Robust Representation of Spatial Sound in Stereo-to-Multichannel Upmix