Right. I agree (mostly--just quibbling about nirvana with what Jo has said.

To rephrase, there is no explicit backlash compensation in Linear, but rather the algorithm tries to set itself up in a situation without backlash.
It does this by first moving far "out", then moving back "in" by quite a bit, in hopes that the inward motion eats up all the backlash.
It then tries to sample a v-curve, moving "in" by step size, sampling the HFR, moving in again ... eventually find where the bottom of that curve is
(by moving past the minimum and noticing the HFR has started to increase). It really doesn't care what the absolution position is, just the value of
the HFR at the bottom of the V. It then performs its 2nd pass, whose goal is to find the minimum again and stop there (at whatever
absolution position, doesn't matter). It does that 2nd pass by again moving back out quite a bit, and again reversing to inward, moving quite a bit to
again hopefully eat up all the backlash. Hopefully, the position it finds itself in at that point is outside the minimum HFR position.
It then performs its 2nd inward sweep, iterating the following: moving in my 1/2 step size, measuring an HFR, and deciding if its done.
It's done if the measured HFR is close enough to the minimum HFR from the 1st pass (and its no longer decreasing).
It may decide that something went wrong, and instead of going off to nirvana, repeating the backward/forward motion.
It always terminates after 30 steps.

It's not perfect, but can often overcome backlash.

Read More...