings highlight the feasibility of data-driven post-processing for improved radar localization and lay groundwork for robust, adaptable targets localization in complex clutter environments.

Abstract

Leveraging the advanced functionalities of modern radio frequency (RF) modeling and simulation tools, specifically designed for adaptive radar processing applications, this paper presents a data-driven approach to improve accuracy in radar target localization post adaptive radar detection. To this end, we generate a large number of radar returns by randomly placing targets of variable strengths in a predefined area, using RFView, a high-fidelity, site-specific, RF modeling & simulation tool. We produce heatmap tensors from the radar returns, in range, azimuth [and Doppler], of the normalized adaptive matched filter (NAMF) test statistic. We then train a regression convolutional neural network (CNN) to estimate target locations from these heatmap tensors, and we compare the target localization accuracy of this approach with that of peak-finding and local search methods. This empirical study shows that our regression CNN achieves a considerable improvement in target location estimation accuracy. The regression CNN offers significant gains and reasonable accuracy even at signal-to-clutter-plus-noise ratio (SCNR) regimes that are close to the breakdown threshold SCNR of the NAMF. We also study the robustness of our trained CNN to mismatches in the radar data, where the CNN is tested on heatmap tensors collected from areas that it was not trained on. We show that our CNN can be made robust to mismatches in the radar data through few-shot learning, using a relatively small number of new training samples.

Figures (13)

• Figure 1: The map of the matched case RFView® example scenario. The blue triangle is the platform location and the red region is the range-azimuth area for radar processing. The elevation heatmap overlaying the left image depicts the simulation region.
• Figure 2: The map of the mismatched case RFView® example scenario. Each of the displaced platform locations (1 km North, 1 km South, 1 km East, and 1 km West) are shown in green, and their relative range-azimuth areas for radar processing are depicted in orange. The original platform location (blue) and range-azimuth area (red) are recycled from the matched case.
• Figure 3: (Left) Size $5\times26$ heatmap tensor example from the matched case RFView® example scenario (see Section \ref{['Sec3.2']}). Each size $1 \times 26$ 'array in azimuth' pertains to one of the $\kappa = 5$ unique range bins. (Right) Size $5 \times 26 \times 31$ heatmap tensor example from the Doppler processing RFView® example scenario (see Section \ref{['Sec3.4']}). Each size $26 \times 31$ heatmap image has an azimuth and velocity dimension, and pertains to one of the $\kappa = 5$ unique range bins.
• Figure 4: Baseline regression CNN architecture for azimuth step size $\Delta \theta = 0.4^{\circ}$ and depth parameter $\kappa$.
• Figure 5: Doppler CNN architecture for default grid step size $(\Delta r, \Delta \theta, \Delta v) = (30 \ \text{m}, 0.4^{\circ}, 0.5 \ \text{m/s})$.