Ive recently been playing with a new analytical solution to the multilateration problem, but its complicated and I dont fully understand the maths behind it yet because it was Kevin’s work, not mine. Ive been implementing it into the disdrometer simulation software but results are poor so far. The largest problem is that the principle error checking and selecting technique is to see if the proposed drop occured inside the tank area. When many pulses exist in time, the chances are very high that some combination(s) would be inside that area. So because of this lack of resolution, I cannot include that result in the hit statistics, hence the resulting accuracy is low (typically < 50% at 10mm/h).
I have added some more tests, like quadrant analysis to make sure that the results are in the correct quadrant, not only in the tank and I have also another idea to ultimatly produce an error variable. [Like the fminsearch routines before that picked the combination with the lowest error even if there were more than 1 possible combination]. This would perform a reverse multilateration algorithm to recalculate what the pulse positions really should be if that position is correct. The only problem is that this is using the inaccuracy of the analytical method. Strictly speaking it should use the error of the true vs. suspected data.
Finally, as always (and typically the case), it could be because of a bug in the code. I’m struggling for focus/motivation at the moment and I’m busy this weekend. I dont think I will have my car on monday so I will have to get emma to take me into uni in the morning. A whole day (minus labs) in my office with no distractions should do the trick. Off for a break now..
Post a Comment