Real-Time Noise Cancellation With Deep Learning

Measurements are often contaminated with large amounts of non-stationary noise which require effective reduction techniques. We present a new deep learning algorithm learns continuously in real-time while filtering the data. The requires reference signal alongside noisy itself. As poof concept we demonstrate its performance by reducing electroencephalogram (EEG) help custom 3D printed compound electrode using flexible materials. With this setup achieved an average 20dB improvement to ratio EEG for both removing eye-blink and muscle-artefacts. This has potential not only improve but can be applied wide range industrial consumer applications such as sensing or cancelling headphones.