Gaussian Mixture Model-Enabled Perturbed Lattice-Based Crowd Sensing for Radio Environment Mapping

In the upcoming AI-enabled B5G and 6G wireless networks, the network operations would heavily rely upon actual on-ground information regarding network resource utilization. This can be provided by a Crowd-Sensed Radio Environment Map (CSREM). In a CSREM, some of the idle devices are selected to assess the resource utilization and report it back to a central entity, which utilises this real-time information to construct a Radio Environment Map using an interpolation algorithm, typically Ordinary Kriging (OK) and takes resource decisions accordingly. In this work, spatial statistics have been utilized for the adequate selection of crowd sensors so that the entire region is covered by the least number of sensors and that the sensors are evenly distributed across the region. The sensors thus selected form a perturbed lattice structure. Additionally, the adopted variational Bayesian-based GMM-based Kriging (VK) interpolation algorithm utilises a more accurate GMM-based representation of the radio environment. Hence, the combination of the adopted perturbed lattice sensor deployment and VK interpolation technique results in a substantial accuracy improvement in interpolation performance as compared to the prevalent OK method.