Feasibility of deep reinforcement learning for standard-compliant joint OFDMA and MU-MIMO scheduling

Ascertain whether and how the deep reinforcement learning-based approach proposed by Noh et al. can effectively explore the combinatorial user-group and resource-unit assignment space while satisfying the user–resource-unit association constraints defined by IEEE 802.11ax.

Background

Deep reinforcement learning has been proposed for joint OFDMA and MU-MIMO optimization. However, the joint problem involves a vast combinatorial search space and intricate 802.11ax user–RU association rules (e.g., each user at most one RU, MU-MIMO only on certain RUs), making it unclear whether such methods can both scale and guarantee compliance.

Clarifying this issue would help assess DRL’s suitability for practical 802.11ax scheduling and guide the design of constraint-aware learning architectures or hybrid optimization-learning methods.