Learning Power Allocation and Channel Estimation for Superimposed Pilot-Assisted Multiuser MIMO

Superimposed pilot (SIP) enhances the spectral efficiency of multiple-input multiple-output (MIMO) communication by superimposing pilots upon data at specific power levels, without reserving orthogonal resource for pilots. However, it is challenging to optimize the pilot-data power (PDP) allocation and address the nonlinear pilot-data interference (PDI), especially in multiuser scenarios. In this letter, we propose a deep learning (DL)-based SIP scheme, aiming to enhance data transmission rates of multiuser MIMO systems. Specifically, we develop a model-driven SIP network (SIPNet) for joint optimization of the PDP allocation and channel estimation, which spares the appropriate data power and gains the refined channel estimates. When deploying the well-trained SIPNet, we incorporate an iterative channel estimation and data detection (ICEDD) design for effective alternate PDI elimination. Simulation results show that SIPNet allocates more pilot power, staggered among users, in the middle orthogonal frequency division multiplexing (OFDM) symbols, and the proposed scheme with fewer iterations outperforms the traditional pilot scheme with exclusive orthogonal pilots and existing SIP scheme with ICEDD, especially under challenging channel conditions characterized by high mobility and low signal-to-noise ratio (SNR).