Direction of Arrival Estimation and Phase-Correction for Noncoherent Subarrays: A Convex Optimization Approach

Tom Tirer, Oded BialerIEEE Trans. Aerosp. Electron. Syst 2022 · OpenReview

TL;DR

This paper introduces a novel convex optimization approach for estimating the direction of arrival (DOA) of multiple sources using antenna arrays composed of non-coherent sub-arrays. The method effectively handles unknown and varying phase shifts between sub-arrays, offering improved accuracy and robustness compared to existing techniques, particularly in challenging scenarios.

What Problem Does This Paper Address?

Estimating the direction of arrival (DOA) of signals is crucial for applications like radar, localization, and communication. A common approach uses antenna arrays, where more elements generally lead to better accuracy. However, synchronizing all elements of a large array (making them "coherent") is expensive and complex. An alternative is to use multiple "non-coherent" sub-arrays, where elements within each sub-array are synchronized, but there are unknown phase shifts between sub-arrays that can even vary from one snapshot (observation) to the next. This variability makes traditional DOA estimation methods difficult to apply. Existing methods either ignore these phase shifts (leading to performance loss) or require complex, non-convex optimization that is sensitive to initialization. This paper tackles the challenge of accurately estimating DOAs from such non-coherent sub-arrays, especially when phase shifts are unknown and change over time.

Methods

The core of the proposed solution lies in a two-stage convex optimization approach.

  1. Joint Sparse and Low-Rank Matrix Reconstruction: The problem is first formulated as reconstructing two matrices: a sparse matrix representing the signal sources and a low-rank matrix representing the array's structure. This formulation is then relaxed into a convex optimization problem. To efficiently solve this for multiple snapshots, the authors combine the Alternating Direction Method of Multipliers (ADMM) and the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA). This custom optimization scheme is significantly faster than general-purpose solvers for this type of problem.
  2. Phase-Correction Strategy: Instead of directly estimating DOAs from the first stage's solution, the authors propose a more refined approach. They use a computationally lighter convex relaxation to estimate the unknown sub-array phase shifts. These estimated phase shifts are then used to "correct" the observed data, effectively making the sub-arrays appear coherent. A standard DOA estimation method (like MUSIC) can then be applied to this phase-corrected data, leading to improved performance.

Datasets

Not discussed in the paper. The paper focuses on simulated data generated according to the described model.

Key Insights and Ideas

  • Leveraging Non-Coherence: The paper demonstrates that instead of fighting the non-coherence of sub-arrays, it can be effectively modeled and exploited. The unknown phase shifts, which are a nuisance in traditional methods, become a key piece of information to estimate.
  • Two-Stage Approach for Superior Performance: The most significant insight is that directly estimating DOAs from the initial convex relaxation is good, but estimating the phase shifts first and then performing DOA estimation on "phase-corrected" data yields even better results. This highlights the importance of accurately handling the phase information.
  • Efficient Convex Optimization for Scalability: The development of a specialized ADMM+FISTA algorithm is crucial. It allows the proposed methods to scale much better with the number of snapshots, making them practical for real-world applications where multiple observations are available.
  • Practical Tightness of SDP Relaxation: For phase shift estimation, the paper uses a Semidefinite Programming (SDP) relaxation. Empirically, they observe that this relaxation is "practically tight," meaning it yields the true optimal solution to the original non-convex problem, which is a strong theoretical and practical result.

Key Results

  • Improved DOA Estimation Accuracy: Both proposed methods (Proposed1: direct DOA estimation, Proposed2: phase-correction) significantly outperform non-coherent MUSIC and other sparsity/low-rank based methods.
  • Proposed2 Outperforms Proposed1: The phase-correction strategy (Proposed2) consistently achieves better DOA estimation accuracy than the direct estimation method (Proposed1), especially at higher Signal-to-Noise Ratios (SNRs).
  • Competitive with State-of-the-Art: Proposed2 demonstrates performance competitive with, and often superior to, methods like Joint-AM, which is a state-of-the-art non-convex approach.
  • Significant Speedup: The custom ADMM+FISTA optimization scheme for the initial convex problem is orders of magnitude faster than general-purpose solvers (like CVX) for multiple snapshots. Algorithm 2 for phase shift estimation also provides a substantial speedup over CVX.
  • Robustness to Source Number: The methods are designed to handle a large, potentially unknown, number of sources without requiring good initializations, unlike some existing non-convex methods.
  • Phase Shift Estimation Quality: The RMSE of the estimated phase shifts is shown to be relatively stable across different scenarios and improves monotonically with SNR, validating the effectiveness of the phase estimation stage.

Limitations and Future Work

  • Computational Complexity for Multiple Snapshots: While the proposed ADMM+FISTA scheme is much faster than general solvers, it still scales with the number of snapshots. For extremely large numbers of snapshots, further optimizations might be needed.
  • Hyper-parameter Tuning: The methods involve several hyper-parameters (e.g., $\beta, \mu, \lambda, C$) that need to be tuned. While the paper provides some guidance and experimental settings, optimal tuning can still be a challenge.
  • Future Work: The paper implicitly suggests that the developed optimization framework and the phase-correction strategy can be extended to other related problems in array signal processing. The authors also mention that their approach can be used to initialize non-convex optimization methods, potentially improving their performance. Further exploration of the theoretical guarantees for the SDP relaxation's tightness in this specific context could also be a future direction.

Summary updated

Abstract

Estimating the direction of arrival (DOA) of sources is an important problem in aerospace and vehicular communication, localization, and radar. In this article, we consider a challenging multisource DOA estimation task, where the receiving antenna array is composed of noncoherent subarrays, i.e., subarrays that observe different unknown phase shifts at every snapshot (e.g., due to waiving the demanding synchronization of local oscillators across the entire array). We formulate this problem as the reconstruction of joint sparse and low-rank matrices and solve the problem’s convex relaxation. To scale the optimization complexity with the number of snapshots better than general-purpose solvers, we design an optimization scheme, based on integrating the alternating direction method of multipliers and the accelerated proximal gradient techniques, which exploits the structure of the problem. While the DOAs can be estimated from the solution of the aforementioned convex problem, we further show how an improvement is obtained if, instead, one estimates from this solution only the subarrays’ phase shifts. This is done using another, computationally light, convex relaxation that is practically tight. Using the estimated phase shifts, “phase-corrected” observations are created and a final plain (“coherent”) DOA estimation method can be applied. Numerical experiments show the performance advantages of the proposed strategies over existing methods.
BibTeX
@article{tirer2022direction,
  author = {Tom Tirer and Oded Bialer},
  title = {{Direction of Arrival Estimation and Phase-Correction for Noncoherent Subarrays: A Convex Optimization Approach}},
  journal = {IEEE Trans. Aerosp. Electron. Syst.},
  year = {2022},
  url = {https://openreview.net/forum?id=Ijs-kpbSZgB},
}
Figure