CoRe-Net: Co-Operational Regressor Network with Progressive Transfer Learning for Blind Radar Signal Restoration

TL;DR

CoRe-Net introduces a cooperative regression framework for blind radar signal restoration, replacing adversarial training with a pair of interacting regressors: an Apprentice Regressor (AR) that restores degraded radar signals and a Master Regressor (MR) that provides task-specific quality feedback. The AR is trained to minimize a multi-term restoration loss, guided by MR’s predicted normalized PSNR, while MR learns to accurately estimate PSNR for both clean and restored signals. Progressive Transfer Learning (PTL) enables multi-pass refinements by cascading CoRe-Nets and feeding outputs forward with inherited parameters, yielding substantial SNR gains on the BRSR benchmark. Across experiments, CoRe-Net achieves state-of-the-art restoration, surpassing BRSR-OpGAN by over 1 dB in single pass and more than 3 dB with PTL, while maintaining a compact, real-time capable 1D Self-ONN architecture suitable for resource-constrained platforms.

Abstract

Real-world radar signals are frequently corrupted by various artifacts, including sensor noise, echoes, interference, and intentional jamming, differing in type, severity, and duration. This pilot study introduces a novel model, called Co-Operational Regressor Network (CoRe-Net) for blind radar signal restoration, designed to address such limitations and drawbacks. CoRe-Net replaces adversarial training with a novel cooperative learning strategy, leveraging the complementary roles of its Apprentice Regressor (AR) and Master Regressor (MR). The AR restores radar signals corrupted by various artifacts, while the MR evaluates the quality of the restoration and provides immediate and task-specific feedback, ensuring stable and efficient learning. The AR, therefore, has the advantage of both self-learning and assistive learning by the MR. The proposed model has been extensively evaluated over the benchmark Blind Radar Signal Restoration (BRSR) dataset, which simulates diverse real-world artifact scenarios. Under the fair experimental setup, this study shows that the CoRe-Net surpasses the Op-GANs over a 1 dB mean SNR improvement. To further boost the performance gain, this study proposes multi-pass restoration by cascaded CoRe-Nets trained with a novel paradigm called Progressive Transfer Learning (PTL), which enables iterative refinement, thus achieving an additional 2 dB mean SNR enhancement. Multi-pass CoRe-Net training by PTL consistently yields incremental performance improvements through successive restoration passes whilst highlighting CoRe-Net ability to handle such a complex and varying blend of artifacts.

Figures (6)

• Figure 1: Sample signals from the BRSR dataset demonstrating the impact of various artifacts and the effectiveness of the proposed Core-Net approach. In each example, the clean signal (blue) is shown in the top plot, whereas the corrupted (red) and restored (green) signals are overlaid in the bottom plot. The middle area in the bottom plot is highlighted to emphasize the restored signal. Artifact combinations include AWGN + Interference (a), Echo (b), Interference (c), and a blend of all artifacts (d), with annotated SNR improvements showcasing the restoration performance.
• Figure 2: Overall methodology illustrating the CoRe-Net framework, where the AR restores the distorted radar signal ($\hat{s}$), and the MR predicts the PSNR for both restored and clean signals.
• Figure 3: Progressive Transfer Learning (PTL) strategy for CoRe-Net. Each training pass ($k$) starts with a distorted signal ($r_k$) and produces a restored signal ($\hat{s}_k$), which becomes the input for the next pass ($r_{k+1}$). Parameters ($\theta_k$) are initialized using Xavier uniform initialization for $k=0$ and updated iteratively using the best-performing weights from the previous pass based on the validation SNR.
• Figure 4: Architecture of CoRe-Net showcasing the Apprentice Regressor (AR) and the Master Regressor (MR). The AR, on the left, is divided into encoder and decoder parts. The MR is shown in the center. Both components share building blocks such as ResDownBlock, ResUpBlock, and the Regression Block (right). $N$ represents the number of filters per block, while $d$ indicates the use of dropout. Skip connections are depicted with dotted lines, and feature dimensions are labeled at each stage. The Regression Block in the MR predicts PSNR (scaled between 0 and 1). The color coding of AR and MR matches that of the methodology figure for consistency.
• Figure 5: Restoration performance of CoRe-Net across training passes and modulation types. Mean SNR across four restoration passes with PTL (a), demonstrating significant gains in restoration quality. Mean SNR for radar signal modulation type (b), comparing the performance of CoRe-Net across multiple passes and BRSR-OpGAN.