Thank you indulis, I tried to put feedback capacitor and it improved behavior, but 0.1uF was not enough so we tried 320uF and it is quite good. I have done measurement (with 50kHz of sampling frequency) of transient responses and part of captured data you can see on 1. figure. On 2. figure is shown detail where is see what is my problem now. When the system is in zero voltage state and the transient has to be from this state to arbitrary other, this transient is delayed. Can anybody help me? Thank you PS: Scheme is on figure 3.