Hardware Implementation Of The Particle Filter Based On FPGA

Particle Filter is based on Monte Carlo method and recursive Bayesian estimation theory, which can represent the required posterior probability density function by a set of weighted particles. Because of the advantages in the aspect of state and parameter estimation, particle filters showed a great practical value and broad application in nonlinear and non-Gaussian systems. However, due to the high computation complexity, it is difficult to use particle filter in the real-time system. In order to deal with such problem, we designed an efficient architecture according to the optimization method. The new hardware-based implementation can be used in the nonlinear non-Gaussian target tracking system with the property of high speed and low resource utilization.In this paper, we summarize the development of particle filter and point out the research significance firstly. Then, we introduce the basic particle filter with the Bayesian theory and Monte Carlo method as the theoretical foundation. In order to implement particle filter with efficient hardware architecture, some work has been done to optimize the algorithm by removing the redundant operation and approximate calculation. We also proposed a high efficiency redundancy storage resampling methods to relieve the pressure of memory access. Finally, a parallel architecture of particle filter is proposed to finish the complex calculation with a high throughput. The novel architecture is based on the redundancy storage strategy and reduce the weights imbalance phenomenon.We implement the new architecture based on the FPGA platform and deal with the bearing-only tracking problem in two-dimensional nonlinear and non-Gaussian system. The experiment results show that the designed hardware is extremely suitable for real-time implementation while maintaining a high performance. The proposed hardware architecture can be widely used in the target tracking, wireless communication, navigation and other area with nonlinear and non-Gaussian style.