Precoder Learning for Weighted Sum Rate Maximization
Mingyu Deng, Shengqian Han
TL;DR
This work tackles weighted sum-rate maximization (WSRM) in multiuser MIMO downlinks by designing a deep neural network that respects joint unitary and permutation equivariance, enabling efficient precoder learning with UE weights. Linear weight formulations fail to capture the weight-induced policy, so the authors introduce a nonlinear weight function learned via a dedicated Edge-GNN sub-network that operates on pairwise interaction features $E_1=h_i^H x_j^{(l-1)}$ and $E_2=\alpha_i h_i^H x_j^{(l-1)}$, while preserving the symmetries. The proposed approach demonstrates superior learning and generalization performance compared with baselines (WMMSE, Model-GNN, Edge-GNN, Graph-transformer, EGAT) and achieves lower training and inference complexity. The work advances practical precoder learning for fair, low-latency downlink transmissions by embedding structural priors directly into the network design.
Abstract
Weighted sum rate maximization (WSRM) for precoder optimization effectively balances performance and fairness among users. Recent studies have demonstrated the potential of deep learning in precoder optimization for sum rate maximization. However, the WSRM problem necessitates a redesign of neural network architectures to incorporate user weights into the input. In this paper, we propose a novel deep neural network (DNN) to learn the precoder for WSRM. Compared to existing DNNs, the proposed DNN leverage the joint unitary and permutation equivariant property inherent in the optimal precoding policy, effectively enhancing learning performance while reducing training complexity. Simulation results demonstrate that the proposed method significantly outperforms baseline learning methods in terms of both learning and generalization performance while maintaining low training and inference complexity.