Optimal pre-filter design for SINS based on particle swarm optimization

The accuracy of strapdown inertial navigation system (SINS) generally cannot achieve an acceptable level for the high accuracy SINS. Common method to solve the problem is to improve the accuracy of the inertial sensor or the navigation algorithm. However, this paper focuses on the influence that pre-filter applies on the accuracy of SINS. The minimum influence can be obtained at less cost by adjusting the pre-filter. A novel approach is proposed to design an optimal pre-filter to match the four sample attitude algorithm for improving SINS accuracy. A pre-filter which is often used to remove the high frequency noise in the inertial sensors outputs has a shaped frequency characteristic of inertial sensors outputs. The characteristic leads to an influence of the pre-filter on the accuracy of SINS. Different types and different parameters of pre- filters that shape inertial sensor outputs differently have different influences on the accuracy of SINS. For designing an optimal pre- filter, the presented approach utilizes the particle swarm optimization (PSO) algorithm to find the optimal type and parameters for the optimal pre-filter design. The PSO algorithm is an intelligence algorithm widely used for the optimization of continuous nonlinear function. It can speed up the designing process owing to a fast convergence. This paper just chooses finite impulse response (FIR) filter and wavelet filter as the optimized objects to validate the effectiveness of the approach. The optimal pre-filter design procedure is presented in detail. In order to analyze influences of different types and different parameters of the pre-filter, the yaw error is regarded as the measurement of the influence. The PSO algorithm is used for the optimization of the parameters of FIR filter and wavelet filter. Because coning error is affected by the pre-filter, it is used to construct the fitness function which is needed in the PSO algorithm to evaluate parameters. The derivation of fitness function is provided in this paper.