TL;DR
This paper tackles the complex challenge of maximizing data rates in a 5G/6G network that combines three cutting-edge technologies: UAVs (drones), Visible Light Communications (VLC), and NOMA (Non-Orthogonal Multiple Access). The authors introduce a nature-inspired "Harris Hawks Optimization" (HHO) algorithm to solve the difficult math of drone placement and power allocation, then use that same algorithm to train a neural network for lightning-fast, real-time network management.
What Problem Does This Paper Address?
As we move toward 6G, we need "massive connectivity"—the ability to serve a huge number of devices simultaneously. Visible Light Communication (VLC) is a promising solution because it uses the unregulated optical spectrum (LEDs) to provide high-speed data. However, VLC is strictly limited by line-of-sight; if something blocks the light, the connection drops.
Mounting VLC transmitters on UAVs solves this by providing mobile, high-altitude "light bulbs" that can maintain clear lines of sight to ground users. To squeeze even more capacity out of this setup, the authors use NOMA, which allows multiple users to share the same frequency/time resource.
The catch? Jointly optimizing where the drone should hover and how much power to give each user is a "non-convex, NP-hard" problem. Standard engineering shortcuts (like Successive Convex Approximation) fail here because the mathematical model for how light spreads (the Lambertian emission model) is too complex for them to handle.
Methods
The authors propose a two-stage approach:
- HHOPAP (Harris Hawks Optimization for Power Allocation and Placement): They adapt the Harris Hawks Optimization algorithm—a swarm intelligence metaheuristic that mimics the cooperative hunting strategies of hawks (waiting, encircling, and pouncing). Unlike gradient-based methods, HHO doesn't need "smooth" math to work; it can navigate the "rugged" mathematical landscape of VLC channel models to find near-optimal drone positions and power levels.
- HHOFNN (HHO-trained Feedforward Neural Network): Direct optimization is slow. To make this work in the real world, the authors use HHO to train a neural network. Instead of using standard backpropagation (which often gets stuck in "local minima" or mediocre solutions), they use the "hawks" to find the best weights and biases for the network. Once trained, this FNN can take user locations as input and instantly output the optimal drone position and power settings.
Datasets
- Synthetic VLC Data: The primary evaluation uses simulated environments where 10 to 20 ground users are randomly distributed in a 10m x 10m area.
- Benchmark Datasets: To prove that their HHO-based neural network trainer is superior to other methods, they tested it against the well-known Iris and Cancer datasets from the UCI Machine Learning Repository.
Key Insights and Ideas
- The "Local Minima" Trap: Traditional training algorithms for neural networks are like a hiker trying to find the lowest point in a valley by only looking at their feet; they often get stuck in a small dip halfway up the mountain. HHO acts more like a flock of birds surveying the whole range, making it much more likely to find the true "global" optimum.
- VLC and NOMA are a Perfect Match: NOMA performs best when the Signal-to-Noise Ratio (SNR) is high. Because VLC operates over short distances with clear lines of sight, it naturally provides the high SNR that NOMA needs to thrive.
- The Altitude Paradox: You might think lower is always better for light, but the paper shows a "sweet spot" for UAV altitude. If the drone is too low, the angles of light incidence become too sharp for the ground sensors; if it's too high, the signal weakens due to distance.
Key Results
- Superior Performance: The proposed HHOPAP algorithm consistently outperformed standard OFDMA (the tech used in 4G/5G) and other NOMA variants (like GRPA) in terms of total network speed (sum-rate).
- Efficiency: The HHO-trained neural network (HHOFNN) achieved results within 10% of the computationally heavy direct optimization, but it can provide these answers in real-time.
- Optimization Accuracy: When compared to other popular swarm algorithms like Particle Swarm Optimization (PSO) and Genetic Algorithms (GA), the HHO trainer achieved a significantly lower Mean Squared Error (MSE) on benchmark datasets.
- Convergence: The HHO algorithm is shown to be stable, typically reaching a final solution within about 250 iterations.
Limitations and Future Work
- Fixed-Wing UAVs: This study focused on "rotary-wing" drones that can hover. Future work needs to address "fixed-wing" drones that must stay in constant motion, requiring complex trajectory optimization rather than just static placement.
- Energy Constraints: UAVs have limited battery life. The authors note that future models should include "energy-efficient" designs that balance data rates with the power required to keep the drone airborne.
- Security: Because light can be intercepted by anyone in the "footprint," the authors suggest investigating physical layer security for these UAV-VLC systems.
Summary updated