Well, understanding the math of a Kalman filter was way beyond my pay grade, but based on the description in http://bilgin.esme.org/BitsBytes/KalmanFilterforDummies.aspx, a 1d implementation reduces the complexity considerably. According to this guide, we can assume simple float values for most of the coefficients for most purposes.
Based on your suggestion, I think I will incorporate an "analyze mode" into the external itself to calculate the noise parameters and set them automatically. I'll let you know when that's in git.
Joel
ChuckI like the approach generally--you have parameters for the assumed noise model and methods to set them (better than trying to build a monolith that does both the measurement and filtering). Do you have another patch or abstraction to analyze the sensor data and calculate those parameters? If so, you should add it to git.Hey JoelI was very interested to see your implementation. It's drastically simpler than I thought it would be. Well, you did mention it was simple :) However, I thought the math was pretty expensive to do and complex to program.
On Thu, Feb 28, 2013 at 12:47 PM, Joel Matthys <jwmatthys@gmail.com> wrote:
I just completed a very simple 1D Kalman filter Pd external. I haven't really done any documentation on it, but it seems pretty robust for cleaning up 1D sensor inputs.
The source is here:
https://github.com/jwmatthys/kalman-pd
Joel
_______________________________________________
Pd-list@iem.at mailing list
UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list