This article presents the design, the implementation, and the perceptual evaluation of a multichannel loudspeaker array for the creation of 3D virtual sound environments. Dedicated to perceptual and postural studies, this system consists of 42 loudspeakers equally distributed around the listener on a full sphere (3 meter diameter), allowing to synthetize sounds moving within the whole 3D space. The spatial distribution of the loudspeakers makes it possible to use two different 3D rendering approaches: a physical approach through the use of 3D Higher Order Ambisonics (HOA) method up to fifth order and a perceptual approach through the use of Vector-Based Amplitude Panning (VBAP) method. A localization experiment was carried out in order to investigate the performance of the present system in terms of localization accuracy. Furthermore, this experiment aimed to explore the perceptual impact of the rendering methods (e.g. HOA and VBAP), and to evaluate the improvement of localization performance with respect to the ambisonic order (from third to fifth order). Differences between rendering methods are investigated in terms of azimuth and elevation localization errors, and front-back and up-down confusions. Finally, virtual sound source localization is compared to real sound source localization generated by 42 loudspeakers.