A Neural-ADC-Compatible Fully-Dynamic Lossless Adaptive Resolution Compression Technique for Energy-Constrained Bio-Signal Recording

This paper presents a lossless adaptive-resolution compression technique intended for energy-constrained implantable neural interface microsystems. The proposed method seamlessly integrates with the widely-adopted ADC-direct front-end architectures with minimal overhead and no need for architectural modification, making it universally-compatible. It maintains an always-Nyquist sampling rate and employs a single-bit quantizer to overcome the limitations of previously-reported adaptive sampling and adaptive resolution methods (e.g., missing short-time events, nonlinearity, power/area overdesign). It also features continuous DC variation calculation and adjustable hysteresis bands. The method is optimized and validated through MATLAB simulations, followed by synthesis and implementation in a standard 130nm CMOS process, resulting in an active area of 0.0529 mm² and total power consumption of 1.9µW @ 128kHz. Post-layout verification was conducted using prerecorded EEG and ECG datasets, demonstrating a minimum compression ratio of 83.75% of the theoretical maximum (i.e., $\frac{{{N_H}}}{{{N_L}}}$), highlighting its effectiveness for implantable neural recording devices.

I. Introduction

Neurological disorders presently impact over one billion individuals globally, and this figure is projected to rise due to the aging population [1]. Continuous recording of neuro-electrical activity has been proven highly valuable for several disorders for diagnostic and treatment purposes [2]- [8], [11]. The pursuit of more precise diagnosis and management of these disorders has driven the demand for neural recordings with higher spatio-temporal resolutions, which poses substantial challenges in terms of energy efficiency as discussed below.