Abstract

Both Kalman and Lainiotis filters arise in linear estimation and are associated with linear systems. In this paper, we investigate the partition of linear systems into lower dimensioned independent linear systems. We study the effects of partitioning on filter behaviour and computational burden. It is concluded that the partition does not affect the behaviour of Kalman and Lainiotis filters. Simulation results show that Kalman and Lainiotis filters compute the same outputs (estimation and estimation error covariance), both in the single and in the partitioned model. It is also concluded that the partition leads to significant speedup in both Kalman and Lainiotis filters. Specifically, it is shown that the partition into p independent models leads to super-linear speedup in the range (p, p2) for both filters.

Keyword

Linear systems, Kalman filter, Lainiotis filter, Partitioning.

Cite this article

Refference

[1][1]Anderson BDO, Moore JB. Optimal filtering. Dover Publications, New York, USA; 2005.

[2][2]Kalman RE. A new approach to linear filtering and prediction problems. Journal of Basic Engineering. 1960; 82(1):35-45.

[3][3]Assimakis N, Adam M. Discrete time Kalman and Lainiotis filters comparison. International Journal of Mathematical Analysis. 2007; 1(13):635-59.

[4][4]Lainiotis D. Partitioned linear estimation algorithms: Discrete case. IEEE Transactions on Automatic Control. 1975; 20(2):255-7.

[5][5]Kim GM, Lin DC, Kim JH, Yoon SH. Robust head tracking based on hybrid color histogram and random walk Kalman filter. North Carolina Agricultural and Technical State University Greensboro department of Electrical Engineering; 2002.

[6][6]Assimakis N, Adam M. Global systems for mobile position tracking using Kalman and Lainiotis filters. The Scientific World Journal. 2014: 1-10.

[7][7]Assimakis A, Adam M. Mobile position tracking in three dimensions using Kalman and Lainiotis filters. The Open Mathematics Journal. 2015; 8(1):1-6.