Citation

Jamal, Arshad and Kanesaraj Ramasamy, R. and Abdullah, Junaidi and Thanjappan, Sivasutha and Hamzah, Faizal Amri and Jamal, Shamsuriani Bt Md (2025) Energy-Efficient Biomedical Signal Decoupling Via Optimized Generative AI Models. Energy-Efficient Algorithms and Systems in Computing, 629. pp. 57-73. ISSN 2198-4182

Full text not available from this repository.

Abstract

The non-invasive diagnosis of cardiopulmonary diseases is reliant upon a separation of respiratory and cardiac sounds. Nevertheless, these bioacoustics signal frequently coincide in both the temporal and frequency domains, making precise separations a technically demanding endeavour. Traditional signal processing techniques and deep learning models show promise, but they usually require many resources, which limits their application in low-energy environments like wearable health monitors. This paper proposes an energy-efficient method utilising variable autoencoders (VAEs), a generative deep learning model, for the blind source separation of cardiac and pulmonary sounds. The suggested VAE architecture is streamlined and optimised for low-power devices while maintaining separation precision. It acquires compact latent representations of mixed bioacoustics sounds and reconstructs individual components through a probabilistic decoder. We assess the model using publicly accessible datasets and compare it with conventional and deep learning methodologies. The results indicate that our VAE model attains enhanced signal separation performance, as measured by Signal-to-Distortion Ratio (SDR) and Signal-to-Interference Ratio (SIR), while utilising considerably reduced power and memory resources. The model attains inference latency under 60 ms and functions at less than 2.5 watts on edge devices like the Raspberry Pi 4 and NVIDIA Jetson Nano. These results illustrate the feasibility of generative AI models for real-time, energy-efficient biomedical signal processing. The proposed technology presents a viable approach to intelligent, portable, and scalable cardiopulmonary diagnostic systems applicable in both clinical and remote environments.

Item Type:	Article
Subjects:	H Social Sciences > HD Industries. Land use. Labor > HD28-70 Management. Industrial Management > HD30.2 Electronic data processing. Information technology. Including artificial intelligence and knowledge management
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	10 Feb 2026 03:07
Last Modified:	10 Feb 2026 03:07
URII:	http://shdl.mmu.edu.my/id/eprint/15279

Downloads

Edit (login required)