An inertial neural network approach for robust time-of-arrival localization considering clock asynchronization

Area of Science:

• Optimization Algorithms
• Signal Processing
• Machine Learning

Background:

• Source localization is crucial in various applications, often relying on Time of Arrival (TOA) techniques.
• Existing methods face challenges with noisy data, outliers, and clock asynchronization, limiting accuracy.
• The l1-norm objective function offers robustness but requires efficient optimization methods.

Purpose of the Study:

• To develop and analyze an inertial neural network (INN) for robust source localization using the TOA technique.
• To investigate the convergence and stability properties of the proposed INN.
• To address the complexities of clock asynchronization in TOA-based localization and enhance real-world applicability.

Main Methods:

• An inertial neural network (INN) is proposed for solving the l1-norm source localization problem.
• Lyapunov function method is employed to analyze the convergence and stability of the INN.
• An iterative INN approach is utilized to optimize solutions by exploring different inertial parameters.
• The model is extended to incorporate clock asynchronization for practical scenarios.

Main Results:

• The INN demonstrates convergence and stability, validated through Lyapunov analysis.
• Numerical simulations with Gaussian noise and uniform outliers show superior performance.
• Experiments using ultra-wideband (UWB) hardware confirm the method's effectiveness.
• The proposed approach consistently yields more accurate source positions compared to existing algorithms, with or without clock asynchronization.

Conclusions:

• The developed inertial neural network provides a more effective and accurate solution for TOA-based source localization.
• The method exhibits robustness against various noise types and clock asynchronization issues.
• This work advances practical source localization by offering a reliable algorithm for real-world applications.