On the Construction of Channel Gain Map: Model-Based or Model-Free Approach?

Channel gain map (CGM) is a promising technique that enables the environment-aware communications by providing a priori channel gain information for users at arbitrary locations. The construction of CGM can be viewed as spatial channel prediction from the communication perspective, which motivates the channel model-based construction algorithm. On the other hand, it can also be viewed as spatial data interpolation from the geostatistic perspective, where a bunch of model-free methods can be adopted. This paper compares model-based and model-free approaches for CGM construction. We extend the existing model-based channel prediction by introducing a multi-mode channel model to address the spatial inconsistency issue in conventional models. The model-based methods are then compared with various model-free spatial interpolation methods, including a deep learning method. The results show that the channel model may introduce substantial bias for CGM construction in complex environments.