Joint DoA-Range Estimation Using Space-Frequency Virtual Difference Coarray

Zihuan Mao, Shengheng Liu, Yimin D. Zhang, Leixin Han, Yongming HuangIEEE TSP 2022 · OpenReview

TL;DR

This paper introduces a novel method for simultaneously estimating the direction and range of targets using a Frequency Diverse Coprime Array (FDCA). By leveraging a "space-frequency virtual difference coarray" and advanced signal processing techniques like Decoupled Atomic Norm Minimization (DANM) and Cyclic Rank Minimization (CRM), the authors significantly increase the number of detectable targets and improve estimation accuracy, especially in challenging scenarios.

What Problem Does This Paper Address?

Accurate localization of targets in both direction-of-arrival (DoA) and range is crucial for many applications, including radar, sonar, and wireless communications. Traditional methods often struggle to achieve high resolution and detect a large number of targets simultaneously, especially with limited sensor hardware. Frequency Diverse Arrays (FDAs) offer a way to achieve joint DoA-range estimation by exploiting frequency increments across array elements. However, the resulting "space-frequency virtual difference coarray" in FDCA systems suffers from missing elements ("holes"), which complicates signal reconstruction and limits the number of degrees-of-freedom (DoFs). Existing methods like CMT-BCS are grid-based and can be computationally expensive or suffer from performance loss due to basis mismatch. This paper aims to overcome these limitations by proposing a gridless, low-complexity interpolation framework that effectively fills these "holes" and enhances DoF.

Methods

The core of the proposed solution lies in effectively reconstructing the "space-frequency virtual difference coarray" derived from the FDCA.

  • Virtual Coarray Derivation: The paper first derives the virtual coarray signal from the FDCA's covariance matrix. This virtual coarray, while offering more DoFs, contains "holes" due to the coprime structure.
  • Decoupled Atomic Norm Minimization (DANM): To address the computational burden of reconstructing the doubly-Toeplitz covariance matrix and to interpolate the "holes," the authors propose DANM. This method converts the coarray signal into a matrix form, decoupling spatial and frequency information, and then uses atomic norm minimization for reconstruction.
  • Cyclic Rank Minimization (CRM): To avoid approximation errors introduced by convex relaxation in DANM, the paper reformulates the problem as a rank-minimization problem. This is then solved using a cyclic rank minimization approach, which is shown to be more accurate and closer to the Cramér-Rao bound.
  • ADMM-based Solver: For further efficiency, an Alternating Direction Method of Multipliers (ADMM) based solver is developed, providing closed-form solutions for the optimization problems.
  • Subspace-based Spectral Estimation: After reconstructing the interpolated coarray signal, standard subspace-based spectral estimation techniques like 2D MUSIC are applied for joint DoA-range estimation.

Datasets

The paper does not introduce new datasets. Instead, it relies on simulated data generated according to the described FDCA signal model for evaluation. The simulations involve varying numbers of targets, signal-to-noise ratios (SNRs), and snapshots to test the performance of the proposed methods.

Key Insights and Ideas

  • Space-Frequency Virtual Difference Coarray: The concept of creating a unified "space-frequency virtual difference coarray" from FDCA signals is key to increasing DoFs.
  • Interpolation for "Holes": Effectively filling the "holes" in the virtual coarray is crucial for unlocking its full potential. DANM and CRM are presented as sophisticated ways to achieve this.
  • Gridless Estimation: The proposed methods are gridless, meaning they don't require pre-defined grids for target locations, which is a significant advantage over methods like CMT-BCS and can lead to higher accuracy.
  • CRM's Superiority: Reformulating the problem to directly minimize rank (CRM) instead of using convex relaxation (DANM) leads to more accurate reconstruction and better performance, especially in challenging scenarios.
  • Trade-off Management: The CRM method allows for explicit control over the trade-off between DoA and range estimation accuracy through adjustable coefficients, which is valuable for practical applications.

Key Results

  • Increased DoFs: The proposed methods, particularly DANM and CRM, significantly increase the number of detectable targets compared to standard techniques like SST (which discards non-consecutive coarray lags) and CMT-BCS. For instance, in a scenario with 132 targets, CMT-BCS struggles, while DANM and CRM can still resolve most targets.
  • Improved Accuracy: Numerical simulations show that the proposed CRM method achieves superior Root Mean Square Error (RMSE) for both DoA and range estimation, closely approaching the theoretical Cramér-Rao Bound (CRB). The CRM consistently outperforms DANM and other baseline methods.
  • Computational Efficiency: While DANM and CRM involve more complex optimization than simpler methods, the ADMM-based solver for CRM significantly reduces computational complexity, making it comparable to other advanced methods.
  • Robustness: The methods demonstrate robustness to grid mismatch and are effective even when the number of targets exceeds the DoFs of the physical FDCA.

Limitations and Future Work

  • Identifiability: While the proposed methods can identify more targets than theoretically possible with the coarray without interpolation, the identifiability is still limited by the number of non-negative unique lags and can be affected by grid mismatch. The paper notes that the number of sources K should not exceed D² + 2D for guaranteed identifiability.
  • Coefficient Tuning: While the paper provides guidelines, optimal selection of coefficients (like γp, γf, and μ) might still require some empirical tuning, especially in multi-target scenarios where a trade-off between DoA and range estimation is managed.
  • Future Work: The paper suggests exploring multiple measurement vector (MMV) formulations for scenarios with multiple snapshots and further investigating the impact of coefficient settings. The ADMM-based solver's efficiency is highlighted, suggesting its potential for real-time applications. The theoretical analysis of identifiability and performance bounds provides a strong foundation for future research in this area.

Summary updated

Abstract

In this paper, we address the problem of joint direction-of-arrival (DoA) and range estimation using frequency diverse coprime array (FDCA). By incorporating the coprime array structure and coprime frequency offsets, a two-dimensional space-frequency virtual difference coarray corresponding to uniform array and uniform frequency offset is considered to increase the number of degrees-of-freedom (DoFs). However, the reconstruction of the doubly-Toeplitz covariance matrix is computationally prohibitive. To solve this problem, we propose an interpolation algorithm based on decoupled atomic norm minimization (DANM), which converts the coarray signal to a simple matrix form. On this basis, a relaxation-based optimization problem is formulated to achieve joint DoA-range estimation with enhanced DoFs. The reconstructed coarray signal enables application of existing subspace-based spectral estimation methods. The proposed DANM problem is further reformulated as an equivalent rank-minimization problem which is solved by cyclic rank minimization. This approach avoids the approximation errors introduced in nuclear norm-based approach, thereby achieving superior root-mean-square error which is closer to the Cramér-Rao bound. The effectiveness of the proposed method is confirmed by theoretical analyses and numerical simulations.
BibTeX
@article{mao2022joint,
  author = {Zihuan Mao and Shengheng Liu and Yimin D. Zhang and Leixin Han and Yongming Huang},
  title = {{Joint DoA-Range Estimation Using Space-Frequency Virtual Difference Coarray}},
  journal = {IEEE Trans. Signal Process.},
  year = {2022},
  url = {https://openreview.net/forum?id=AxKq9Hg0_JEd},
}
Figure