Transformer-Based Site-Specific Channel Estimation

The rapid increase in the number of mobile devices and the diversity of service requirements have escalated complexity in wireless networks. This has fueled the need for seamless wireless access technologies that can support high data rates across heterogeneous devices. To address these challenges, massive multiple-input multiple-output (MIMO) systems have been considered in the 5G and upcoming 6G cellular systems. However, the independent channel estimation process for each antenna element for all base stations (BSs) is impractical because it requires a substantial pilot overhead. In response to this limitation, we propose a transformer-based site-specific channel estimation algorithm to infer the channel of a BS based on the UE position or the channels of other BSs, without pilot transmission. The algorithm effectively captures the nonlinear characteristics of channels using a multi-head attention (MHA) module. We evaluated the performance using a channel dataset collected from a 3D ray tracing environment replicating a large-scale real-world city, which includes multiple BSs and UEs. The proposed algorithm successfully depicts the channel characteristics of the communication environment.