The operation of future radio systems will benefit from any available information about the radio environment, e.g., to better allocate radio resources, and to predict the radio conditions of users based on their locations. Thereby, radio environment maps (REMs), i.e., the information about the radio channel per location, can assist future radio systems. However, measuring large-area REMs with high spatial resolution results in enormous effort and it is more efficient to estimate REMs from sparse observations. In this work, we present a deep neural network (DNN)-based interpolation technique that is capable of recovering spatial fading patterns through interpolation by extracting position-dependent channel correlations. Our approach solely relies on the sparsely sampled REM that is to be interpolated. By systematically studying DNN structures and input features, we extract a favorable structure for the spatial interpolation of anisotropic environments. Based on a simulated indoor REM with varying fading structures, we demonstrate that our approach is superior to conventional methods in recovering spatial fading structures. Using reconstructed REMs for radio applications in future work will yield application-specific metrics to further assess the reconstruction quality.
Recovering High-Resolution Fading Patterns from Sparsely Sampled Indoor REMs
2024-06-24
1573539 byte
Conference paper
Electronic Resource
English
Texture recognition from sparsely and irregularly sampled data
British Library Online Contents | 2006
|Multi-Sensor Excision of Sparsely Sampled Nonstationary Jammers for GPS Receivers
British Library Conference Proceedings | 2015
|