Distributed Multiuser Precoding via Unfolding WMMSE for Multicell MIMO Downlinks

Multiuser precoding design in multicell-cooperation multiple-input multiple-output (MU-MIMO) systems is a challenging task due to the inevitable information exchange among cooperative base stations (BSs). Based on the productive deep unfolding technique, this paper proposes a multicell-cooperation iterative framework, namely an unfolding synchronous distributed weighted mean-squared-error minimization (USD-WMMSE), to enable distributed and synchronous precoding calculations at the BSs. We develop the USD-WMMSE framework by employing the best response (BR) parallel iteration to decompose the coupling WMMSE precoding in multicell-cooperation MU-MIMO systems into a synchronous distributed scheme. The framework is unfolded into a sequential of neural networks with the iteration step sizes as trainable parameters to accelerate the convergence via unsupervised training. The convergence behavior of the USD-WMMSE is also theoretically proved. Simulation results validate that the proposed USD-WMMSE approaches the sum rate of the centralized WMMSE method with a computational complexity reduction from <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${\mathcal { O}}(B^{3})$ </tex-math></inline-formula> to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${\mathcal { O}}(B^{2})$ </tex-math></inline-formula> with respect to the number of cooperative BSs, B. The proposed USD-WMMSE also exhibits generalizability to various tested scenarios such as different transmit power and iteration numbers.