A close examination of Multiple Model Adaptive Estimation versus single extended Kalman filter for precision attitude determination

With today's advanced nonlinear filtering, e.g., Multiple Model Adaptive Estimator (MMAE) or Particle Filter (PF), and noise identification for filter update at the roll-off level (via the filter process noise covariance matrix) and noise cancellation at the measurement level approach at a reasonable technology maturation level, it is strongly believed that an extremely high precision attitude determination can be practically achieved using commercial low cost low grade MEMS inertial sensors (i.e., MEMS gyros and accelerometers and/or MEMS IMU). This paper revisits the MMAE design developed in the past with a close examination of its performance using high fidelity models of the gyros and star tracker to determine its viability for a possible design and implementation of a new attitude determination system using low-cost low-grade MEMS gyros and CMOS star trackers. The proposed MMAE design with gyro noise identification (i.e., Angular Random Walk (ARW) and Rate Random Walk (RRW) are primary elements to be estimated for update and cancellation) is evaluated against the single EKF based design for a performance measure of how well the proposed MMAE and noise identification scheme improve over the baseline design. The design is evaluated using several simulations runs.