Kalman Filters
-
just wondering if anyone here knows anything about this technique called Kalman Filters that is used for signal processing including radars.
-
-
LOL
This word has become a meme within the team
-
thanks for the reference
-
@Alfred Itâs a technique where you want to estimate something that you donât know the true value of. Could be a temperature or it could be the position of an enemy aircraft.
You start out with an initial guess, eg. âI think the temperature is 75, plus/minus 2â. You then do an actual measurement and note the value of that observation. Your measurement has uncertainty, mainly from the method you use for the observation. A thermometer is slightly inaccurate, a radar scope also.
From the uncertainty of the initial guess, and the uncertainty of the measurement, you calculate something called the Kalman Gain. Together with the initial guess, the Kalman Gain allows you to refine or improve the measurement you did, in the direction of coming closer to the true value you are after.
You then do another measurement, repeat the process but this time your previous (and refined) measurement takes the place of the initial guess. Rinse and repeat.
By doing iterations this way, you will converge towards the true value of what you are looking for. The challenge is to know when to stop measuring since it takes time and/or computer cycles.
Plenty of examples online, so lends itself pretty well to a small testcase in Excel or similar
-
thank you very much jayb, your description pretty much says it all. I read about kalman gain, so I have a slight idea of what it is. I think the other challenge is to start with the best initial guess.
-
@Alfred As you go through the iterations (more and more observations) the original initial guess becomes less important. It is only there to get the iterations started. You should make sure that the uncertainty that you state for the initial guess (which you use for the first value of the Kalman Gain) is large enough to encompass the true value that you are looking for
-
There is a Kalman filter in the forum algorythm.
-
Kalman Filters technique is exactly how I pick up ladies in bars, but the more I drink, the harder it is to calculate the Kalman Gain - so I just repeat the process every night hoping to eventually converge on the perfect pickup-line.
-
@SemlerPDX - actually, the way to go about it is to note and buffer the gals that DONâT work out and how to eliminate the the Type from the ID algorithm in order to converge on the one(s) (plural, if youâre employing Tryst While Scan mode) that will. Thatâs the RL functional GainâŚ