To suppress the systematic errors of the X-ray pulsar navigation system (XPNAV) based on the phase and Doppler frequency shift, and eliminate singularity phenomena, a double-differenced autonomous navigation method is proposed. By considering the intrinsic spin phase and time propagation delay of the pulsar, the observation pulse phase models on the spacecraft and Mars were established. To linearize the measurement model, a phase-differenced model was developed by subtracting between the two phase models. To suppress the systematic errors in the phase-differenced model, a double-differenced model based on phase and Doppler frequency shift is proposed, which is the difference between the phase-differenced model at the neighbor filtering time. For a deep-space spacecraft, the magnitude of the systematic error is significant but changes slowly over the filtering period. Therefore, the double-differenced model can suppress the majority of systematic errors. In order to solve the problem of correlation between process and measurement noise, an improved extended Kalman filter (IEKF) method is proposed. The simulation results reveal that the proposed method can effectively suppress the impact of systematic errors, including the pulsar direction error and linearization error.