Semidefinite relaxation method for unified near-Field and far-Field localization by AOA

Xianjing Chen,Gang Wang,K.C. Ho

DOI: https://doi.org/10.1016/j.sigpro.2020.107916

IF: 4.729

2021-04-01

Signal Processing

Abstract:<p>This paper investigates the unified near-field and far-field localization using angle-of-arrival measurements, where the knowledge that the source is in the near-field or far-field is not required. We propose a semidefinite relaxation method to solve this problem by using the modified polar representation of the source. By approximating the measurement model, we formulate the localization problem as a weighted least squares problem, which is a non-convex constrained optimization problem. This problem is then relaxed to a convex semidefinite program (SDP) by applying semidefinite relaxation. The relaxed SDP problem is tightened by adding a set of second-order cone constraints to reach a tighter mixed semidefinite and second-order cone program. The proposed method provides an accurate position estimate for the near-field source, and yields a precise direction of arrival estimate for the far-field source. To reduce the estimation bias introduced by the approximation of the measurement model, we further derive the analytical expression of the bias and subtract it from the solution to obtain an approximately unbiased estimate. Both simulations and theoretical analysis show that the proposed method reaches the Cramer-Rao lower bound when the noise is not large under Gaussian noise, and has a comparable bias with the maximum likelihood estimator.</p>

engineering, electrical & electronic

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the unified positioning of near - field and far - field targets. Specifically, the authors propose a method based on semidefinite relaxation (SDR) to achieve high - precision positioning of near - field and far - field targets through angle - of - arrival (AOA) measurements.

In traditional positioning methods, the positioning of near - field and far - field targets usually requires different models and algorithms, which increases the complexity and computational cost of the system. The SDR method proposed in this paper aims to provide a unified framework that can handle the positioning problems of near - field and far - field targets simultaneously, thereby simplifying the system design and improving the positioning accuracy.

### Main problem description:
1. **Distinction between near - field and far - field targets**: In traditional methods, the positioning models for near - field and far - field targets are different, resulting in the need to design and optimize algorithms separately.
2. **Requirement for high - precision positioning**: In practical applications, both near - field and far - field targets require high - precision positioning results.
3. **Computational complexity**: Traditional positioning methods may involve complex non - linear optimization problems and are difficult to solve efficiently.

### Solution:
The authors propose a unified positioning method based on semidefinite relaxation. The core idea is to transform the original non - convex optimization problem into a convex optimization problem, making the problem easier to solve. Specific steps include:

- **Problem modeling**: Establish the mathematical relationship between the target position and the sensors through AOA measurements.
- **Semidefinite relaxation**: Relax the original non - convex optimization problem into a semidefinite programming problem and solve it using convex optimization techniques.
- **Unified framework**: This method can handle the positioning problems of near - field and far - field targets in the same framework without having to design different algorithms separately.

Through this method, the authors not only simplify the complexity of the positioning system but also improve the positioning accuracy and robustness.

### Formula representation:
Suppose the position vector of the target is \(\mathbf{p}\), the position vector of the sensor is \(\mathbf{s}_i\), and the AOA measured by the \(i\) - th sensor is \(\theta_i\), then the following relationship can be established:

\[
\mathbf{p}-\mathbf{s}_i = r_i\begin{bmatrix}\cos(\theta_i)\\\sin(\theta_i)\end{bmatrix}
\]

where \(r_i\) is the distance from the target to the \(i\) - th sensor. By introducing appropriate relaxation variables and constraint conditions, the above non - convex problem can be transformed into a semidefinite programming problem:

\[
\min_{\mathbf{X}}\quad\text{tr}(C\mathbf{X})
\]
\[
\text{s.t.}\quad\mathbf{X}\succeq0,\quad\text{rank}(\mathbf{X}) = 1
\]

where \(\mathbf{X}\) is a matrix variable, \(\text{tr}(\cdot)\) represents the trace of the matrix, \(\mathbf{X}\succeq0\) indicates that \(\mathbf{X}\) is a positive semidefinite matrix, and \(\text{rank}(\mathbf{X}) = 1\) means that the rank of \(\mathbf{X}\) is 1.

By solving the above optimization problem, the optimal position estimate of the target can be obtained.