Abstract: Nonlinearity of power amplifier always results in spectrum spreading on input signal. When using digital predistortion technology to compensate power amplifier's nonlinearity, the required sampling frenquence for the output signal of amplifier is times of the original signal. According to the undersampling digital predistortion method based on Zhu Generalized Sampling Theorem, the amplifier output signal can be sampled as the sampling frequency of original baseband signal Nyquist rate. Using the undersampling digital predistortion method, the digital predistortion performance of non-memory power amplifier is satisfactory. But for amplifier with significant memory effect, the digital predistortion performance is worsened in comparison to amplifier with non-memory effect. Furthermore it is poorer than that of oversampling. Simulations show that the higher the sampling frequency, the better the digital predistortion performance. As for engineering application, if the requirements of digital predistortion performance are met, a sampling frequency should be chosen as low as possible for cost reduction.