Adaptive Particle Filter for Nonparametric Estimation with Measurement Uncertainty in Wireless Sensor Networks

Area of Science:

• Signal Processing
• Wireless Sensor Networks
• Estimation Theory

Background:

• Particle filters (PFs) are crucial for nonlinear signal processing in wireless sensor networks (WSNs).
• Measurement uncertainty in WSNs degrades the accuracy of traditional PFs.
• Existing methods often pre-assume distribution models for likelihood calculation, limiting performance.

Purpose of the Study:

• To propose a novel PF method with an improved likelihood calculation for WSNs.
• To enhance estimation performance by addressing measurement uncertainty and noise.
• To develop adaptive PFs adaptable to varying environmental conditions.

Main Methods:

• A dynamic Gaussian model was used to characterize nonparametric measurement uncertainty.
• A likelihood adaptation method was developed, incorporating prior information and a belief factor to mitigate measurement noise.
• The optimal belief factor was determined by minimizing Kullback-Leibler divergence.
• The method was integrated into PFs for a WSN target tracking system, creating three adaptive PF versions.

Main Results:

• The proposed likelihood adaptation method significantly improved PF estimation performance in high-noise environments.
• Adaptive PFs demonstrated high adaptability to environmental changes.
• The method did not introduce significant computational complexity.

Conclusions:

• The novel likelihood adaptation method effectively enhances PF performance in WSNs, particularly under high noise conditions.
• Adaptive PFs offer a robust and computationally efficient solution for WSN signal processing.
• This approach provides a valuable enhancement for target tracking and other WSN applications.