We propose a new scheme combining the calibration of the path imperfections and the compensation of HPA nonlinearity in the downlink OFDM smart antenna systems. We use a two term third-order polynomial (without second-order term) and the indirect learning architecture for calibration and compensation, to make each path of the antenna array have equal characteristics. We test our scheme with computer simulations. The result shows that, with the addition of only one third-order term, the adverse nonlinear effects as well as the those of linear imperfections can be effectively compensated.