[Paper Review] Sensor Network Localization, Euclidean Distance Matrix Completions, and Graph Realization

Yichuan Ding, Nathan Krislock+2 more|ArXiv.org|Dec 14, 2006

Sparse and Compressive Sensing Techniques17 references4 citations

TL;DR

This paper reframes sensor network localization (SNL) as a Euclidean Distance Matrix (EDM) completion problem, demonstrating that SDP relaxations for SNL are equivalent to known EDM completion formulations. By treating anchors as a clique and projecting onto the minimal face of the SDP cone, the method reduces problem size, restores Slater's condition, and improves numerical stability. The key contribution is a more stable, lower-dimensional SDP formulation using a quadratic constraint, which outperforms linearized Schur complement approaches in conditioning and convergence speed.

ABSTRACT

We study Semidefinite Programming, \SDPc relaxations for Sensor Network Localization, \SNLc with anchors and with noisy distance information. The main point of the paper is to view \SNL as a (nearest) Euclidean Distance Matrix, \EDM, completion problem and to show the advantages for using this latter, well studied model. We first show that the current popular \SDP relaxation is equivalent to known relaxations in the literature for \EDM completions. The existence of anchors in the problem is {\em not} special. The set of anchors simply corresponds to a given fixed clique for the graph of the \EDM problem. We next propose a method of projection when a large clique or a dense subgraph is identified in the underlying graph. This projection reduces the size, and improves the stability, of the relaxation. In addition, viewing the problem as an \EDM completion problem yields better low rank approximations for the low dimensional realizations. And, the projection/reduction procedure can be repeated for other given cliques of sensors or for sets of sensors, where many distances are known. Thus, further size reduction can be obtained. Optimality/duality conditions and a primal-dual interior-exterior path following algorithm are derived for the \SDP relaxations We discuss the relative stability and strength of two formulations and the corresponding algorithms that are used. In particular, we show that the quadratic formulation arising from the \SDP relaxation is better conditioned than the linearized form, that is used in the literature and that arises from applying a Schur complement.

Motivation & Objective

• To reframe sensor network localization (SNL) with anchors as a Euclidean Distance Matrix (EDM) completion problem to leverage well-established EDM theory.
• To demonstrate that anchors are not special but correspond to a fixed clique in the EDM graph, enabling unified treatment with other sensors.
• To improve numerical stability and reduce problem size by projecting the SDP relaxation onto the minimal face of the cone when a clique or dense subgraph exists.
• To compare the robustness and conditioning of two SDP formulations: the quadratic constraint $ Y - XX^T \succeq 0 $ and the linearized Schur complement form.
• To develop and test a primal-dual interior-exterior path-following algorithm for the SDP relaxation, emphasizing numerical efficiency and accuracy.

Proposed method

• Reformulate the SNL problem as a nearest EDM completion problem, where sensor positions correspond to points realizing a partial distance matrix.
• Use semidefinite programming (SDP) relaxations to solve the EDM completion, showing equivalence to existing SDP formulations in the literature.
• Apply minimal cone projection to reduce the SDP problem size when a clique (e.g., anchors or dense subgraphs) is identified, restoring Slater's constraint qualification.
• Propose a new sensor position estimation method based on principal component analysis (PCA), improving accuracy over direct extraction from the SDP solution.
• Implement a Gauss-Newton primal-dual interior-exterior path-following algorithm to solve the SDP relaxation, using both quadratic and linearized (Schur complement) constraints.
• Compare the two formulations using numerical tests on randomly generated sensor networks with varying parameters, measuring convergence speed, duality gap, and solution accuracy.

Experimental results

Research questions

• RQ1 Can sensor network localization with anchors be equivalently modeled as a Euclidean Distance Matrix (EDM) completion problem?
• RQ2 Does treating anchors as a clique in the EDM graph enable improved numerical stability and problem size reduction?
• RQ3 Is the quadratic SDP constraint $ Y - XX^T \succeq 0 $ more numerically stable than the linearized Schur complement form $ \begin{bmatrix} I & X^T \\ X & Y \end{bmatrix} \succeq 0 $?
• RQ4 Can the minimal face projection technique be generalized to other dense subgraphs to further reduce problem dimension and improve conditioning?
• RQ5 How does the choice of objective norm (ℓ₂ vs ℓ₁) affect solution accuracy and algorithmic performance in SNL?

Key findings

• The SDP relaxation for SNL is mathematically equivalent to known EDM completion relaxations, validating the reformulation approach.
• Projection onto the minimal face of the SDP cone after identifying a clique (e.g., anchors or dense subgraphs) reduces problem size and restores Slater's condition, improving stability.
• The quadratic constraint formulation $ Y - XX^T \succeq 0 $ is better conditioned than the linearized Schur complement form, requiring fewer iterations to achieve a given duality gap tolerance.
• Numerical tests show the quadratic formulation achieves higher accuracy with fewer iterations, confirming theoretical expectations about central path distortion in the linearized version.
• The PCA-based sensor position estimation method consistently outperforms direct extraction from the SDP solution, reducing average error by over 80% in tested instances.
• The Gauss-Newton primal-dual path-following algorithm demonstrates robust performance, especially when applied to the quadratic formulation, even on ill-conditioned problems.