The ability to form an image without using spatial encoding gradients allows for cheaper MRI manufacturing costs, particularly in small low field devices suitable for point-of-care imaging. The Bloch-Siegert shift effect provides an RF alternative to spatial encoding with gradient coils. This work presents a new approach to evaluate and optimize nonlinear RF encoding schemes generated by transmit phased arrays. From thousands of possible encoding patterns, the algorithm finds the set that minimizes correlations between spatial patterns while maximizing the spatial encoding. Simulated images demonstrate that the nonlinear spatial encodings provided by Bloch-Siegert RF pulses can provide high image quality.