Abstract
Smart building (SB) technology integrates diverse systems across multiple layers to sense, process, and act upon environmental data. As urbanization increases the demand for connected nodes, traditional transmission techniques face congestion and inefficiency. This paper proposes a 6G-enabled framework incorporating indoor visible light communications (VLC) and machine learning (ML) to enhance SB intelligence. We envision a framework with reliable data transmission and powerful reasoning capabilities, supported by simulation results demonstrating high localization accuracy.
Introduction
The rapid expansion of Smart Buildings faces two primary technical hurdles: connectivity and intelligence. Traditional Radio Frequency (RF) systems like WiFi are increasingly congested and vulnerable to security breaches. Simultaneously, the sheer volume of data generated by thousands of sensors requires more than traditional processing; it requires autonomous, real-time decision-making.
To address these, we propose leveraging 6G technologies. Indoor VLC offers a high-bandwidth, secure alternative to RF by using existing LED lighting infrastructure. When paired with ML, these systems can transform "big data" into actionable intelligence, creating a "brain" for the building. This relationship is reciprocal: VLC and ML enable the SB, while the SB provides the stable, controlled environment necessary for these technologies to thrive.
Proposed Architecture
The framework is structured into a seven-layer functional architecture designed to manage the lifecycle of data from sensing to service execution.
The Sensing and Network Layers utilize VLC and Power Line Communications (PLC) for secure, high-speed data transport. The Semantic and Software Layers employ ML to interpret user inputs, such as voice or gestures. Higher-level Processing and Reasoning Layers use ML algorithms to filter data redundancy and perform complex logic, finally culminating in the Service Layer where actuators like smart lights or HVAC systems execute commands.
Case Study: Indoor Localization
To validate the framework, we conducted a case study on indoor localization using a simulated dataset from the CandLES (Communication and Lighting Emulation Software) platform. The simulation modeled a 10m x 10m x 3m room with a grid of 99x99 receivers collecting Received Signal Strength (RSS) data. We compared three configurations: a 4-LED setup (VLC-4), an 81-LED setup (VLC-81), and a standard 4-access-point WiFi setup. Three ML models—Support Vector Machines (SVM), Neural Networks (NN), and K-Nearest Neighbors (KNN)—were tested across 50 simulation runs.
Results and Discussion
The results highlight the significant advantages of VLC over traditional WiFi for indoor positioning. The VLC-81 system achieved an accuracy rate exceeding 95% for a Prediction Error Distance (PED) of 0.3 meters.
Even the minimal VLC-4 setup using the KNN algorithm outperformed the WiFi baseline. A key insight is that the high directionality of light creates more distinct signal signatures at different locations compared to the omnidirectional nature of RF, which directly enhances ML-based localization accuracy. Furthermore, while performance varied at high precision, the results for different ML models converged as the allowed error distance increased, indicating that even simpler models are effective for general room-level tracking.
Conclusion and Future Work
While VLC and ML provide a robust foundation for SBs, challenges remain. Managing interference between multiple LEDs requires the development of Heterogeneous Networks (HetNets) that coordinate VLC, RF, and PLC. Additionally, the high observability of residents' movements necessitates strict data protection legislation. Future research will also explore SB-Edge-Cloud architectures to balance the latency and computational demands of massive ML tasks.
Summary updated