Bibliography

(eng): Abstract Rao-Blackwellied particle filter is a stochastic filter combining Kalman filters with particle filters. It is suitable for models that could be decomposed into linear and nonlinear part. Since the conditionally linear part can be solved by the Kalman filter, the sequential Monte Carlo is run only on the non-linear subspace. The resulting algorithm is a parallel evaluation of multiple Kalman filters with resampling. The parallel nature of this algorithm allows for very efficient implementation in hardware supporting parallel computation processes. In this contribution, we present implementation of the algorithm in the Field Programmable Gate Array (FPGA). Due to the used model and optimized implementation, the execution time of the filter is in units of microseconds and scales very favorably with the number of particles. This is demonstrated experimentally on a laboratory prototype of sensorless drive with permanent magnet synchronous machine (PMSM) of rated power of 10.7kW.