Deep Learning Denoising Based Line Spectral Estimation

Yuan Jiang, Hongbin Li, Muralidhar Rangaswamy

Research output: Contribution to journalArticlepeer-review

42 Scopus citations

Abstract

Many well-known line spectral estimators may experience significant performance loss with noisy measurements. To address the problem, we propose a deep learning denoising based approach for line spectral estimation. The proposed approach utilizes a residual learning assisted denoising convolutional neural network (DnCNN) trained to recover the unstructured noise component, which is used to denoise the original measurements. Following the denoising step, we employ a popular model order selection method and a subspace line spectral estimator to the denoised measurements for line spectral estimation. Numerical results show that the proposed approach outperforms a recently introduced atomic norm minimization based denoising method and offers a substantial improvement compared with the line spectral estimation results obtained by directly applying the subspace estimator without denoising.

Original languageEnglish
Article number8822737
Pages (from-to)1573-1577
Number of pages5
JournalIEEE Signal Processing Letters
Volume26
Issue number11
DOIs
StatePublished - Nov 2019

Keywords

  • deep learning
  • line spectral estimation
  • signal denoising

Fingerprint

Dive into the research topics of 'Deep Learning Denoising Based Line Spectral Estimation'. Together they form a unique fingerprint.

Cite this