Deep-Learning-Guided Particle Swarm Optimisation for Multi-Objective Indoor Access Point Placement

Indoor access point (AP) placement critically determines coverage, throughput, and traffic fairness in wireless local area networks, yet obstacle-aware propagation and inter-AP load balance are rarely optimised jointly. This paper presents a deeplearning-guided particle swarm optimisation (PSO) framework for multi-objective indoor AP placement in IEEE 802.11ac networks. A ray-casting method extracts the euclidean distance and type of obstacle counts (corridor wall, partition wall, door) crossing between every candidate AP position and host, and two convolutional recurrent deep learning (DL) models convert these features into predicted received signal strength (RSS) and throughput without the cost of simulation. A three-tier fitness function guides PSO to jointly maximise worst-case user quality of service, mean network throughput, and inter-AP load balance, the last enforced through a scale-invariant load-variance penalty and reported via Jain’s Fairness Index. To keep the search computationally efficient, the building-wide candidate grid is pre-selected to 600 representative positions using a coveragebased selection procedure before the deep learning models and PSO are involved. The framework is benchmarked against a greedy sequential baseline using spatial heatmaps, cumulative distribution functions (CDF), and a statistical validation table. PSO-optimised placements exceed the baseline at every density, with fitness gains of 2.8 to 37.5%, a worst-case throughput improvement of up to 360.7% at K = 2, and a load-variance reduction of up to 92.0% at K = 6