MULTIANTENNA CHANNEL MAP ESTIMATION USING DEEP SPATIAL INTERPOLATION

Abstract

The radio maps of multiantenna channel state information (CSI) are constructed using deep learning. The desired CSI is predicted for arbitrary locations in a geographical area based on the measurements collected at sampling locations. Such maps can be used to significantly reduce the overhead associated with CSI acquisition. A novel deep architecture is proposed, consisting of an encoder/decoder pair for transforming high-dimensional CSI features to lower-dimensional embeddings, together with a deep embedding interpolator for exploiting the spatial dependency of the CSI. Two important problem classes are tackled in a unified fashion, namely, CSI interpolation and prediction. Practical scenarios involving missing information are also considered. The efficacy of the proposed methods is verified by numerical tests.