1: \begin{abstract}

2: An active-sensing-based learning algorithm is proposed to solve the near-field beam alignment problem with the aid of wavenumber-domain transform matrices (WTMs).

3: Specifically, WTMs can transform the antenna-domain channel into a sparse representation in the wavenumber domain.

4: The dimensions of WTMs can be further reduced by exploiting the dominance of line-of-sight (LoS) links.

5: By employing these lower-dimensional WTMs as mapping functions, the active-sensing-based algorithm is executed in the wavenumber domain, resulting in an acceleration of convergence.

6: Compared with the codebook-based beam alignment methods, the proposed method finds the optimal beam pair in a ping-pong fashion, thus avoiding high training overheads caused by beam sweeping.

7: Finally, the numerical results validate the effectiveness of the proposed method.    

8: \end{abstract}

9: