Towards Green AI-Native Networks: Evaluation of Neural Circuit Policy for Estimating Energy Consumption of Base Stations

TL;DR

The paper tackles estimating base-station energy consumption from performance counters in radio access networks and compares sparse Neural Circuit Policies (NCPs) against LSTMs. By leveraging continuous-time LTC-based NCPs with 90% sparsity, the authors demonstrate lower training energy and smaller parameter footprints while maintaining competitive accuracy (R^2 around $0.736$ for NCPs vs $0.748$ for LSTMs). The results indicate substantial reductions in energy consumption and model size for NCPs, with robustness to hyper-parameter variations, suggesting easier MLOps for telecom deployments. The findings support the broader goal of sustainable, AI-native network design by promoting compact, energy-efficient architectures that still deliver reliable energy estimations under data drift and noise.

Abstract

Optimization of radio hardware and AI-based network management software yield significant energy savings in radio access networks. The execution of underlying Machine Learning (ML) models, which enable energy savings through recommended actions, may require additional compute and energy, highlighting the opportunity to explore and adopt accurate and energy-efficient ML technologies. This work evaluates the novel use of sparsely structured Neural Circuit Policies (NCPs) in a use case to estimate the energy consumption of base stations. Sparsity in ML models yields reduced memory, computation and energy demand, hence facilitating a low-cost and scalable solution. We also evaluate the generalization capability of NCPs in comparison to traditional and widely used ML models such as Long Short Term Memory (LSTM), via quantifying their sensitivity to varying model hyper-parameters (HPs). NCPs demonstrated a clear reduction in computational overhead and energy consumption. Moreover, results indicated that the NCPs are robust to varying HPs such as number of epochs and neurons in each layer, making them a suitable option to ease model management and to reduce energy consumption in Machine Learning Operations (MLOps) in telecommunications.

Figures (6)

• Figure 1: NCP architecture and functionality. b2
• Figure 2: Flow chart illustrating the e2e pipeline of the ML models in training and inference phases. It also illustrates each process is potentially located in the mobile network.
• Figure 3: Snapshot of normalized energy consumption of a base station is given in the figure (top: trainingset, bottom: testset), where there exists data drift between the training and testset.
• Figure 4: Comparison of model performances using R2-score and MSE in different experiment settings.
• Figure 5: Energy consumption and model parameter size of LSTM and NCP models in different scenarios.