Broadband Off-Grid DOA Estimation Using Block Sparse Bayesian Learning for Nonuniform Noise Variance

Youngmin Choo, Haesang Yang

Research output: Contribution to journalArticlepeer-review

1 Scopus citations


In this article, an extended block sparse Bayesian learning (SBL) is applied to a linear system for estimating the direction of arrival (DOA). DOA estimation using conventional SBL deteriorates when inconsistent noise occurs (e.g., some sensors in the array are malfunctioning). The SBL is extended to accommodate the nonuniform noise variance and shows superior robustness in DOA estimation owing to its selective usage of sensors under normal conditions. Furthermore, a block SBL framework is adopted to exploit the commonality of DOAs over broadband signal frequency components and their correlation at the array. The transformation matrix in the linear system is composed of replicas for preset arrival angles as its basis. Practically, the DOAs deviate from the on-grid angles in the matrix, which results in degradation by basis mismatch. An off-grid SBL incorporating an inconsistent noise variance and block sparsity is formulated to estimate the DOAs reliably. The extended SBL is applied to simulated and measured data (SWellEx-96), and its results are compared with those from conventional approaches. It exhibits better performance in terms of resolution and denoising. In particular, when a part of the array is malfunctioning, the extended SBL can still estimate the DOAs clearly, whereas other approaches fail.

Original languageEnglish
Pages (from-to)1-17
Number of pages17
JournalIEEE Journal of Oceanic Engineering
StateAccepted/In press - 2022


  • Basis mismatch
  • block correlation
  • Broadband communication
  • Direction-of-arrival estimation
  • Estimation
  • Frequency measurement
  • Linear systems
  • off-grid direction-of-arrival (DOA) estimation
  • Sea measurements
  • Sensors
  • sparse Bayesian learning (SBL)


Dive into the research topics of 'Broadband Off-Grid DOA Estimation Using Block Sparse Bayesian Learning for Nonuniform Noise Variance'. Together they form a unique fingerprint.

Cite this