Nowadays 3-D reconstruction for natural scene is an important aspect of the Earth observation with SAR Tomography (TomoSAR). The general method for pattern sidelobe suppression, during the tomographic processing, is usually implemented with the weighting of uniform linear array, however at the expense of mainlobe broadening. In this paper, a minimax optimization model for cells configuration is constructed based on non-uniform linear array, to achieve the optimal peak sidelobe ratio with a fixed mainlobe width for any beam direction in the range of perspectives. For this, an objective function rasterisation is proposed and then the optimal solution can be figured out by the Sequence Quadratic Programming (SQP) with differential evolution for its unique ability of global memory. Finally, experimental results with PolSARpro polarimetric TomoSAR data validate the effectiveness of the proposal for natural scene 3-D image reconstruction. ©, 2015, Science Press. All right reserved.