IoT-Enabled Privacy-Preserving Indoor VLC Localization With Scale-Invariant AOA and Federated Learning

Abstract

This paper presents an Internet-of-Things (IoT)-enabled indoor localization framework for visible light communication (VLC) in smart environments. The method combines scale-invariant angle-of-arrival (AOA) estimation, temporal densification, and federated learning (FL) so distributed clients can infer position without centralizing raw trajectories. At the front end, an angle-diversity receiver (ADR) estimates AOA from calibrated photodiode gains normalized under the intensity modulation/direct detection (IM/DD) Lambertian model, reducing sensitivity to unknown light-emitting-diode (LED) power scaling under dominant line-of-sight (LOS) conditions. A reliability score that combines gain consistency and received-power support guides anchor selection, while a temporal convolutional network–long short-term memory (TCN-LSTM) module reconstructs sparse or irregular AOA features. The resulting features are used by a regression model trained with Federated Averaging (FedAvg). Monte Carlo simulations in a multi-LED indoor scenario show that, at 20 dB optical signal-to-noise ratio (SNR), the proposed pipeline reduces mean absolute error from 17.0 cm to 9.0 cm relative to FL without densification. All results are simulation-based; no hardware-testbed measurements are claimed.