DC-Net: A Distant-range Content Interaction Network for Radio Map Construction

Abstract The Radio Map (RM) stands as a powerful tool in the field of wireless communication and radio regulation. Currently, several algorithms based on Convolutional Neural Networks (CNNs) have been proposed. However, due to the inherent limitations of vanilla convolutions, capturing distant-range information of the radio propagation environment becomes challenging, ultimately resulting in poor RM construction accuracy. Therefore, this letter introduces a novel RM construction model called Distant-range Content interaction Network (DC-Net) and validates its effectiveness using the radiomapseer dataset. Experiment results reveal that the proposed DC-Net achieves State-Of-The-Art (SOTA) performance in mapping accuracy for both known and unknown transmitter locations in RM construction. Additionally, the model demonstrates rapid and stable convergence during the training process, leading to reduced computational resource consumption.