You cant. If you integrate acceleration you get the speed, integrating the speed you get the position. If you want to get an angle you should integrate the output of an inertial (IMU) sensor, so integrating angular rate.

I understand what you mean, this can’t be if we are in a 3 axes freedom, but can’t it be extrapolated if we “lock” an axis :

as a simple example if the measurements are taken from a bike : the Z axis is not used and for simplicity consider that we only want to know how much degrees the driver is doing righ or left :

If you intend to use LIS sensor as an static measuring device, than this could work fine, you should convert cartesian coordinate system in spherical coordinate system (look Matlab’s cart2sph):

theta=atan2(y,x)

phi=atan2(z,sqrt(x^2+y^2)

r=sqrt(x^2+y^2+z^2)

Where theta and phi are azimuth and elevation angles, r is the absolute value of acceleration (sum of all three axes, it should be 1g if it is intended for static measuring).

atan2 is a enhanced function of arctangens(x) in other words its a four-quadrant arctangens.

If the measurements are taken from a bike you would get the dynamic components superimposed. At least you should implement a very large averaging filter.

My english is not very good, with superimposed I meant that dynamic signal is added over a static signal.

Yes, what I have suggested you is an inclinometer. But inclinometer will not work on biclycle when driving, because if you ride in a circle then the centrifugal force keeps you on the bicycle and you will not measure the right angle. It remembers me the a video of a Dakota pilot making loopings and airleron rolls with a glass of water on the instrumental panel without spiling water.