PURPOSE: To overcome the trouble that the signal from a non-linear sensor cannot be directory processed and the signal from the sensor harmfully influences a process model of a Karman filter.

CONSTITUTION: A sample Kalman filter comprises a variable amplifier 1, and the amplifier 1 is, through an adder 3, connected to a delay device 2. The output is, through a negative feedback loop constituted with two matrix calculators 4 and 5, connected to an input of the variable gain amplifier 1, the output of the calculator 5 is, through an addition circuit 6, connected to an input of the amplifier 1. The addition circuit 6 comprises two inputs, and the first input is connected to the output of the matrix calculator 5, and the second input receives an observation value or a measurement value vector Z (n). At initialization stage, the value of a status vector V<0 of 0> of the filter is equal to an expected value E (XO) of a variable X, since an estimator has not captured a knowledge of the measurement value.