In this paper, a new learning scheme named Discrete Time Virtual Feedback Error Learning (DT-VFEL) is proposed. Even if the controlled systems with high relative degree are minimum phase continuous-time system, by sampling at a fast rate, non-minimum phase discrete-time systems can often appear. Therefore, it is difficult to extend VFEL from the continuous-time version to the discrete-time one, because the stable inverse system can not be learned in conventional Discrete Time Feedback Error Learning (DTFEL) schemes. To overcome the drawback, 2-Delay input control is applied. And variable-gain-type learning law is adopted to shorten the time of learning. Finally, the paper shows that the tracking error converges to zero under a PE condition of the discrete-time version, and verifies its effectiveness based on simulation.