Fine-grain parallel processing implementations of Kalman filter algorithms

To implement the Kalman filter, a number of matrix and vector manipulations must be executed within each sample period. Consequently, a drawback of the algorithm is that it is computationally intensive, requiring O(n³) operations for each estimate of the n element state vector. Thus, parallel versions of the Kalman filter have been developed in order to reduce execution times and enable the algorithm to be implemented in an increased range of real-time applications. The paper investigates the potential of a `fine-grain´, parallel processing architecture, i.e. the GAPP chip (geometric arithmetic parallel processor), for implementing the Kalman filter algorithm