Distributed Cooperative Positioning in Dense Wireless Networks: A Neural Network Enhanced Fast Convergent Parametric Message Passing Method

Yue Cao,Shaoshi Yang,Zhiyong Feng

2024-12-22

Abstract:Parametric message passing (MP) is a promising technique that provides reliable marginal probability distributions for distributed cooperative positioning (DCP) based on factor graphs (FG), while maintaining minimal computational complexity. However, conventional parametric MP-based DCP methods may fail to converge in dense wireless networks due to numerous short loops on FG. Additionally, the use of inappropriate message approximation techniques can lead to increased sensitivity to initial values and significantly slower convergence rates. To address the challenging DCP problem modeled by a loopy FG, we propose an effective graph neural network enhanced fast convergent parametric MP (GNN--FCPMP) method. We first employ Chebyshev polynomials to approximate the nonlinear terms present in the FG-based spatio-temporal messages. This technique facilitates the derivation of globally precise, closed-form representations for each message transmitted across the FG. Then, the parametric representations of spatial messages are meticulously refined through data-driven graph neural networks (GNNs). Conclusively, by performing inference on the FG, we derive more accurate closed-form expressions for the a posteriori distributions of node positions. Numerical results substantiate the capability of GNN--FCPMP to significantly enhance positioning accuracy within wireless networks characterized by high-density loops and ensure rapid convergence.

Signal Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in dense wireless networks, in the case where the factor graph (FG) has a large number of short loops, the traditional parametric message passing (MP) method in the distributed cooperative positioning (DCP) method may not converge and is sensitive to initial values. Specifically:

1. **Limitations of the traditional MP method**:
- In high - density wireless networks, the factor graph usually contains many short loops, which will lead to inaccuracies in message passing and make it difficult for the algorithm to converge.
- Using inappropriate message approximation techniques will increase the sensitivity to initial values and significantly reduce the convergence speed.

2. **Specific manifestations of the problem**:
- When dealing with factor graphs with a large number of short loops, traditional methods may lead to a decrease in positioning accuracy and a slowdown in convergence speed.
- Some methods rely on the knowledge of the global graph structure, which is usually not feasible in a distributed wireless network environment.

To solve these problems, the author proposes a graph neural network - enhanced fast - converging parametric message passing method (GNN - FCPMP) to improve positioning accuracy and ensure fast convergence. This method improves the traditional MP method in the following ways:

- **Chebyshev polynomial approximation**: It is used to approximate the nonlinear terms of spatio - temporal messages in the factor graph, thereby reducing the dependence on the initial position estimate and deriving an exact closed - form representation for each message.
- **Data - driven GNN optimization**: Fine - tune the parameterized spatial messages through the graph neural network to further improve the accuracy and robustness of message passing.
- **Closed - form posterior distribution**: By reasoning about the factor graph, derive a more accurate closed - form posterior distribution expression for the node positions.

These improvements enable GNN - FCPMP to significantly improve positioning accuracy and ensure fast convergence in wireless networks with high - density loops.