On the premise of unknown “total” disturbance model, aiming at tracking accuracy of time-varying signals for linear extended state observer, a feedforward observation compensator was designed. After analyzing the observation principle of linear extended state observer, by linearizing the disturbance and neglecting the dynamic response of the error system, the quantitative expression of the static observation error was derived in the time domain. Then by using the estimated value of the disturbance instead of the true value, the feedforward compensation for the observation was made. The feasibility of substitution was analyzed. The results proved that this compensator can reduce the amplitude of the observation error and has a certain pre-correction effect on the phase lag of the state observation. This compensation idea was then extended to the nonlinear extended state observer. The simulation validates the effectiveness of the compensator to improve the observation precision and speed up the convergence of the error. The experimental results further indicate that the introduction of the compensator can significantly improve the control precision of the whole active disturbance rejection control system. Thus the feasibility of this kind of compensation idea gets proved.