LAUSR

Publishing measurement results without elaborating on the uncertainty included is a common practice in the wireless community. Therefore, it is not surprising that the application of state-of-the-art variance reduction techniques to wireless research is rarely seen. In this paper, we discuss, exemplarily, the benchmarking of the average link capacity of a $4\times 4$ multiple-input multiple-output (MIMO) downlink resulting from different base station (BS) antenna configurations. We show that the capacity of this downlink is not only dependent on the antenna spacing (the effect we want to measure), but also on the position of the antennas (an effect we need to control for). Unfortunately, once we change the antenna spacing, we cannot avoid changing the position of the antennas as well. We show that by applying the variance reduction technique of randomizing BS antenna positions, we get more accurate results. We further show that without increasing the effort, our favored method of matching completely eliminates the systematic error of position dependence. Both methodologies can also be applied to throughput measurements and larger antenna arrays, especially to virtual antenna array measurements for massive MIMO.