Home // UBICOMM 2025, The Nineteenth International Conference on Mobile Ubiquitous Computing, Systems, Services and Technologies // View article
Authors:
Simon Boka
Oladayo Bello
Innocent Davidson
Keywords: Automatic modulation classification; Signal to noise ratio; RISC-V; interference.
Abstract:
Deep learning has redefined automatic modulation classification by replacing traditional hand-engineered features with end-to-end neural networks that process raw signal data, thus demonstrating high accuracy at moderate-to-high signal-to-noise ratios. While contemporary convolutional and hybrid recurrent network architectures achieve excellent performance, they often incur significant computational costs that hinder deployment on resource-constrained Internet of Things edge devices. To address this challenge, this work proposes and presents a low-cost, open-source radio platform that performs signal acquisition and utilizes vector extensions for accelerated inference. The platform integrates a commodity Realtek software-defined radio with reduced instruction set computer – five processors. The workflow methodology for the proposed approach is a reproducible, end-to-end pipeline for deploying signal classification models on resource-constrained devices in IoT networks. The pipeline's primary strength is its deterministic dataset assembly. The workflow establishes a coherent baseline for embedded classification under strict memory and processing power constraints typical in IoT devices.
Pages: 14 to 19
Copyright: Copyright (c) IARIA, 2025
Publication date: September 28, 2025
Published in: conference
ISSN: 2308-4278
ISBN: 978-1-68558-288-3
Location: Lisbon, Portugal
Dates: from September 28, 2025 to October 2, 2025