Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Q-Learning for 3D Coverage in VCSEL-based Optical Wireless Systems

Hossein Safi, Rizwana Ahmad, Iman Tavakkolnia, Harald Haas

TL;DR

The paper tackles the challenge of maintaining reliable 3D indoor coverage in VCSEL-based optical wireless networks where receiver height causes non-optimal static beam divergence. It proposes a model-free reinforcement learning approach, specifically Q-learning, to map discrete receiver heights to optimal divergence settings, eliminating the need for explicit channel modeling or exhaustive search. Simulation results show the method achieves near-optimal coverage (up to 92% at low heights) with significantly reduced online computation and robust performance under varied conditions, while also highlighting limitations of single-beam-per-user and the potential for multi-beam extensions. The work demonstrates a scalable, real-time, energy-efficient beam-control paradigm suitable for dense VCSEL array deployments in future 6G OWC systems.

Abstract

Beam divergence control is a key factor in maintaining reliable coverage in indoor optical wireless communication (OWC) systems as receiver height varies.Conventional systems employ fixed divergence angles, which result in significant coverage degradation due to the non-convex tradeoff between optical power concentration and spatial spread. In this paper, we introduce a reinforcement learning (RL)-based framework for dynamic divergence adaptation in vertical-cavity surface-emitting laser (VCSEL)-based OWC networks. By continuously interacting with the environment, the RL agent autonomously learns a near-optimal mapping between receiver height and beam divergence, thereby eliminating the need for analytical modeling or computationally intensive exhaustive search. Simulation results demonstrate that the proposed approach achieves up to 92% coverage at low receiver heights and maintains robust performance under challenging conditions, enabling scalable, real-time, and energy-efficient beam control for dense VCSEL array deployments in next-generation OWC systems.

Q-Learning for 3D Coverage in VCSEL-based Optical Wireless Systems

TL;DR

The paper tackles the challenge of maintaining reliable 3D indoor coverage in VCSEL-based optical wireless networks where receiver height causes non-optimal static beam divergence. It proposes a model-free reinforcement learning approach, specifically Q-learning, to map discrete receiver heights to optimal divergence settings, eliminating the need for explicit channel modeling or exhaustive search. Simulation results show the method achieves near-optimal coverage (up to 92% at low heights) with significantly reduced online computation and robust performance under varied conditions, while also highlighting limitations of single-beam-per-user and the potential for multi-beam extensions. The work demonstrates a scalable, real-time, energy-efficient beam-control paradigm suitable for dense VCSEL array deployments in future 6G OWC systems.

Abstract

Beam divergence control is a key factor in maintaining reliable coverage in indoor optical wireless communication (OWC) systems as receiver height varies.Conventional systems employ fixed divergence angles, which result in significant coverage degradation due to the non-convex tradeoff between optical power concentration and spatial spread. In this paper, we introduce a reinforcement learning (RL)-based framework for dynamic divergence adaptation in vertical-cavity surface-emitting laser (VCSEL)-based OWC networks. By continuously interacting with the environment, the RL agent autonomously learns a near-optimal mapping between receiver height and beam divergence, thereby eliminating the need for analytical modeling or computationally intensive exhaustive search. Simulation results demonstrate that the proposed approach achieves up to 92% coverage at low receiver heights and maintains robust performance under challenging conditions, enabling scalable, real-time, and energy-efficient beam control for dense VCSEL array deployments in next-generation OWC systems.
Paper Structure (16 sections, 17 equations, 3 figures, 3 tables, 1 algorithm)

This paper contains 16 sections, 17 equations, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model
  3. Geometric Configuration and Transmitter Array
  4. Gaussian Beam Propagation and Received Irradiance
  5. Receiver Front-End Model and Final Received Power
  6. Optimization Framework for Divergence Angle Control
  7. Problem Formulation
  8. Analytical Challenges and Dynamic Nature
  9. Reinforcement Learning Formulation
  10. Markov Decision Process Formulation
  11. Q-Learning Framework
  12. Implementation and Convergence Behaviour
  13. Simulation Results and Discussion
  14. Convergence Behavior and Learning Stability
  15. Results and Learned Divergence Policy
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Configuration of VCSEL-Based Downlink Transmission
  • Figure 2: Coverage performance for $\theta_{\text{divergence}} = 15^\circ$ at different heights. Here, $\Gamma_{\text{th}} = 5$ dB, and the null spaces represent regions where SINR $< \Gamma_{\text{th}}$.
  • Figure 3: Convergence behavior of the proposed RL-based divergence control framework. The blue curve shows the evolution of the average coverage reward per episode (with a 50-episode moving average (MA)), while the red curve represents the exploration rate ($\epsilon$).