In this paper, we study a single-user multiple-input single-output (MISO) system with constant envelope (CE) transmission. To enable the nonlinear mapping from a fixed receiver signal constellation to the transmitter CE signal vectors, the availability of perfect channel state information at the transmitter (CSIT) is assumed in existing literature. However, traditionally, CSIT needs to be acquired at the cost of additional channel training and feedback overhead, which increases with the number of transmit antennas. In this paper, we propose a novel adaptive online signal constellation design for MISO CE transmission with significantly reduced training time and feedback complexity compared to the traditional training with fixed constellation set. Numerical results show that our proposed scheme outperforms the traditional scheme in terms of average throughput and yet with less training time required.