The processes in question are jump processes and processes with jumping velocity. We estimate the current position of the stochastic process based on past discrete-time observations (non-linear discrete filtering problem). We obtain asymptotic rates for the expected square error of the filter when observations become frequent. These rates are better than those of a linear Kalman filter. For jump process, our method is asymptotically free of the process parameters. Also, estimation of process parameters is addressed. |