Impact of Environmental Granularity on CNN-Based Wireless Channel Prediction

Accurate wireless channel models are essential in design and optimization of wireless communication systems. Deep learning provides a promising approach for wireless channel modeling with the help of environmental information. One important application is satellite image-based path loss prediction, which has attracted much attention recently. For path loss prediction, environmental characteristics play a crucial role, with the most intuitive manifestation on images being coverage and resolution, referred to as <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">environmental granularity</i>. By adopting absolute measurement metric <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m/pixel</i>, this paper investigates the impact of environmental granularities on deep learning-based path loss prediction through extensive experiments. Results indicate that with increasing environmental granularity, network prediction error exhibits a non-monotonic change. When environment granularity is low, increasing image resolution helps reduce error. However, when environment granularity is high, further increasing resolution actually leads to reduced prediction accuracy. These insights offer guidance for designing deep learning based networks for wireless channel prediction.