Skip to Main Content
Real-time recognition of multichannel, continuous-time physiological signals has been crucial for the development of implantable biomedical devices. This work investigates the feasibility of using the diffusion network, a stochastic recurrent neural network, to recognise continuous-time biomedical signals. In addition, a hardware-friendly approach for achieving real-time recognition is proposed and tested with both artificial and real biomedical data. Based on this approach, the diffusion network is demonstrated to exhibit great tolerance against noise and drifts in continuous-time signals being classified.