In this paper, a novel method based on the memetic algorithm for downlink Spatial Division Multiple Access (SDMA) optimization technique of smart antennas by is proposed. Via the phase-amplitude perturbation method, the radiation pattern formulas available for searching optimal solutions are deduced to search the optimal weighting vector of the array factor of a smart antenna. In a smart antenna system, one user's signal must be considered as other users' interfering signal. An optimal radiation pattern of a smart antenna can not only adjustably suppress interferers by placing nulls at the directions of the interfering sources but also provide a maximum main lobe in the direction of the desired signals at the same time, i.e., to maximize the Signal to Interference Ratio (SIR). In order to achieve this goal, a new convergent method referred as the two-way convergent method for memetic algorithms is proposed. The optimal radiation pattern design of downlink smart antennas can be achieve Spatial Division Multiple Access (SDMA) optimization by using the phase-amplitude perturbation method.