Thesweet spot can be interpreted as the region where acoustic sources create a spatial auditory illusion. We study the problem of maximizing this sweet spot when reproducing a desired sound wave using an array of loudspeakers. To achieve this, we introduce a theoretical framework for spatial sound perception that can be used to define a sweet spot, and we develop a method that aims to generate a sound wave that directly maximizes the sweet spot defined by a model within this framework. Our method aims to incorporate perceptual principles from the onset and it is flexible: it imposes little to no constraints on the regions of interest, the arrangement of loudspeakers or their radiation pattern. However, the perceptual models must satisfy a convexity condition, which is fulfilled by state-of-the-art monaural perceptual models, but not by binaural ones. Proof-of-concept experiments show that our method, when implemented with van de Par's monaural model, outperforms state-of-the-art sound field synthesis methods in terms of their binaural azimuth localization and binaural coloration properties.