The recently proposed signal space separation (SSS) method can transform the multichannel magnetic measurements of brain (MEG) into parts that correspond to inner sources and undesired external interferences. In this paper, we extend this method by decomposing the signal into deep and superficial regions. This is realized by manipulating the SSS coefficients using a scheme that exploits beamspace methodology. It relies on estimating a linear transformation which maximizes the power of the source space of interest over the power of remaining part. We demonstrate that this method yields a simple and direct way to decompose the signal into deep and/or superficial parts.