Coverage aware access point placement for hybrid WiFi and LiFi indoor networks via MHCP guided constrained soft actor critic

The growing congestion of Radio-Frequency (RF) spectrum has made hybrid Wireless Fidelity/Light Fidelity (WiFi/LiFi) indoor networks an attractive solution for high-capacity wireless access. A central design challenge is determining the minimum number of LiFi access points (APs) needed to complement existing WiFi coverage while satisfying data-rate, illuminance, and overlap constraints. Prior work has addressed this problem as a Mixed-Integer Nonlinear Program (MINLP) and proposed an open-loop Matérn Hard-Core Point process (MHCP)-based finite-termination heuristic; however, such heuristic methods require re-solving the combinatorial problem for each new environment, limiting their scalability and adaptability. In this paper, we formulate LiFi AP placement as a Constrained Markov Decision Process (CMDP) and propose a closed-loop, learning-based framework: MINLP-guided Constrained Discrete Soft Actor-Critic (CDSAC). A MHCP is first employed to regularize the action space by generating spatially diverse candidate AP locations, mitigating combinatorial explosion. A discrete SAC agent then learns a deployment policy that maximizes coverage and data rate while satisfying constraints via a Lagrangian primal-dual method. Once trained, the policy adapts to unseen environments without re-solving the underlying optimization problem. Simulation results demonstrating superior performance over the reference MHCP-based heuristic, greedy coverage expansion, random sampling, and exact ILP baselines in both small- and large-scale indoor environments. The proposed method delivers 100% data-rate coverage while requiring fewer APs, highlighting the advantages of learning-based over open-loop heuristic placement.