Adative memory and particle based sequential implementation resampling for particle filtering

The particle filter provides numerical approximation to a nonlinear filtering problem, especially during signal or data transmission. In a heterogeneous environment, reliable state estimation is a critical issue due to the unbalanced particle distribution called sample degeneracy and impoverishment. To address such a problem, sequential implementation resampling (SIR) considers the cause and environment of every specific resampling task decision. However, SIR only considers the cause and environment in a specific situation, which cannot generates reliable state estimation during filtering process. Apart from that, the developed SIR may suffer with unbalanced memory usage, which is reflected in the overall consumed system memory and time. Therefore, this research designed a resampling scheme that generates reliable state estimation and balances the resampling memory usage during particle filtering. To achieve this aim, an adaptive memory and particle sequential implementation resampling (AMPSIR) scheme was designed for different sample impoverishment environments, introduced three enhanced schemes to ensure reliable final state estimation and balanced theresampling memory allocation. The first scheme was the adaptive noise and sample size special strategies resampling (ANSSSR), which combined resampling task from three different types of special strategies resampling, and then reduced state estimation error in different situations in high sample impoverishment. Secondly, the scheme known as adaptive noise and sample size sequential implementation resampling (ANSSIR) combined resampling tasks from three different types of sequential implementation resampling, and then produced a reduction of state estimation error in different stages of sample impoverishment. Finally, the third scheme was the adaptive memory single distribution resampling (AMSDR), which combined resampling tasks from two different types of single distribution resampling, and then generated optimization of resampling memory. All of these enhanced schemes reacted based on measurement detection of particle noise, particle sample size and resampling memory. Simulation results showed that AMPSIR scheme achieved improved performance in termsof reducing state estimation error in different situations in high sample impoverishment by 7.26%, reduced state estimation error in different stages of sample impoverishment by 24.78%, and optimized resampling memory by 28.73% as compared to the existing resampling schemes. The findings showed that the AMPSIR scheme has the capability to do different kinds of resampling tasks, and choose a suitable scheme based on detected noise, sample size and memory measurements. In conclusion, the AMPSIR scheme has been proven to be a valuable solution for different sample impoverishment environments and different resampling memory usage. Besides, it has the ability to adapt the end user’s application memory usage with the scheme to determine the most suitable resampling scheme based on the application memory usage