Energy Efficient Models for Future Communication Networks

Spiking Neural Networks (SNNs), inspired by the sparse and event-driven nature of biological neural systems, offer a highly energy-efficient alternative to Deep Learning (DL) models. DL is widely used in telecommunications for applications such as beamforming, network slicing, anomaly detection, channel prediction, and antenna tilt optimization. However, DL's high energy requirements restrict sustainability and scalability. In this paper, the potential of SNNs in telecommunication applications is examined, which shows notable improvements in energy efficiency. SNNs were able to accurately rebuild channel information with fewer antennas in channel prediction, achieving up to $41 \times$ energy efficiency. SNNs were $10 \times$ more efficient than artificial neural networks (ANNs) for network slicing and $16 \times$ more efficient for anomaly identification. The efficiency of SNNs in reducing computational costs is further shown by experiments on beamforming and antenna tilt datasets. Our results demonstrate that SNNs may provide comparable accuracy with significantly less energy consumption than DL in important telecommunication use cases. The potential of SNNs to drive sustainable energy-efficient AI solutions in next generation telecommunication networks (advanced 5G, Future 6G) is highlighted by this work.