Efficient Service Differentiation and Energy Management in Hybrid WiFi/LiFi Networks

TL;DR

The paper tackles energy-efficient service differentiation in hybrid WiFi/LiFi networks by integrating network slicing with a learning-based slice predictor. A two-stage approach first uses mATRIC to predict per-user slices from KPI data via a resilient backpropagation model trained offline, then optimizes transmit precoding to maximize end-to-end EE under latency and power constraints. It introduces sequential convex programming with inner approximations to convexify non-convex rate and power terms, deriving concave bounds for Shannon terms and convex bounds for channel dispersion, and solves the resulting problems via CVX to obtain suboptimal solutions. Simulations show notable EE improvements over traditional precoding schemes and a WiFi-only baseline, with fast convergence and clear benefits from adding LiFi LEDs, highlighting the method’s practical potential for multi-service deployment in real-world hybrid networks.

Abstract

In this paper, we propose an innovative predict-and-optimize algorithm designed for hybrid WiFi/LiFi networks, aiming to achieve service differentiation while maximizing energy efficiency (EE). The proposed framework utilizes multi-access technology real-time intelligent controller (mATRIC) to dynamically predict the appropriate network slice for each user based on historically monitored key performance indicators (KPIs). This prediction is facilitated by a deep learning model trained using the resilient backpropagation algorithm, with training conducted on KPIs data at the universal non-real time RAN intelligent controller (non-RT RIC). This trained model enables real-time slice selection by mATRIC. In the subsequent phase, the algorithm focuses on optimizing EE of hybrid network as a function of precoding vectors for the predicted slices by employing techniques from sequential convex approximation and the inner approximation method. We introduce novel approximations to convert non-convex objective functions and constraints into convex forms, and develop an iterative algorithm to achieve sub-optimal solutions. Additionally, the EE maximization problem, ensures alignment with end-to-end latency requirements. It also addresses the various constraints inherent to hybrid systems, such as input signal limitations for LiFi LEDs, data rate restrictions, and power budget considerations. Simulation results validate the effectiveness of the proposed algorithm, demonstrating significant improvements in EE while ensuring service differentiation within hybrid network environments.

Figures (9)

• Figure 1: Illustration of System Model
• Figure 2: Training Accuracy of the Proposed Model.
• Figure 3: Training Loss of the Proposed Model.
• Figure 4: Convergence of the proposed algorithm for maximizing EE as function of precoding vectors
• Figure 5: Transmit power minimization based on proposed algorithm.