| With the advent of the Internet of Things(IoT)era,embedded devices with integrated Inertial Measurement Unit(IMU)will be more widely integrated into people's lives,but IMUs widely used in the market have low cost and performance Poor,non-uniform standards.Previous studies have focused on high-precision,unified-standard scenarios.Migrating algorithms to modem mobile phones and other IoT terminals often suffers from slow convergence and poor practicality.To overcome the above problems,this thesis will focus on the selection and design of IMU data processing algorithms in the above scenarios,the problem of dynamic system process noise parameter estimation,and propose a more efficient Maximum Likelihood Estimation(MLE)algorithm.The main work and results are as follows:(1)Aiming at the poor performance and inconsistent standards of the current mobile terminal IMU,this thesis models the IMU data processing problem as a dynamic system state estimation problem and proposes an IMU filtering algorithm based on kalman filtering.The kalman filter model of IMU is obtained by fuzzy processing the gyroscope and accelerometer motion reference models.IMU data measured from five mobile phone terminals at the same time,in the same place,and in the same physical process were used to analyze the sampling frequency and noise distribution.Finally,simulation analysis verifies that the algorithm has better noise reduction effect than the filtering algorithm used by mobile phone terminal manufacturers.(2)Aiming at the Qalman filter Q estimation problem,sufficient conditions for the MLE solvability are found,and the solvability of the MLE algorithm is proved.Aiming at the slow convergence speed of the MLE algorithm in the Q estimation algorithm,a golden section-based MLE algorithm was proposed,and a sequence model of the MLE algorithm was established.The convergence algorithm in the likelihood function was replaced by a golden section one-dimensional search algorithm.The convergence ratio ensures faster algorithm convergence.The simulation results show that compared with the gradient descent based MLE,the golden section based MLE proposed in this thesis can converge to the optimal estimate of Q faster.(3)To slove the problem that the MLE algorithm cannot be applied to nonlinear systems,an MLE approximation algorithm for non-linear systems is proposed.The linear interval is selected,the MLE algorithm is used to obtain the Q estimate,and the stationary parameter is used to obtain the approximate Q parameter value of the system.Aiming at the problem of the slow convergence speed of the MLE algorithm in a non-linear system,the convergence speed is improved by replacing the convergence algorithm in the likelihood function with the golden section algorithm.The simulation data shows that the nonlinear MLE can also converge to the Q-optimal estimate,and the MLE based on the golden section under nonlinear conditions converges faster than the MLE based on gradient descent. |
