Abstract

The use of an extended Kalman filter (EKF) as a nonlinear speed and position observer for permanent-magnet synchronous motor drives is a mature research topic. Notwithstanding, the shift from research prototype to a market-ready product still calls for a solution to some implementation pitfalls. The major and still unsolved topic is the choice of the EKF covariance matrices. This paper replaces the usual trial-and-error method with a straightforward matrices choice. These matrices, possibly combined with a novel self-tuning procedure, should put the EKF drive much closer to an off-the-shelf product. The proposed method is based on the complete normalization of the EKF algorithm representation. Successful experimental results are included in the paper.