Coverage prediction has always been of great concern for mobile network operators. Yet the prevalent approach using analytical models assisted by drive testing based measurements is inherently inaccurate and expensive. We consider a promising alternative for coverage mapping involving crowdsourced measurements and spatial interpolation. In particular, we empirically study the accuracy of wide range of spatial interpolation techniques in different scenarios that capture the unique characteristics of crowdsourced measurements (inaccurate locations, sparse and non-uniform measurements, etc.), and find ordinary kriging to be a fairly robust technique.
展开▼
