A Transformer-Based Network for Unifying Radio Map Estimation and Optimized Site Selection
Yi Zheng, Cunyi Liao, Ji Wang, Shouyin Liu
Abstract
Signal quality and coverage area in cells are greatly impacted by the location of base stations (BS). BS deployment relies on high-precision radio maps (RMs). This study proposes the Optimized Site Selection Network (OSSN), an end-to-end Transformer-based deep learning model for quickly recommending the best BS location of given candidate points to maximize the signal coverage. Utilizing a building map and any candidate points for BS deployment, OSSN rapidly computes the best BS location among these candidates. OSSN contains a visual Transformer-based feature fusion module for integrating a building map and given BS, an information fusion pyramid module for reconstructing a high-precision RM, and a recommendation module for directly finding the best BS location among the given candidates. Meanwhile, a knowledge distillation learning method is introduced to train the parameters of the recommendation module. The results show that OSSN can significantly reduce the runtime of the BS siting.