A Gap between Simulation and Practice for Recursive Filters: On the State Transition Noise

In order to evaluate and compare different recursive filters, simulation is a common tool and numerous simulation models are widely used as 'benchmark'. In the simulation, the continuous time dynamic system is converted into a discrete-time recursive system. As a result of this, the state indeed evolves by Markov transitions in the simulation rather than in continuous time. One significant issue involved with modeling of the system from practice to simulation is that the simulation parameter, particularly e.g. the state Markov transition noise, needs to match the iteration period of the filter. Otherwise, the simulation performance may be far from the truth. Unfortunately, quite commonly different-speed filters are evaluated and compared under the same simulation model with the same state transition noise for simplicity regardless of their real sampling periods. Here the note primarily aims at clarifying this problem and point out that it is very necessary to use a proper simulation noise that matches the filter's speed for evaluation and comparison under the same simulation model. Keywords—Recursive filters, discrete-time dynamic system, Markov transition T. Li’s research interest is nonlinear filtering (especially particle filters), multi-target tracking, information fusion, artificial intelligence. He is currently with the School of Mechatronics, Northwestern Polytechnical University, Xi’an, 710072, China. He is also with the Center for Automated and Robotics NDT, London South Bank University, London, SE1 0AA, UK Contact email: [email protected], [email protected] A Gap between Simulation and Practice for Recursive Filters: On the State Transition Noise