Efficient Modeling of Indoor Electromagnetic Signal Propagation in NLOS Environments Using a Gated MLP

Accurate modeling of indoor electromagnetic signal propagation is crucial for location-based services (LBS) in complex electromagnetic environments. However, Non-Line-ofSight (NLOS) conditions and multipath fading introduce significant noise to Received Signal Strength Indication (RSSI) data, making real-time and high-precision modeling a challenge. Traditional fingerprinting methods like K-Nearest Neighbors (KNN) suffer from high computational latency, while standard regression models often fail to capture high-dimensional nonlinear features. To address this, this paper proposes an Enhanced Gated Multilayer Perceptron (Gated MLP) for efficient indoor signal modeling. We introduce a learnable gating mechanism that adaptively weights input features, effectively suppressing noise from unstable Access Points (APs). Extensive experiments on the UJIIndoorLoc dataset compare the proposed method against six baseline algorithms, including Random Forest, XGBoost, and SVR. Results demonstrate that the Gated MLP achieves a high determination coefficient (R-squared) of 0.9567, comparable to state-of-the-art ensemble methods, while reducing inference latency to 0.001s-approximately 5000 times faster than KNN. This work provides an optimal trade-off between precision and computational efficiency, making it suitable for real-time deployment on resource-constrained embedded devices.