Noise modeling and analysis of an IMU-based attitude sensor: improvement of performance by filtering and sensor fusion

We describe the characterization and removal of noise present in the Inertial Measurement Unit (IMU) MPU-6050. This IMU was initially used in an attitude sensor (AS) developed in-house, and subsequently implemented in a pointing and stabilization platform developed for small balloon-borne astronomical payloads. We found that the performance of the IMU degrades with time due to the accumulation of different errors. Using the Allan variance analysis method, we identified the different components of noise present in the IMU and verified the results using a power spectral density analysis (PSD). We tried to remove the high-frequency noise using smoothing filters, such as moving average filter and Savitzky-Golay filter. Although we did manage to filter some of the high-frequency noise, the performance of these filters was not satisfactory for our application. We found the distribution of the random noise present in the IMU using a probability density analysis, and identified the noise to be white Gaussian in nature which we successfully removed by a Kalman filter in real time.