Design of subband-level MMSE transceiver for multiuser MIMO-OFDM uplink systems and neural network implementation

Abstract

This paper presents two subband-level algorithms for the joint optimization of the precoder and decoder in multiuser MIMO-OFDM uplink systems, while considering a per-antenna power constraint. The two proposed algorithms formulate the joint design of MMSE transceiver as a non-convex, multi-variable coupled optimization problem under power constraint. This problem is modeled using the Lagrange multiplier method, and the analytical solution is obtained by continuously updating the gradient through a generalized iterative method. This method achieves a specific design of subband-level precoding matrix that balances performance, computational complexity and feedback overhead. Additionally, this paper proposes the implementations of the above algorithms in terms of neural network structures. The deep learning-based precoding algorithm significantly reduces computation complexity compared to the traditional iterative algorithms. Furthermore, the robustness of our proposed implementation can be greatly improved by adjusting the network structure and training dataset, and its performance gain can be comparable to or even better than those of the iterative algorithms. Finally, the link simulations are performed to verify the performance gains of our algorithms.