Direction Of Arrival Estimation For Non-Coherent Sub-Arrays Via Joint Sparse And Low-Rank Signal Recovery

Tom Tirer, Oded BialerICASSP 2021 · OpenReview

TL;DR

This paper tackles the challenging problem of pinpointing the direction of multiple radio sources using an array of antennas that aren't perfectly synchronized. They propose a novel method that combines sparse signal recovery with low-rank matrix factorization, formulated as a convex optimization problem, to achieve more accurate direction-of-arrival (DOA) estimates, especially when dealing with imperfectly synchronized antenna subarrays.

What Problem Does This Paper Address?

Estimating the direction of arrival (DOA) of signals is crucial in many applications, from radar to telecommunications. A common approach for coherent antenna arrays (where all elements are perfectly synchronized) is sparse signal reconstruction, which identifies DOAs from peaks in a high-dimensional signal estimate. However, synchronizing a large number of antenna elements can be prohibitively expensive and complex.

This paper focuses on a more practical scenario: using an array composed of multiple non-coherent sub-arrays. In this setup, elements within each sub-array are synchronized, but there are unknown, varying phase shifts between the sub-arrays due to unsynchronized local oscillators. This lack of perfect coherence significantly complicates DOA estimation, especially when only a single snapshot of data is available. Existing methods for non-coherent sub-arrays often struggle with single snapshots or require prior knowledge of the number of sources.

Methods

The authors propose a two-pronged approach based on convex optimization and signal reconstruction.

  1. Joint Sparse and Low-Rank Matrix Recovery: They formulate the problem as reconstructing a matrix that is both "sparse" (in terms of source locations) and "low-rank" (related to the sub-array structure). This is achieved by defining a grid of possible DOAs and then formulating a convex optimization problem. This problem uses a combination of the L1-norm (for sparsity) and the nuclear norm (for low-rankness) as regularization terms, aiming to recover a matrix Z that captures the underlying signal structure.
  2. Phase-Shift Estimation and Coherent Refinement: While the DOAs can be estimated directly from the solution of the first convex problem, the authors found that a better approach is to first use the recovered matrix Z to estimate the unknown phase shifts between the sub-arrays. Once these phase shifts are estimated, they can be used to "correct" the observations, effectively creating a pseudo-coherent observation vector. This corrected data is then fed into a standard sparsity-based DOA estimation method (similar to those used for fully coherent arrays), leading to improved accuracy.

Datasets

Not discussed in the paper. The evaluation is performed using simulated data with controlled parameters (number of sources, angles, SNR, array configuration).

Key Insights and Ideas

  • The Power of Joint Sparsity and Low-Rankness: By formulating the problem as a joint sparse and low-rank matrix recovery, the method can effectively handle the complexities introduced by non-coherent sub-arrays, outperforming methods that only consider sparsity.
  • Phase Correction is Key: Estimating the unknown phase shifts between sub-arrays and then applying a coherent DOA estimation technique yields significantly better results than directly estimating DOAs from the initial non-coherent solution. This highlights the importance of addressing the phase discrepancies explicitly.
  • Convex Optimization for Robustness: The use of convex relaxation makes the problem tractable and robust, meaning it doesn't rely on good initializations, which is a common pitfall for non-convex optimization methods. This also allows for handling a potentially large or unknown number of sources.

Key Results

The proposed method, particularly the second strategy (Proposed2) involving phase correction, consistently outperformed existing sparsity-based and MUSIC-based methods in terms of Root Mean Square Error (RMSE) across various Signal-to-Noise Ratios (SNRs).

  • In scenarios with two sources, Proposed2 showed a significant RMSE reduction compared to SparsityOnly and MUSIC, especially at higher SNRs.
  • With an increased number of sources (four sources), the advantage of the proposed methods became even more pronounced, with Proposed2 again achieving the best performance.
  • The paper also shows visualizations of the "spectrums" (signal strength estimates across possible angles) for different methods, illustrating how the proposed methods produce sharper and more accurate peaks corresponding to the true source locations.

Limitations and Future Work

  • Computational Cost: While the convex optimization is tractable, it still involves solving a relatively large optimization problem. The paper mentions that it took "only a few seconds" on a specific laptop, but for real-time applications with very large arrays or high data rates, further optimization might be needed.
  • Assumption of Narrowband Signals: The formulation assumes narrowband signals, which might not hold for all applications.
  • Future Work: The authors suggest that the improved performance with more sources might be further exploited. Additionally, the robustness of the phase estimation step could be investigated further. The paper also briefly touches upon the benefit of combining nuclear and L1 norms for estimating a limited set of parameters, hinting at potential for other related problems.

Summary updated

Abstract

Estimating the directions of arrival (DOAs) of multiple sources from a single snapshot obtained by a coherent antenna array is a well-known problem, which can be addressed by sparse signal reconstruction methods, where the DOAs are estimated from the peaks of the recovered high-dimensional signal. In this paper, we consider a more challenging DOA estimation task where the array is composed of non-coherent sub-arrays (i.e., sub-arrays that observe different unknown phase shifts due to using low-cost unsynchronized local oscillators). We formulate this problem as the reconstruction of a joint sparse and low-rank matrix and solve its convex relaxation. While the DOAs can be estimated from the solution of the convex problem, we further show how an improvement is obtained if, instead, one estimates from this solution only the phase shifts, creates "phase-corrected" observations and applies another final (plain, coherent) sparsity-based DOA estimation. Numerical experiments show that the proposed approach outperforms strategies that are based on non-coherent processing of the sub-arrays as well as other sparsity-based methods.
BibTeX
@inproceedings{tirer2021direction,
  author = {Tom Tirer and Oded Bialer},
  title = {{Direction Of Arrival Estimation For Non-Coherent Sub-Arrays Via Joint Sparse And Low-Rank Signal Recovery}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2021},
  url = {https://openreview.net/forum?id=bh6WAwRTGH},
}
Figure