Citation

Iftikhar, Ubaid and Tahir, Hassam Ahmed and Tahir, Hamna and Mahmud, Azwan (2025) Wireless Signal Generation via Modulation-Aware GANs for Low-Data AI. In: 2025 Multimedia University Engineering Conference (MECON), 21-23 July 2025, Cyberjaya, Malaysia.

Text 490.pdf - Published Version Restricted to Repository staff only Download (2MB)

Abstract

AI performance in wireless systems is limited by data scarcity and dynamic channel conditions. This paper introduces Modulation-Aware GANs (MA-GANs), a framework that synthesizes wireless signals by embedding modulation-specific constraints and modeling stochastic channel effects like Rayleigh fading and phase noise. Our design integrates domain-informed layers such as differentiable pulse shaping and spectral normalization to ensure physical-layer realism. We propose a hybrid training strategy that combines synthetic pre-training with minimal real-data fine-tuning, evaluated using wireless-centric metrics: Modulation Fidelity Score (MFS) and Spectral Compliance Index (SCI). MA-GAN narrows the sim-to-real gap to 1.8%, reduces EVM by 4.1×, and maintains BER below 10−3 under dynamic conditions. With only 20% real data, it outperforms fully real-data-trained baselines by 1.6%, cutting data costs by 80%. This work enables reliable AI-driven wireless systems for 5G, 6G, and IoT applications.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Wireless communication , Training, Measurement, 6G mobile communication, Technological innovation, 5G mobile communication, Stochastic processes, Data models, Internet of Things, Artificial intelligence
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5101-6720 Telecommunication. Including telegraphy, telephone, radio, radar, television
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	17 Mar 2026 02:51
Last Modified:	19 Mar 2026 01:36
URII:	http://shdl.mmu.edu.my/id/eprint/15465

Downloads

Edit (login required)