Neural Recording Analog Front-End Noise Reduction with Digital Correlated Double Sampling

Amplifiers in biomedical sensing systems play a crucial role in elucidating often weak biosignals such as those originating from the brain in electroencephalograms (EEG) but can suffer from flicker (1/f) and thermal noise. Correlated double-sampling (CDS) is a method that can be used to reduce low frequency noise. Conventionally implemented as an analog reset of the amplifier between samples of interest, the CDS operation adds kT/C sampling noise which may exceed the low-frequency noise being removed, especially if the reset is not allowed adequate time to settle. Analog CDS therefore, puts a limitation on the sampling frequency of such sensing systems, making them less ideal for the fast, high-bandwidth acquisition required for applications such as neural interfaces. Digital CDS is presented here as a technique which relies on intermittent sampling of an internal reference voltage taken during brief disconnections from the sensor between real samples for a noise correction performed digitally post-acquisition. This work demonstrates the implementation of digital CDS on a neural interface system-on-chip (NISoC) and details the optimization of a windowed weighting correction algorithm to achieve a 71% improvement in noise performance. In-Ear EEG and EOG recordings were performed with and without CDS for comparison.