OK, kwaiweng. Now you got the idea. Our control variable here is angular velocity.
But I insist; due to the small integration constant it takes a large and long (in time) stick movement for the reference "level" to become tilted more than a fraction of degree. If you make a quick stick movement and then return to center, the DF should level back to about the same "horizontal" plane than before. If not; maybe your integration constant is too large. HINT:
I integrate every 1/50 sec or 20ms. Maybe you're doing this calculation more often? In that case you should reduce the constant accordingly.