Research On Signal Processing Of MEMS Gyroscope Based On Deep Recurrent Neural Network

This paper mainly studies the insufficient accuracy of MEMS(micromachined)gyroscopes.Based on the characteristics of cyclic neural networks that can process time series data,the algorithm filtering can effectively reduce various errors.Provide theoretical support and research reference in MEMS gyroscope signal processing.First of all,according to the working principle of MEMS and different error models and measure the performance index of the gyroscope.The Allan variance analysis method is used to realize the key identification of the static drift error of the MEMS gyroscope,and distinguish the three main random errors in the output data of the original MEMS gyroscope.Then,a Kalman filter is designed based on the time series to filter the MEMS static drift data.After Kalman filtering,the most important zero-bias instability coefficients of the X,Y,and Z axes are respectively reduced by 6.7%,16.8%,and 40.9%compared to the original data,indicating that Kalman filtering can effectively reduce the static drift of MEMS gyroscopes error.Secondly,in view of the insignificant reduction of the Kalman filter error coefficient,the output is assumed to be zero-mean white noise in an ideal state.It is proposed to construct a long and short memory network combined with a gated recurrent unit(LSTMGRU)and a double-layer long and short memory neural network(LSTM-LSTM)to reduce the noise of the static drift error of the MEMS gyroscope’s X-axis,Y-axis,and Zaxis,using Allan variance Analyze the main random error coefficient after noise reduction.The simulation results show that after LSTM-GRU noise reduction,the zero-bias instability coefficients of the X-axis,Y-axis,and Z-axis are reduced by 91.4%,83.7%,and 86.3% respectively;after the LSTM-LSTM noise reduction,the three-axis zero-bias is unstable The coefficient of sex decreased by 98.1%,96.1%,and 89.7%,respectively.The experimental results show that the recurrent neural network is effective in dealing with the static random drift error of the MEMS gyroscope.The noise reduction effect based on LSTM-LSTM is better than that of LSTM-GRU,and the recurrent neural network can achieve better noise reduction than the traditional Kalman filter.effect.Finally,in order to further verify the effectiveness of the proposed method,a dynamic turntable experiment is set up.The experimental results are: the two filtering methods and the standard deviation of the original dynamic data have been reduced,but the application of the LSTM method to achieve the noise reduction of the MEMS gyroscope is better than GRU.