We propose a method for the optimal design of acquisition geometry in terms of imaging quality by improving the distribution of the wavenumber content of the approximate full waveform inversion gradient at a specified target. It has been successfully developed in 2D and we present here the extension to the 3D case. A study through diffraction tomography shows that for a regular layout of the acquisition, the distribution of the wavenumbers is not regular and for a fixed maximum offset those wavenumbers will vary inside an envelope which can be approximated by a spherical dome. We are interested in finding the positions of sources and receivers that give a regular wavenumbers sampling inside this envelope. As in 2D, we express this problem as an instance of Centroidal Voronoi Tessellation. By minimizing a newly designed objective function we can find acquisition geometries that improve our regularity criterion. It is also possible to add constraints on the deployment area and take into account the effects on the illumination from the heterogeneity of the initial model. We believe that those new developments open up new perspectives for target-oriented full waveform inversion and optimal post-acquisition data selection.