Work

A Five-Tier Digital Twin Framework with Physics-Informed Neural Networks for Predictive Battery Management in 5G User Equipment: Joint RAN-Aware Degradation Modeling and Autonomous Power Saving

Year: 2026
Type: article
Abstract: The deployment of 5G New Radio (NR) introduces millimeter-wave communications, massive MIMO, and carrier aggregation, which significantly increase the power consumption of user equipment (UE) modems and accelerate lithium-ion battery degradation. Existing Battery Management Systems (BMS) optimize charge/discharge cycles without awareness of radio access network (RAN) traffic patterns, while the RAN configures Discontinuous Reception (DRX) and Bandwidth Part (BWP) parameters without knowledge of the battery’s State of Health (SoH). This paper proposes DT-5GBat, a five-tier Digital Twin framework that closes this gap by integrating a physics-informed neural network (PINN) with a 5G network simulator for predictive, cross-layer battery management. Tier 1 provides a reduced-order Single Particle Model (SPM) of the Li-ion cell. Tier 2 fuses real-time BMS sensor data with RAN metrics (RSRP, CQI, RRC state, DRX duty cycle). Tier 3 employs a PINN-LSTM architecture to predict SoH trajectory and Remaining Useful Life (RUL). Tier 4 solves a multi-objective optimization problem trading off energy consumption, latency, and battery aging. Tier 5 executes the optimized policy in a closed control loop via UE middleware or O-RAN RIC. It is validated by co-simulation between NS-3/5G-LENA and the Rakhmatov non-linear battery model using NASA Ames, Oxford, and CALCE datasets augmented with synthetic 5G load traces. DT-5GBat extends UE autonomy by 21.3% and reduces capacity fade by 19.8% over 500 equivalent cycles, with a RUL prediction MAPE of 3.42% and a computational overhead below 1.8% of the total UE energy budget.
Language: English
Open Access status: gold
Estimated APC paid: $1,930
OpenAlex ID: W7236245183
FWCI:
Cites: 33
Cited by:
Related to: 10