Optimization Algorithims in RAN

Until now, I was in favour with using Spring algorithm in RAN. Long time back, I tested Spring algorithm against Simplex Downhill and found that they perform with comparable accuracy. But Spring requires much lesser amount of coding and theoretically has low order of complexity.

However, for curiosity shake (or to further validate my argument), I implemented Simplex algorithm too for landmark aided positioning (LAP). And suprisingly found out my simulation sped-up hundreds of times faster (10 min compared to four hours)!

When discussed with prof, we agreed that even though Simplex is more complex in theory, practically it can be fast. This is what many literature claims too (I spent some times yesterday reading some books om Simplex). On the other hand, Spring can be slow in practice even though it is very easy to implement.

Further discussion with prof revealed that I can actually change my stopping creteria for the Spring algorithm to make it faster. Earlier I was checking the convergence after each move imposed by each landmarks. Actually it is the resultant movement, i.e. the movement after adjustments for all the landmarks, is the one to be checked for convergence. I have to compare the results of this new spring algorithm and simplex algorithm.