Analog-based Compressive Sensing of Multichannel Neural Signals: Systematic Design Approaches

Abstract

Compressive Sensing (CS) is an emerging data compression method in the neurorecording application to decrease the transferring data rate as well as the power consumption. It can be implemented in both analog and digital domain, but analog has the potential of reducing the power more due to decreasing the sampling frequency in frontend. In analog, CS is usually achieved by switched capacitor circuits. Non-ideal specifications of Operational Transconductance Amplifier (OTA) of CS integrator such as finite gain, bandwidth, slew rate and output swing induce error and reduce the total SNR. In this paper, we simulate these non-idealities in Matlab and Simulink with the assumption that all other elements in this system are ideal. The results demonstrate that the SNR of the whole system is very sensitive to the gain, bandwidth and output swing of OTA. In other words, the bottleneck to achieve a high SNR in the neurorecording system is the CS encoder. For neurorecording implant applications, CS is mainly achieved with reasonable reported SNR between 8 and 24 dB. However, we can obtain an improved CS recording performance using main blocks, i.e. LNA and ADC with much relaxed specifications in order to reduce the power consumption as well as the silicon area.