A Sub-<inline-formula> <tex-math notation="LaTeX">$\mu\text{W}$ </tex-math> </inline-formula>/Channel, 16-Channel Seizure Detection and Signal Acquisition SoC Based on Multichannel Compressive Sensing

Abstract

An accurate 16-channel seizure detection system-on-chip is presented, which is based on extracting features from the compressed recorded data using multichannel compressive sensing (MCS). MCS is used to reduce the transmission data rate of sparse biological signals and to lower the power consumption of the resource-constrained recording and detection systems. Conventional MCS architecture introduces several issues such as requiring high-resolution analog-to-digital-converter (ADC) and offset-proportional random signal generation at the output of MCS block. We introduce a new multi-input single-output compressive sensing (MISOCS) block that uses a straightforward technique to embed the data of all channels in each sample of the compressed signal. Hence, the proposed MISOCS block provide more information of recorded data than the conventional MISOCS block, mathematically. This technique is shown to require less ADC resolution in comparison to the conventional MCS technique. Furthermore, the problem of offset-proportional random signal generation at the output of MCS block is solved in the proposed architecture. The system is implemented in a UMC 0.18- $ \mu \text{m}$ CMOS technology. The proposed seizure detection system is tested over 420 h of clinical iEEG data including 23 seizures and reaches a perfect sensitivity of 100% and an average false alarm rate of 0.09 $\text{h}^{-1}$ for artifact-free channels.