Blind source separation (BSS) has many applications: sound scene analysis, speech recognition, medical signal processing, etc. However, most of these applications concern the temporal separation of signals. Studies have shown the effectiveness of separation in the ambisonic domain with spherical microphone recordings. Thanks to the ambisonic approach, it is possible to separate the directions of arrival of the sources. As such, BSS becomes a promising tool for sound field reproduction with loudspeakers arrays (WaveField Sythesis or Higher-Order Ambisonics). Thanks to spherical microphone arrays and Ambisonics principle, both spatial and temporal information are available. Therefore, it would be possible to reproduce the individual sound field of each separated source. Thus, one can remove a given source from a recording and reproduce the remaining sound field. The main objective of this work is to reproduce the sound field of one of the captured sources by removing the rest of it (sources or noise). The first part of the paper presents the methods for the BSS and corresponding sound field reproduction. The second part presents simulation results and investigates effect of measurement noise, spatial source separation, and reflection. [Work supported by NSERC Discovery grant.]