Noise modeling with Markov modeling

5 replies · 2K views · last reply
ramesses
Messages
16
Reaction score
0
Hi
I'm using accelerometer & horizontal gyroscope in order to replace GPS. Now, I'want to model the noise with first order markov process, to use it in kalman filter.
I recorded measurement on all axes and computed auto-correlation.
This picture represents auto-correlation on one of axes.
http://picpaste.com/pics/autocorrelation_x-qjpnbYJk.1437477728.png

Now, I know that the first order markov process takes the following equation :
w = white noise which has the same variance.
and P is the correlation

My problem is how to fix the value of "P" (know as correlation) ?
thank you
 
Last edited by a moderator:
Noise is usually modeled as a stationary process, not a Markov process.
 
In general, a noise process has a mean of 0. A Markov process has a mean, at a given time, the value at the last known sample.
 
The Gauss-Markov process gives a good result as you see in this picture.
Now, I don't understand why I need to use kalman filter in-order to estimate the position's error ?
why we don't integrate directly the Gauss-Markov sequence ?
Line plot with two curves against an unlabeled horizontal axis from 0 to 16000: a blue curve decreasing almost linearly from about 0.9 to about -0.2, and a green curve decreasing more irregularly from 1.0 to about -0.45.
 
Sorry - Ican't answer your specific questions. I have not worked with the specific process or Kalman filters.