Non-Orthogonal Pilot Design Exploiting Spatial Non-Stationarity for TDD XL-MIMO Transmission

Extremely large-scale multiple-input multiple-output (XL-MIMO) is a fundamental technology for 6G. The surge in the number of base station (BS) antennas has introduced new characteristics such as spatial non-stationarity in the wireless channel, which means users at different locations probably ‘see’ different parts of the BS antenna array, namely their visible regions (VRs). The concept of VR has been widely employed to simplify the design of XL-MIMO transmission. However, the VR information has yet to be utilized in the design of non-orthogonal pilots. Therefore, we aim to leverage the VR information to reduce the pilot overhead. To this end, we first establish the relationship between the VRs of users and the spatial correlation of the corresponding channel. Subsequently, we formulate the non-orthogonal pilot scheduling problem for XL-MIMO systems with spatial non-stationarity. Assuming knowledge of all users' VRs, we propose a graph-based non-orthogonal pilot design approach called the two-stage graph partitioning (TSGP), which jointly exploits the VR and angle domain information. Numerical results demonstrate that the proposed TSGP algorithm can effectively reduce the pilot overhead without compromising sum rate performance compared to the conventional orthogonal training approach.