Biotelemetry Signal Denoising Using Generative AI

In this paper, we introduce the application of diffusion models, a new type of generative deep learning algorithm, as an innovative solution for biotelemetry signal denoising. Traditionally, techniques like Least-Mean-Square (LMS) adaptive filtering and Wavelet Transform (WT) have been used for denoising, but their effectivness are often limited in Low Signal-to-Noise Ratio (SNR) regimes. The diffusion model is capable of denoising signals, and has the advantages of simple training process. Through simulations and ex vivo experiments with porcine tissue model in backscatter communication-based biotelemetry systems, we evaluated the performance of dif-fusion models against aforementioned conventional denoising algorithms. The Bit-Error-Rate (BER) was used as the primary metric of evaluation. Furthermore, given the time-consuming nature of iterative sampling process of diffusion model, this work also investigated a non-iterative approach to diffusion model. The results presented in this paper indicate that both iterative and non-iterative diffusion models offer a more effective method for reducing noise in biotelemetry than conventional denoising algorithms. Furthermore, it was observed that the non-iterative method outperforms the iterative method, however at cost of a lengthy training time and susceptibility to delayed convergence if the training parameters are incorrectly selected. The results presented in this work will aid the design of biotelemetry system capable of functioning in low SNR regimes.