Phase gradient autofocus (PGA) algorithm can be used to estimate and correct phase errors effectively in echo signals of synthetic aperture lidar. The wavelength of laser is three to four magnitudes shorter than that of the microwave, so the vibration of experimental platform can bring great phase errors during the process of synthetic aperture lidar (SAL) imaging, which results in the reduction of imaging quality. To improve the image quality, a PGA algorithm with an improved method of windowing compared to the conventional PGA method is applied in SAL and is experimentally verified by an SAL demonstrator on a 64 m-distance-target to not only realize the accuate compensation to the phase errors in the defocusing data, but also have fewer iteration times and the same compensation effect.