This paper presents a novel digital predistortion (DPD) scheme based on indirect learning architecture (ILA), where only the in-phase (I) or quadrature (Q) component is needed to demodulate. To facilitate the DPD implementation, the power amplifier's (PA) complex baseband output, free from measurement noise, is first recovered by a PA forward modeling by acquiring only the I component of the PA's output. Then the post-inverse of the PA, used as the predistorter, can be identified utilizing the estimated PA output. Compared with the classical ILA scheme, both the theoretical analysis and the tested results manifest that the proposed method can not only reduce the cost significantly, but also achieve superior linearization performance, as the measurement noise can be mitigated effectively.