Reinforcement Learning Based Load Balancing for Hybrid LiFi WiFi Networks

Abstract

Light fidelity (LiFi) is an emerging communication technology, which utilizes the lightemitting diodes (LEDs) for high-speed wireless communications. Due to its huge unlicensed bandwidth, LiFi is capable of supporting high data rates. The quality of the LiFi channel fluctuates across the room due to interference, reflection from walls or blockage. On the other hand, WiFi is another wireless communication technology that is capable of providing moderate data rates with ubiquitous coverage. As the electromagnetic spectrum of LiFi does not overlap with WiFi, both of them can coexist to form a hybrid LiFi and WiFi network for seamless and high-throughput connectivity. The performance of a hybrid system significantly depends upon the access point (AP) assignment and resource allocation strategies. In this paper, a downlink hybrid system with one WiFi AP and four LiFi APs is considered, and a reinforcement learning (RL) algorithm is implemented in order to determine an optimal AP assignment strategy, which maximizes the long-term system throughput while ensuring the required users fairness and satisfaction. Furthermore, two different scenarios based on the random waypoint model with uniform and non-uniform distribution of users have been studied. The performance of the proposed system is compared against state-of-the-art benchmark approaches e.g., signal strength strategy (SSS), exhaustive search, and an iterative optimization method. The results are reported in terms of the average system throughput, user satisfaction, fairness, and capacity outage probability. It is shown that the proposed RL method performs closer to the exhaustive search scheme at fairly low complexity. The RL method also outperforms the SSS scheme and the iterative algorithm in most scenarios INDEX TERMS Hybrid LiFi WiFi, Light Fidelity (LiFi), Load balancing, Reinforcement learning (RL), Trust region policy optimization (TRPO).

Hybrid LiFi and WiFi Networks: A Survey

“…Instead, it is feasible to use machine learning to cope with the uncertainties in network deployment, user distribution, traffic situations, etc. A LB method based on reinforcement learning was introduced in [115]. It is shown that this method can outperform iterative algorithms in most scenarios.…”

Hybrid NOMA-Based ACO-FBMC/OQAM for Next-Generation Indoor Optical Wireless Communications Using LiFi Technology

“…Additionally, the NLOS component is the superposition of all non-LOS components that are due to one or more reflections at the wall surfaces. The frequency dependence NLOS optical impulse response for a room corresponds to a first-order low-pass filter with transfer function is given by [38] [39]…”

Towards Terabit LiFi Networking

“…In recent years, machine learning (ML) techniques have been studied for solving NP-hard optimization problems with different contexts (Sun et al, 2019;Zhu et al, 2020). For instance, some of the previous works in the literature related to the optimization problem formulated in this paper (Elgamal et al, 2021;Shrivastava et al, 2020;Ahmad et al, 2020) consider the implementation of reinforcement learning (RL) to solve various optimization problems in optical and RF systems where RL can interact with an environment to learn an optimal policy that makes right decisions. Specifically, in (Elgamal et al, 2021), reinforcement learning is used for assigning each user to an exclusive wavelength in a WDMA-based optical wireless network.…”

“…In (Shrivastava et al, 2020), a deep Q-network (DQN) learning-based algorithm is proposed for solving an optimization problem that aims to maximize the sum rate of a hybrid optical/RF network through allocating power and bandwidth in addition to user association. In (Ahmad et al, 2020), a RL-based load balancing approach is proposed for maximizing the sum rate of users in a hybrid LiFi/ WiFi network. It is shown that ML techniques can provide sub-optimal solutions while avoiding complexity.…”