Physics- Informed Machine Learning Models for Indoor Wi-Fi Access Point Placement

Abstract

One of the main challenges in optimally placing indoor Wi-Fi access points in a complex indoor environment is the estimation of the received signal strength (RSS) given different access point locations. This paper proposes a deep learning approach, a modification to the classic Deep Convolutional Generative Adversarial Network (DCGAN), to generate accurate power maps for a specific indoor geometry. It has been demonstrated that this model consistently outperforms a benchmark ray-tracing simulator in efficiency, maintaining a comparable accuracy.