Enhanced demodulator for 5G NTN using spatio-temporal attention convolutional autoencoder and Akida Brainchip SNN

February 1, 2025·
Swetha Varadarajulu
,
Marcele O. K. Mendonça
,
Geoffrey Eappen
,
Jorge Querol
,
Symeon Chatzinotas
· 1 min read
Abstract
This paper presents a novel approach to overcoming challenges in 5G and 6G mobile satellite systems (MSS) in Low Earth Orbit (LEO), focusing on Non-Line-of-Sight (NLOS) issues in 5G Non-Terrestrial Networks (NTN) that connect directly with handheld devices. We utilize a Convolutional Neural Network (CNN) with a Spatio-Temporal Attention Network (STAN) autoencoder, which is then converted into a Spiking Neural Network (SNN) using Brainchip Akida’s Meta TF Software Framework. This integration of neuromorphic processing aims to enhance energy efficiency, reduce computational demands, and increase data transmission rates, optimizing Channel State Information (CSI) in compliance with 3GPP standards. Our STAN-CNN-SNN architecture achieves a 85% reduction in computational requirements, leading to decrease in power consumption, and increase in data rates within the C-Band spectrum. Simulations with LEO satellite MSS parameters demonstrate significant advancements in communication systems. The numerical results demonstrates substantial computational reductions with minimal capacity trade-offs.
Type
Publication
IET conference proceedings.
publication

Abstract & Overview

This paper presents a novel approach to overcoming challenges in 5G and 6G mobile satellite systems (MSS) in Low Earth Orbit (LEO), focusing on Non-Line-of-Sight (NLOS) issues in 5G Non-Terrestrial Networks (NTN) that connect directly with handheld devices. We utilize a Convolutional Neural Network (CNN) with a Spatio-Temporal Attention Network (STAN) autoencoder, which is then converted into a Spiking Neural Network (SNN) using Brainchip Akida’s Meta TF Software Framework. This integration of neuromorphic processing aims to enhance energy efficiency, reduce computational demands, and increase data transmission rates, optimizing Channel State Information (CSI) in compliance with 3GPP standards. Our STAN-CNN-SNN architecture achieves a 85% reduction in computational requirements, leading to decrease in power consumption, and increase in data rates within the C-Band spectrum. Simulations with LEO satellite MSS parameters demonstrate significant advancements in communication systems. The numerical results demonstrates substantial computational reductions with minimal capacity trade-offs.

Publication Metadata

6G-NTN AI & Neuromorphic SatCom