RASPNet: A Benchmark Dataset for Radar Adaptive Signal Processing Applications

TL;DR

RASPNet addresses a critical gap in adaptive radar research by delivering a large-scale, realistic benchmark with $M=100$ radar scenarios and $K=10{,}000$ clutter realizations per scenario, totaling over $16\,\text{TB}$. Built with RFView, it supports traditional RASP methods and data-driven CVNN approaches, and is validated against measured MCARM data to some extent. The paper introduces the dataset's construction, organization via an $\mathcal{E}$-statistic-based difficulty ordering, and several applications: target localization benchmarking (including the Adaptive Radar Transformer, ART), CVNN benchmarking (including Convolutional Steinmetz Networks), and transfer learning. It also demonstrates that RFView can approximate real-world clutter sufficiently for initial model development, while acknowledging limitations and potential future enhancements. RASPNet is positioned to accelerate research by providing a reproducible testbed that bridges the gap between simulations and real-world adaptive radar performance.

Abstract

We present a large-scale dataset called RASPNet for radar adaptive signal processing (RASP) applications to support the development of data-driven models within the adaptive radar community. RASPNet exceeds 16 TB in size and comprises 100 realistic scenarios compiled over a variety of topographies and land types across the contiguous United States. For each scenario, RASPNet comprises 10,000 clutter realizations from an airborne radar setting, which can be used to benchmark radar and complex-valued learning algorithms. RASPNet intends to fill a prominent gap in the availability of a large-scale, realistic dataset that standardizes the evaluation of RASP techniques and complex-valued neural networks. We outline its construction, organization, and several applications, including a transfer learning example to demonstrate how RASPNet can be used for real-world adaptive radar scenarios.

Figures (9)

• Figure 1: Mapping the $M = 100$ scenarios within RASPNet. Each scenario is indexed by $i \in S$.
• Figure 2: Range-Doppler plots for (a) Flight #5 Acquisition 575 from the MCARM measured dataset, (b) RFView® simulated data, and (c) classical bald-earth model without terrain heights. The colormap depicts the magnitude squared of the radar range-Doppler spectrum (red is higher and blue is lower).
• Figure 3: Categorizing RASPNet scenarios based on difficulty. Each scenario is indexed by $q_i \in Q$.
• Figure 4: Test performance comparison on RASPNet scenarios $i \in \tilde{S}$ using convolutional CVNN, convolutional RVNN, and convolutional Steinmetz and analytic networks [ours]. The x-axis represents the training epochs, while the y-axis indicates the target localization performance.
• Figure 5: Regression CNN target localization performance for scenarios ${i \in \tilde{S} \setminus \{29\}}$ after transfer learning $([Err_{\text{CNN}}]_{\text{TL}})$ and fine-tuning $([Err_{\text{CNN}}]_{\text{FT}})$.

Theorems & Definitions (1)

• Definition 2.1