From Mladen:

RSX 2

Experimenting with RSX

As we know rsx is a smoother rsi. Was thinking if there is some space in its calculation to experiment and eventually improve it (if possible). So I tried this experiment. What is added are the depth of calculation (maximum 10 - default rsx has "depth" 3) and speed. Speed is a parameter that causes it to "overreact" (in positive sense this time, I hope) and to overcome the flattening problem a bit.

To show what is it about, here is an example of a 32 period rsx with default (upper) and with "speed" set to 0.8 (lower).

As it is obvious 2 things happened : with "speed" 0.8 it is faster than the regular one in detecting crosses (a bar or two but sometimes the difference is much, much more) and it reaches the predefined levels while the regular one almost never does.
I converted it to mt5.