SIGNETS: Neural Network Architectures for m-QAM Soft Demodulation

This paper presents a novel approach to Quadrature Amplitude Modulation (QAM) demodulation using neural networks, addressing the limitations of traditional demodulation techniques in complex channel conditions. Through systematic Neural Architecture Search and Hyper-parameter optimization, we develo...

Full description

Saved in:
Bibliographic Details
Main Authors: Aravind R. Voggu, Kanish R, Nishith Akula, Lohitaksh Maruvada, Takanori Shimizu, Madhav Rao
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11071691/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849223110388088832
author Aravind R. Voggu
Kanish R
Nishith Akula
Lohitaksh Maruvada
Takanori Shimizu
Madhav Rao
author_facet Aravind R. Voggu
Kanish R
Nishith Akula
Lohitaksh Maruvada
Takanori Shimizu
Madhav Rao
author_sort Aravind R. Voggu
collection DOAJ
description This paper presents a novel approach to Quadrature Amplitude Modulation (QAM) demodulation using neural networks, addressing the limitations of traditional demodulation techniques in complex channel conditions. Through systematic Neural Architecture Search and Hyper-parameter optimization, we develop a family of Convolutional Neural Network architectures that demonstrate robust performance across challenging channel conditions, including multi-path fading, inter-symbol interference, and non-linear distortions, without requiring explicit channel estimation. We comparatively evaluate the members of the family of networks to find a Pareto-optimal Neural Network Demodulator with a balance of demodulation accuracy and computational cost, achieving an average accuracy of 99.658% across 4 dB to 24 dB SNR while requiring less than 16,000 Floating-Point Operations (FLOPs) for every demodulated QAM-16 symbol. A practical Field Programmable Gate Array (FPGA) implementation that achieves 2.52 Million bits per second throughput while maintaining 99.55% demodulation accuracy through structured pruning and quantization-aware training is presented. Experimental validation over acoustic channels demonstrates superior performance compared to traditional techniques, with performance further enhanced through fine-tuning.
format Article
id doaj-art-fe47a0bea4524bd0b15a3f8a61bbf63c
institution Kabale University
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-fe47a0bea4524bd0b15a3f8a61bbf63c2025-08-25T23:18:39ZengIEEEIEEE Access2169-35362025-01-011311707811708910.1109/ACCESS.2025.358604311071691SIGNETS: Neural Network Architectures for m-QAM Soft DemodulationAravind R. Voggu0https://orcid.org/0000-0002-6242-1164Kanish R1https://orcid.org/0009-0000-1640-5852Nishith Akula2https://orcid.org/0009-0005-6299-0825Lohitaksh Maruvada3https://orcid.org/0009-0004-8244-6846Takanori Shimizu4https://orcid.org/0009-0000-3920-2652Madhav Rao5https://orcid.org/0000-0003-2278-9148International Institute of Information Technology Bangalore, Bengaluru, Karnataka, IndiaInternational Institute of Information Technology Bangalore, Bengaluru, Karnataka, IndiaInternational Institute of Information Technology Bangalore, Bengaluru, Karnataka, IndiaInternational Institute of Information Technology Bangalore, Bengaluru, Karnataka, IndiaSony India Software Centre, Bengaluru, Karnataka, IndiaInternational Institute of Information Technology Bangalore, Bengaluru, Karnataka, IndiaThis paper presents a novel approach to Quadrature Amplitude Modulation (QAM) demodulation using neural networks, addressing the limitations of traditional demodulation techniques in complex channel conditions. Through systematic Neural Architecture Search and Hyper-parameter optimization, we develop a family of Convolutional Neural Network architectures that demonstrate robust performance across challenging channel conditions, including multi-path fading, inter-symbol interference, and non-linear distortions, without requiring explicit channel estimation. We comparatively evaluate the members of the family of networks to find a Pareto-optimal Neural Network Demodulator with a balance of demodulation accuracy and computational cost, achieving an average accuracy of 99.658% across 4 dB to 24 dB SNR while requiring less than 16,000 Floating-Point Operations (FLOPs) for every demodulated QAM-16 symbol. A practical Field Programmable Gate Array (FPGA) implementation that achieves 2.52 Million bits per second throughput while maintaining 99.55% demodulation accuracy through structured pruning and quantization-aware training is presented. Experimental validation over acoustic channels demonstrates superior performance compared to traditional techniques, with performance further enhanced through fine-tuning.https://ieeexplore.ieee.org/document/11071691/CNNcommunicationFPGALLRmachine learningmultipath components
spellingShingle Aravind R. Voggu
Kanish R
Nishith Akula
Lohitaksh Maruvada
Takanori Shimizu
Madhav Rao
SIGNETS: Neural Network Architectures for m-QAM Soft Demodulation
IEEE Access
CNN
communication
FPGA
LLR
machine learning
multipath components
title SIGNETS: Neural Network Architectures for m-QAM Soft Demodulation
title_full SIGNETS: Neural Network Architectures for m-QAM Soft Demodulation
title_fullStr SIGNETS: Neural Network Architectures for m-QAM Soft Demodulation
title_full_unstemmed SIGNETS: Neural Network Architectures for m-QAM Soft Demodulation
title_short SIGNETS: Neural Network Architectures for m-QAM Soft Demodulation
title_sort signets neural network architectures for m qam soft demodulation
topic CNN
communication
FPGA
LLR
machine learning
multipath components
url https://ieeexplore.ieee.org/document/11071691/
work_keys_str_mv AT aravindrvoggu signetsneuralnetworkarchitecturesformqamsoftdemodulation
AT kanishr signetsneuralnetworkarchitecturesformqamsoftdemodulation
AT nishithakula signetsneuralnetworkarchitecturesformqamsoftdemodulation
AT lohitakshmaruvada signetsneuralnetworkarchitecturesformqamsoftdemodulation
AT takanorishimizu signetsneuralnetworkarchitecturesformqamsoftdemodulation
AT madhavrao signetsneuralnetworkarchitecturesformqamsoftdemodulation