Multiple model concepts in navigational applications

In this work we study multi model approaches for the purpose of online EKF tuning in strapdown navigation applications. Based on the weaknesses of a single navigation filter (with fixed parameters) not being able to comply with multiple noise conditions, we suggest a parallel application of multiple filters. Each filter corresponds to a hypothesis of the system noise matrix. The likelihood of each model is determinable using the normalized innovation squared (NIS) in presence of position observations (e.g. via GPS). In detail, we examine two methods commonly applied in target tracking algorithms: the interactive multiple model (IMM) and the autonomous multiple model (AMM) approach. An analysis of the two architectures is presented outlining pros and cons. Finally, results of the filter bank approach are compared to a conventional navigation filter based on synthetic and experimental data.