Research On The Algorithm Of Particle Filter And Its Hardware Implementation In SINS/GPS Integrated System

Particle filters are recursive Bayesian filtering approaches based on sequential Monte Carlo sampling. Particle filter can be applied to any nonlinear models or non-Gaussian noise models. The more nonlinear model, or the more non-Gaussian noise, the more potential particle filters have. Along with the rapid development of computer technology and the cost reduction of computational power, particle filter has attracted more and more attention to the researchers from different areas, with many successful applications in signal processing, digital communications, biology, statistics, medicine, chemistry and social sicences, and many others.They have become effective methods of studying the optimal state estimation of nonlinear and non-Gaussian dynamic systems.But particle filters have some drawbacks, such as particle degeneration, impoverishment, huge computation burden.There exist nonlinear equations in the integrated navigation system of strapdown inertial navigation system (SINS) and GPS, such as inertial navigation nonlinear error models and pseudorange observation equations. On the other hand, the noises of the real integrated navigation systems are very complicated, and the classical Kalman fiter and extended Kalman filter are not appropriate. So particle filters have become popular in the military related areas, such as inertial navigation, integrated navigation, car positioning and target tracking.Considering the demand of the integrated navigation system during high-dynamic conditions in the project team, this paper fulfills the hardware implementations of both the particle filter and the 2-sample strapdown inertial navigation computation. Specifically, the research work is as following:(1) Study particle filter algorithms and propose an improved particle filter by fusing general particle filters and the error correction technique of Dynamic Matrix Control which can overcome the computational complexity of particle filters especially in high dimensional models. Simulations in single dimension target tracking and the application in SINS/GPS with large initial azimuth error show it has better convergence and filtering stability performance with fewer paticles. (2) Study the parallel implementation architecture of the improved particle filter on the basis of generic hardware architecture, such as the optimization of storing unit, the improvement of resampling, the generation of Gaussian noise and the calculation of exponent function. After the accomplishment of the hardware implementation based on the SINS error models presented in Chapter 3, timing analysis and simulation verify that the parallel design can make the particle filter run in realtime systems, which has important significance in the field of the realtime application of particle filters.(3) Study the parallel hardware architecture of the 2-sample navigation computation algorithm presented by Savage and finally fulfill the hardware implementation of the algorithm based on FPGA. In this paper, a simple strapdown inertial navigation system is presented, including a low-cost MEMS ADIS16365, ML506 FPGA evaluation platform and a personal computer. The actual test of the system demonstrates that it can accomplish the collection of IMU data, simply data-processing and data transmission from FPGA board to PC.