Abstract
Direction of arrival (DOA) estimation is a classical problem in signal processing with many practical applications. Its research has recently been advanced owing to the development of methods based on sparse signal reconstruction. While these methods have shown advantages over conventional ones, there are still difficulties in practical situations where true DOAs are not on the discretized sampling grid. To deal with such an off-grid DOA estimation problem, this paper studies an off-grid model that takes into account effects of the off-grid DOAs and has a smaller modeling error. An iterative algorithm is developed based on the off-grid model from a Bayesian perspective while joint sparsity among different snapshots is exploited by assuming a Laplace prior for signals at all snapshots. The new approach applies to both single snapshot and multi-snapshot cases. Numerical simulations show that the proposed algorithm has improved accuracy in terms of mean squared estimation error. The algorithm can maintain high estimation accuracy even under a very coarse sampling grid.
Introduction
In array signal processing, Direction of Arrival (DOA) estimation involves determining the spatial angles from which signals impinge upon a sensor array. Modern approaches have increasingly framed this as a "sparse recovery" problem, leveraging compressed sensing to identify signal sources across a discretized horizon. However, a fundamental limitation persists: physical signals rarely align perfectly with a fixed grid. When a signal falls between grid points, "basis mismatch" occurs, leading to significant estimation errors.
Published in 2012, this paper addresses this "rounding error" by introducing Off-grid Sparse Bayesian Inference (OGSBI). By combining a linearized Taylor series expansion with a robust Bayesian framework, the authors provide a method to accurately locate signal sources even when they fall "between the cracks" of a sampling grid.
The Off-Grid Model
Traditional on-grid methods suffer from a trade-off: a coarse grid leads to high modeling error, while an extremely dense grid increases the correlation between columns in the measurement matrix, violating the conditions required for stable sparse recovery and increasing computational load.
To resolve this, the authors propose a Linearized Off-grid Model. Instead of treating the grid as static, they use a first-order Taylor expansion to approximate the signal's true location. They define an offset parameter, $\beta$, representing the distance from the nearest grid point. This effectively transforms the static grid into a flexible one where each point can "nudge" itself closer to the true physical signal, significantly reducing modeling error without requiring a prohibitively dense grid.
Sparse Bayesian Inference Framework
The OGSBI algorithm treats the signal and the grid offsets as hidden variables within a probabilistic model, solved via an iterative Expectation-Maximization (EM) approach. Two key components define this framework:
- Laplace Prior: The authors assume a Laplace prior for the signals. In Bayesian terms, this encourages "sparsity," forcing the model to explain the data using the fewest possible signal sources.
- Uniform Prior for Offsets: A critical innovation in this paper is the use of a uniform prior for the grid offsets. Previous attempts often assumed Gaussian distributions, which do not accurately reflect the physical reality that a signal is equally likely to be anywhere between two discrete grid points. The uniform prior, combined with the iterative refinement of $\beta$, leads to more robust convergence.
To enhance practical utility, the authors also introduced OGSBI-SVD. This variant utilizes Singular Value Decomposition to reduce the dimensionality of the data, which both speeds up computation and improves the signal-to-noise ratio (SNR) by filtering out noise-only subspaces.
Experimental Results
The methodology was validated through numerical simulations using a Uniform Linear Array (ULA) with 8 sensors. The performance was measured using Mean Squared Error (MSE) across 200 trials.
- Superior Accuracy: OGSBI-SVD consistently outperformed standard benchmarks like $\ell_1$-SVD. In many scenarios, the proposed method achieved an MSE 10-15 dB lower than traditional compressed sensing techniques.
- Grid Robustness: Perhaps the most significant finding is OGSBI's performance on coarse grids. While on-grid methods fail as the grid interval increases (e.g., to 4°), OGSBI maintains high accuracy. This allows practitioners to use much coarser grids, significantly reducing the computational burden without sacrificing precision.
- Breaking the Lower Bound: Because OGSBI is not restricted to discrete points, it effectively "breaks" the theoretical accuracy floor that limits on-grid methods to the nearest grid increment.
Limitations and Conclusion
Despite its advantages, OGSBI is not without limitations. The SVD-based variant is sensitive to measurement outliers; even a 0.2% corruption in data can degrade performance. Additionally, the requirement for matrix inversions at each iteration makes the algorithm computationally intensive for very large-scale sensor arrays.
In conclusion, OGSBI represents a significant advancement in DOA estimation. By accounting for the continuous nature of the physical world within a sparse Bayesian framework, it provides a highly accurate and grid-robust solution for modern signal processing challenges.
Summary updated