End-to-end DOA estimation via self-supervised cascaded DNNs with array errors mitigation
Shuang Wu, Ye Yuan, Hongyu Pu, Tai Fu · 2026
Read End-to-end DOA estimation via self-supervised cascaded DNNs with array errors mitigation by Shuang Wu, Ye Yuan, Hongyu Pu, Tai Fu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13482609.txt.
Description
Wu and colleagues build a two-stage neural system in 2026 for direction-of-arrival estimation, the radar and sonar task of pinpointing where a signal comes from. It works without first modeling the antenna flaws that usually wreck such systems. Real arrays suffer shifted elements, coupling between neighbors, and mismatched gains, and labeled training data for every flaw is impractical. Their cascade cleans the signal with adaptive filters, then feeds a self-taught network imitating the classic MUSIC spectrum method.
The filter stage uses four layers tuned on signal statistics, while the spectrum stage stacks eight layers trained without human-provided direction labels. Because the stages train in parallel and correct errors inside the pipeline, no explicit model of the array's flaws is ever needed. Across unseen test scenarios it beat current methods, holding up especially when signals were faint or snapshots few.
Catalog record
| CREATOR | Shuang Wu, Ye Yuan, Hongyu Pu, Tai Fu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 95 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |