TransDOA: Calibrating Array Imperfections via Transformer-based Transfer Learning

TL;DR

This work tackles robust DOA estimation in the presence of hardware imperfections and adverse signal conditions by introducing TransDOA, a Vision Transformer–based estimator that operates on covariance information. A Covariance Matrix Embedding strategy converts SCMs into Transformer inputs, followed by a Transformer Encoder and a DOA head that supports both 1D and 2D DOA outputs, with a permutation-invariant training loss. To close the gap between ideal simulations and real-world imperfections, the authors propose a supervised transfer learning framework that aligns source (ideal) and target (imperfect) features using paired data and a composite $\mathcal{L}_{\text{total}} = \alpha \mathcal{L}_{\cos} + \beta \mathcal{L}_{\text{mse}}$ objective. Extensive simulations across four scenarios demonstrate that TransDOA achieves superior robustness and accuracy under low SNR, limited snapshots, and multiple array imperfections, and that the transfer-learning approach substantially enhances performance with limited target-domain data, including 2D DOA tasks. This approach provides a practical pathway to deploy advanced DL-based DOA estimators in real-world sensing systems with hardware nonidealities.

Abstract

In practical scenarios, processes such as sensor design, manufacturing, and installation will introduce certain errors. Furthermore, mutual interference occurs when the sensors receive signals. These defects in array systems are referred to as array imperfections, which can significantly degrade the performance of Direction of Arrival (DOA) estimation. In this study, we propose a deep-learning based transfer learning approach, which effectively mitigates the degradation of deep-learning based DOA estimation performance caused by array imperfections. In the proposed approach, we highlight three major contributions. First, we propose a Vision Transformer (ViT) based method for DOA estimation, which achieves excellent performance in scenarios with low signal-to-noise ratios (SNR) and limited snapshots. Second, we introduce a transfer learning framework that extends deep learning models from ideal simulation scenarios to complex real-world scenarios with array imperfections. By leveraging prior knowledge from ideal simulation data, the proposed transfer learning framework significantly improves deep learning-based DOA estimation performance in the presence of array imperfections, without the need for extensive real-world data. Finally, we incorporate visualization and evaluation metrics to assess the performance of DOA estimation algorithms, which allow for a more thorough evaluation of algorithms and further validate the proposed method. Our code can be accessed at https://github.com/zzb-nice/DOA_est_Master.

Figures (23)

• Figure 1: Schematic illustration of environmental perception using front-mounted mmWave radar in a night-time autonomous driving scenario. The radar point clouds are projected onto the camera image for visualization.
• Figure 2: The framework of TransDOA and transfer learning approach proposed for array imperfections.
• Figure 3: The overall structure of TansDOA and the processing procedure of input covariance matrix.
• Figure 4: The framework of transfer learning approach proposed for array imperfections.
• Figure 5: DOA estimation results under Scenarios 1 with $SNR=-5db$ and $snap=10$. Three incident signals are generate and we focus on the first one. The points represent the estimated DOA while the circles denote the true DOAs.