Neural Network Augmentation of Attitude Estimation Using Navigation Satellite Signal Phase
We consider the attitude estimation of an aircraft utilizing navigation satellite carrier phase measurements. An Extended Kalman Filter (EKF) for the Euler angles is augmented by an artificial neural network (ANN) to improve its estimation performance. MLP and RBFN networks are trained for various levels of manoeuvre and measurement noise, and their performance compared under complex manoeuvre scenarios. It is shown that the ANN provides significant improvement in the EKF performance. RBFN scores distinctly over MLP in terms of training time and estimation accuracy. The RBFN is optimized and the improvement through multipoint training is estimated.