A continuous DOA estimation method based on complex transformer and mask reconstruction

By constructing a self-supervised learning complex Transformer network based on complex Transformer and mask reconstruction, the problem of poor robustness of traditional DOA estimation methods under low signal-to-noise ratio is solved, and high-precision and robust continuous DOA estimation is achieved, improving the estimation performance in complex environments.

CN121955867BActive Publication Date: 2026-06-09SHANGHAI UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SHANGHAI UNIV
Filing Date
2026-04-03
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Traditional DOA estimation methods are not robust in low signal-to-noise ratio or insufficient snapshot scenarios, while deep learning methods have insufficient generalization ability in complex environments, making it difficult to achieve high-precision and robust DOA estimation.

Method used

We adopt a method based on complex Transformer and mask reconstruction. By constructing a complex Transformer network, we perform self-supervised learning and fine-tuning training. We utilize complex multi-head attention layers, multi-scale complex convolutional layers and complex feedforward network layers to achieve high-precision and robust continuous DOA estimation.

Benefits of technology

It achieves high-precision DOA estimation without mesh under low signal-to-noise ratio, improves the robustness and real-time performance of the estimation, eliminates discrete mesh mismatch error, and maintains the excellent direction-finding performance of the model in complex environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121955867B_ABST
    Figure CN121955867B_ABST
Patent Text Reader

Abstract

The application discloses a continuous DOA estimation method based on complex number Transformer and mask reconstruction and belongs to the technical field of array signal processing. The method comprises the following steps: constructing a data set, constructing a complex number Transformer network structure, pre-training the complex number Transformer network structure by adopting a self-supervised learning strategy, fine-tuning training the pre-trained complex number Transformer network structure based on the data set, obtaining a final complex number Transformer network structure, and running the final complex number Transformer network structure on test data with different signal-to-noise ratios, using RMSE as a performance evaluation index, and outputting a continuous DOA estimation result. By means of complex number Transformer and mask self-supervised learning, the application realizes high-precision and gridless continuous DOA estimation under low signal-to-noise ratio, and significantly improves estimation robustness and real-time performance.
Need to check novelty before this filing date? Find Prior Art