Most of previous researches in Multiple-Input- Multiple-Output (MIMO) downlink beamforming based on uplink-downlink duality assume sum power constraints. However, the powers of antennas are limited individually in realistic implementations. Per-antenna power constrains are considered in this paper. In the uplink-downlink duality, it is shown the noise power at uplink receive antennas are dual to the weights of downlink transmit antenna powers. Therefore, the noise vector can be applied to adjust transmit antenna powers. Uplink-downlink duality and the noise vector controlled scheme are adopted to update beamformers to optimize the SINR vector. Though the problem is nonconvex, the proposed algorithm could converge to a sub optimal local solution. Simulations show the algorithm effectively control antenna powers during the optimization.