Journal of Fuzzy Systems and Control.
Vol.
No 2, 2025 ISSN: 2986-6537.
DOI: 10.
59247/jfsc.
Optimizing Hybrid LiFi Communication Systems Using Fuzzy Reinforcement Learning for Enhanced Network Performance Fatimah Abdulameer Azeez 1,*.
Bashar Jabar Hamza 2 Department of Technical Communication Engineering.
Al-Furat Al-Awsat Technical University.
Al Najaf.
Iraq Email: 1 fatimah.
ms6@student.
iq, 2 coj.
bash@atu.
*Corresponding Author AbstractAiLight Fidelity (LiF.
technology has emerged as a pivotal solution for high-speed data transmission in modern communication networks.
However, its limitations, such as signal obstruction and coverage gaps, necessitate integration with hybrid systems to ensure seamless connectivity.
This study introduces a novel Fuzzy Reinforcement Learning (FRL) algorithm to optimize hybrid LiFi communication systems, addressing critical challenges like handover inefficiency, load imbalance, and dynamic environment adaptation.
The proposed FRL framework combines fuzzy logic to manage uncertainties in user mobility and channel conditions with reinforcement learning to dynamically adapt network parameters, ensuring optimal performance.
Through comprehensive simulations and real-world validations, the hybrid system demonstrates significant improvements in throughput .
8 Gbp.
, handover latency .
, and coverage .
% user connectivit.
compared to standalone LiFi and traditional RF-based Key contributions include non-linear decisionmaking, long-term performance optimization, and scalable deployment strategies for next-generation wireless systems.
The results highlight the potential of FRL-optimized hybrid LiFi networks to overcome current bandwidth constraints, offering a robust solution for 6G and IoT applications.
This work bridges the gap between theoretical advancements and practical implementation, paving the way for energy-efficient, highperformance communication systems.
KeywordsAiLiFi.
Hybrid Communication.
Fuzzy Logic.
Reinforcement Learning.
Handover Optimization.
Network Performance.
Component INTRODUCTION The exponential growth in global internet traffic, accelerated by remote work, telemedicine, and digital entertainment, has strained traditional radio frequency (RF)based networks, necessitating alternative high-speed communication technologies .
Light Fidelity (LiF.
, a networked extension of Visible Light Communication (VLC) shown in Fig.
1 has emerged as a promising solution, leveraging light-emitting diodes (LED.
for ultra-fast data transmission while offering advantages such as enhanced security, immunity to electromagnetic interference, and energy efficiency .
Unlike RF signals.
LiFi operates in the unlicensed visible light spectrum (OO300 TH.
, mitigating spectrum congestion and enabling ultra-high data rates exceeding 10 Gbps in experimental setups .
However.
LiFiAos reliance on online-of-sight (LoS) transmission introduces critical limitations, including signal blockage by opaque objects and limited coverage range .
Fig.
Schematic diagram for the LI_FI system To overcome these challenges, hybrid LiFi/WiFi networks have been proposed, combining LiFiAos high bandwidth with WiFiAos broader coverage .
Despite their potential, such hybrid systems face several unresolved issues:
Handover inefficiencyAeFrequent and unnecessary handovers degrade performance, particularly in mobile scenarios .
Load imbalance Ae Uneven user distribution across access points (AP.
leads to congestion and reduced network efficiency .
Dynamic environment adaptation Ae Traditional rulebased methods struggle with real-time variations in user mobility and channel conditions .
Recent studies have explored machine learning (ML)based solutions, including reinforcement learning (RL) for dynamic resource allocation and fuzzy logic for handling uncertainty .
, .
However, existing approaches often fail to balance short-term adaptability with long-term optimization, highlighting the need for an integrated solution .
This paper presents the first integration of Fuzzy Reinforcement Learning (FRL) for hybrid LiFi network optimization, addressing the limitations of standalone ML Our key contributions include:
A hybrid FRL framework combining fuzzy logicAos uncertainty handling with RLAos adaptive learning for realtime decision-making.
Dynamic load balancing and handover optimization, reducing unnecessary handovers by 40% compared to RSSI-based methods.
Comprehensive validation via MATLAB/NS-3 simulations, demonstrating superior throughput .
8 Gbp.
, coverage .
%), and fairness (JainAos index = 0.
The proposed system is scalable for 6G and IoT deployments, offering a robust solution for next-generation wireless networks.
This work is licensed under a Creative Commons Attribution 4.
0 License.
For more information, see https://creativecommons.
org/licenses/by/4.
Journal of Fuzzy Systems and Control.
Vol.
No 2, 2025 II.
SYSTEM MODEL
Network Topology Let A=.
1,A,aN}} be the set of NN LiFi APs and aN 1 denote the WiFi AP.
Users U=.
1,A,uM} are randomly distributed in a room of size LyWyH.
Hij = .
Acos m .
j ) 2Ad2ij ILos A M : Lambertian order.
A A
: Photodetector area.
A ij : Incidence angle.
A dij : Distance between uiui and ajaj.
A I LoS : Indicator function .
if LoS exists, else .
ycayc OO .
A , ycA .
where at representing AP selection.
The Reward Function.
ycyc = yc1 .
ycIycAycIycuyceyc Oe yc2 .
yayaycC Oe yc3 .
where w1, w2, w3, w1, w2, w3 are weights.
The Fuzzy Rules can be selected as follows:
A Inputs: iSNR (Positive/Zero/Negativ.
, vv (Slow/Mediu m/Fas.
Load (Low/Medium/Hig.
A Output: Handover probability p OO .
Constraints Coverage Constraint:
yc = .
,2.
A , ycA .
ycIycAycIyc > yuycoycnycu where min is the minimum SNR threshold .
, 15 dB for LiF.
Power Constraint:
Optimization Objectives The Objective 1: Maximize ThroughputAy .
ycEyc < ycEycoycaycu OAyc.
ycEyc yaycnyc ycAycaycu Oc ycoycuyci2 .
OcycoOyc ycEyco yaycnyco yua 2 ycn=1 User-AP Association:
ycA 1 where: Pj: Transmit power of AP aj, and the E2 is the Noise The Objective function with the two conditions is to minimize Handovers is defined as:
ycN Oc ycuycnyc = 1 .
OAycn.
yc=1 where xi OO.
indicates connection.
ycoycnycu Oc yayaycC .
cycn , y.
yc=1 where I HO=1 if a handover occurs at time t, else 0.
Last objective function for load blanking as shown in Fig.
2 is defined as:
SIMULATION SETUP AND ALGORITHM STEPS FOR
FRL-OPTIMIZED LIFI SYSTEM
where uj is the set of users connected to aj.
Simulation Setup First, let us define the environment configuration where the room dimensions are 10y10y3 m (L y W y H).
The LiFi APs: 4 APs mounted on the ceiling at positions .
5,2.
5,2.
, .
5,7.
Users: 8 mobile users with random initial positions and velocities vOO.
m/sec.
Fuzzy Reinforcement Learning (FRL) Formulation The State Space vector for the problem as follows:
Key Parameters The simulation setup parameters are listed in Table 1.
ycA 1 ycoycnycu Oc | yc=1 .
cyc | Oe ycA ycA ! ycyc = .
cIycAycI1 .
A , ycIycAycIycA 1 , ycO, yaycuycaycc1 .
A , yaycuycayccycA 1 } .
ycEyc yaycnyc a ycIycAycI = yua2 .
a v: User velocity, a Loadj: Number of users connected to aj.
Fig.
Simulation setup showing LiFi AP placement .
range squar.
, user distribution .
lack dot.
, and coverage zones .
haded region.
, .
The main action space is defined as:
Table 1.
Parameter LiFi transmit power LiFi bandwidth Noise floor (LiF.
Learning rate () Discount factor () Exploration rate (A) Simulation setup parameters Value 10 dBm 100 MHz -90 dBm Description LED power per AP OFDM-based modulation Photodetector noise Q-learning update step size Future reward importance A-greedy policy Algorithm Steps The main algorithm steps are as follows:
Input: Real-time network use eq.
iSNR=SNRtargetOeSNRcurrent Ie {Negative.
Zero.
Positiv.
Velocity v Ie {Slow.
Medium.
Fas.
AP Load Ie {Low.
Medium.
Hig.
Rule Evaluation: If iSNR is Positive AND v is Slow AND Load is Low.
THEN initiate handover .
= 0.
Defuzzify Output: Crisp handover probability pp using the centroid method.
Output: Optimal AP selection at OO.
,2,3,.
Reinforcement Learning Agent: Initialize Qtable: Q.
=0 for all state-action pairs.
Reward Calculation: using Eq.
Huynh Van Khuong.
Nonlinear Control Law Design for Inverted Pendulum Systems via RBF Neural Networks Journal of Fuzzy Systems and Control.
Vol.
No 2, 2025 Q-Table Update:
cyc , ycayc ) Ia ycE .
cyc , ycayc ) yu .
cyc yu max ycE ycyc 1 , yca Oe ycE.
cyc , ycayc )].
Monitor Performance: Log throughput, handover rate, and load distribution.
Where the performing metric is listed in Table 2.
IV.
Table 4.
Convergence rate Iteration 1,000 5,000 10,000 Avg.
Reward
SIMULATION RESULTS
Based on the simulation setup section.
Fig.
3 shows the performance of the proposed system.
The proposed FRL algorithm outperforms conventional RSSI-based and standalone LiFi/WiFi systems in all key metrics: Throughput:
Achieves 4.
8 Gbps .
2 Gbps for LiFi-only and 1.
1 Gbps for WiF.
Handover Latency: Reduced to 20 ms .
50 ms in RSSI-based method.
Coverage Gap Resolution: Connects 100% of users .
75% in baseline LiF.
The Load Balancing (Table 3 The FRL algorithm distributes users evenly across APs under high traffic .
Table 2.
Metric Performance metric Formula Target Oc ycoycuyci2 .
ycIycAycIycn ) Maximize (> 4 Gbp.
ycA Throughput ycn=1 Handover Rate Fairness Index (JainAo.
ycAycu.
ycuyce Eaycaycuyccycuycyceyc ycNycnycoyce (Ocyc yaycuycaycc ) ya= ycA Ocyc .
aycuycayccyc )2 Table 3.
AP1
AP2
AP3
AP4
Target: 0.
Load balancing Users (RSSI) Fig.
Handover stability Minimize (< 0.
Users (FRL) Table 5.
Fairness index for different methods Method RSSI-based FRL (Propose.
Fairness Index (F) CONCLUSION AND FUTURE WORKS The proposed Fuzzy Reinforcement Learning (FRL)optimized LiFi system addresses critical challenges in modern wireless communication, delivering measurable improvements over standalone LiFi networks.
Key Achievements Throughput:
Achieves 4.
Gbps, standalone LiFi .
2 Gbp.
by 50% and WiFi .
1 Gbp.
3y, enabling ultra-high-speed applications.
Mobility Support: Reduces handover latency to 20 ms, a 60% improvement over traditional RSSI-based methods .
, ensuring seamless connectivity.
Coverage: Eliminates dead zones, providing 100% user coverage even in non-line-ofsight (NLoS) scenarios, compared to 75% with baseline LiFi.
Load Balancing: Distributes traffic efficiently (JainAos fairness index = 0.
93 vs.
72 for RSSI), minimizing congestion and maximizing AP utilization.
Fig.
FRL algorithm performance of conventional RSSI-based and standalone LiFi/WiFi systems Key Observation: FRL reduces AP1/AP2 congestion by 33% while utilizing underloaded APs .
AP.
Handover Efficiency (Fig.
FRL maintains stable connections during mobility .
, triggering 40% fewer unnecessary handovers than RSSI-based methods.
Packet Loss: < 0.
1% with FRL .
2% in baseline.
Now consider the convergence analysis (Table .
The RL agent converges to the optimal policy within 10,000 iterations.
Implication:
Fast convergence suitable for real-time deployment.
Indeed, the fairness Index (JainAo.
Comparison .
isted in Table .
Interpretation: FRL improves load-balancing fairness by 29%.
Broader Implications Scalability: The FRL frameworkAos real-time convergence .
ithin 10,000 iteration.
and adaptability make it suitable for dense 6G and IoT deployments.
Energy Efficiency:
Leverages LiFiAos dual-use capability .
llumination communicatio.
, reducing reliance on power-intensive RF Practical Viability:
Validated MATLAB/NS-3 simulations and testbed experiments under dynamic mobility .
p to 3 m/.
Future Directions These results suggest that FRL-optimized hybrid LiFi systems offer a promising pathway to address bandwidth constraints in next-generation wireless networks, such as:
Huynh Van Khuong.
Nonlinear Control Law Design for Inverted Pendulum Systems via RBF Neural Networks Journal of Fuzzy Systems and Control.
Vol.
No 2, 2025 A Integration with 5G/6G heterogeneous networks for hybrid RF-VLC systems.
A Testing in real-world environments .
, multi-floor buildings, 3D mobility scenario.
A Extension to deep reinforcement learning (DRL) for large-scale network optimization.
REFERENCES