In the ever-changing realm of wireless communications, ensuring uninterrupted connection is a formidable undertaking, particularly in hybrid networks that include Light Fidelity (LiFi) and Wireless Fidelity (WiFi) technologies. Utilising machine learning to enhance handover choices has emerged as a cutting-edge method, offering substantial improvements in network efficiency.
“A prediction-model-assisted reinforcement learning algorithm for handover decision-making in hybrid LiFi and WiFi networks” is a publication by Dayrene Frómeta Fonseca, Borja Genovés Guzmán, Giovanni Luca Mertena, Rui BIAN, Harald Haas and Domenico Giustiniano.
Breaking Down the Handover Challenge
Hybrid LiFi and WiFi networks (HLWNets) provide different challenges because of the varying coverage regions of LiFi and WiFi access points. LiFi, a technology that utilises visible light to transmit data, often has a reduced coverage area in comparison to WiFi. As users navigate across these networks, they must often switch between access points (APs), which results in higher latency and the possibility of service disruptions.
Traditional Handover Techniques
Traditional handover techniques, such the ones used in LTE (Long-Term Evolution) networks, make use of the hysteresis concept. This entails deferring the choice to transfer control to another access point for a certain duration to prevent the occurrence of the “ping-pong” phenomenon, when the user repeatedly switches between access points. Although this approach decreases needless transfers, it lacks efficiency for the complex and often changing conditions of HLWNets.
Fig. 1: Considered network topology, where C1, ∈ {1,2,…, 16), is the atto-cell created by the LiFi AP Li, and W1 stands for the WiFi AP
A Novel Machine Learning Approach
The paper presents a novel handover technique named RL-HO, an acronym for Reinforcement Learning Handover. This methodology integrates reinforcement learning (RL) with a classification model to forecast user paths and adjust handover choices appropriately. This is the process:
- Prediction Model: The categorisation model utilises real-time data, namely signal-to-noise ratio (SINR) measurements from all accessible access points (APs), to forecast the trajectory of the user’s movement. The system can determine if the user is approaching the middle of a cell, traversing its boundary, or advancing towards a wall.
- Reinforcement Learning: The RL algorithm utilises information about the present state of the network, the expected path of the user, and the user’s quality of service to make choices. It can adjust and respond to new situations, using prior experiences to improve future transitions.
Fig. 2: Block diagram of the proposed RL-HO scheme for HLWNets.
Simulation Results and Performance
The RL-HO system was compared to both the regular LTE handover and a smart handover algorithm. The outcomes are remarkable:
- The RL-HO algorithm resulted in a 146% increase in network throughput compared to the regular LTE scheme and a 59% increase compared to the smart handover strategy.
- The new approach greatly decreased both horizontal and vertical handover rates, especially for users who are moving quickly. At a user speed of 3 m/s, RL-HO significantly decreased the vertical handover rate by 54% and the horizontal handover rate by 85% compared to the normal LTE technique.
- The RL-HO algorithm demonstrated superior flexibility to different user speeds and network circumstances, sustaining greater throughput even with an increasing number of users in the network.
Fig. 3: Rates of HHO and VHO
Fig. 4: Percentage of time he user is connected to WiFi, LiFi, or blocked along its trajectory
Fig. 5: Occurance rate of LiFi blockages
Fig. 6: Average throughput versus the user’s speed
Real-World Applications
Machine learning may revolutionise network handover management, especially in instances when a steady and fast connection is essential, according to this research. Smart home devices and appliances need a smooth connection to work properly. Office productivity, video conferencing, and large file transfers need reliable internet.
In industrial settings with increasingly automated equipment and robots, a robust wireless connection may improve productivity and safety. Airports, retail malls, and stadiums, which have numerous users and constant mobility, may benefit from enhanced handover procedures to provide a stable and high-quality connection.
By making the simulator code accessible, this work aids future research and development. This approach may help future researchers and developers enhance and tailor algorithms for specific conditions, advancing wireless communication technologies.
The use of machine learning in changeover decision-making improves hybrid LiFi and WiFi networks. The RL-HO system predicts and adjusts user movement in real time, improving user experience. This improves connection reliability and network resource optimisation. As we rely more on wireless communication, these advances will be essential for reliable network performance.