A hybrid parallel-serial approach to nonlinear filtering

Nonlinear filtering is often accomplished using algorithms, such as the extended Kalman filter, which process data serially by linearizing the state equations about single solution hypotheses. This linearization introduces losses which may ultimately cause these procedures to diverge at low signal-to-noise ratios. Parallel filtering techniques eliminate these linearization losses at the price of more processing. A new hybrid approach is presented which exploits the advantages of both procedures, threshold reduction with parallel filtering and processing efficiency with serial filtering. An example from bearings only target state estimation is provided.