Abstract: An accurate and reliable massive MIMO channel model is crucial for supporting design and performance evaluation of such systems in the future. However, massive MIMO channel sounding systems are cost prohibitive and complicated, where a large number of antenna elements and associated radio frequency transceiver chains are needed. Virtual large-scale arrays have been extensively utilized as an alternative for massive MIMO channel characterization. In this paper, we investigate virtual large-scale array systems formed by repositioning a real subarray system for channel characterization. With this scheme, we have the flexibility to scale between system cost and system channel sounding capability. Based on this scheme, general beampatterns of subarrays and subarray-based virtual large-scale arrays are derived, based on measured complex antenna patterns of antenna elements on the subarray. Three real subarray antenna systems, i.e., a 16-element uniform rectangular array (URA) at 3.5 GHz, an 8-element uniform circular array at 3.5 GHz, and an 8-element uniform linear array (ULA) at 10 GHz, were utilized to form the respective large-scale virtual arrays, i.e., a 128-element URA, a 48-element uniform cylinder array, and a 64 element ULA, respectively. Beamforming analysis based on the measured complex radiation patterns of the real arrays is provided to demonstrate that virtual large-scale arrays can significantly improve the capability of multipath parameter detection compared with the subarrays. Therefore, such cost-effective systems are promising for characterization massive MIMO propagation channels.