Integrated circuit design methods for biosignal measurements with high input impedance requirements
Analog front-end circuits are essential during the acquisition of sensed biosignals such as during electroencephalography (EEG), electrocardiography (ECG) and electroretinography (ERG). EEG is crucial for diagnosing neurological conditions such as epilepsy, sleep disorders, encephalopathies, and coma. For epilepsy, effective treatment requires precise tracking and profiling of seizures to ensure correct medication administration. Furthermore, accurate seizure characterization necessitates continuous EEG monitoring in a way that does not disrupt a patient's daily activities. Wearable devices are well-suited for this purpose, but their relatively high power consumption limits the duration of continuous monitoring. Hence, there is a need for a low-power analog EEG signal processing system that includes a feature extraction circuit capable of deriving spectral power directly within the analog domain.In EEG signal measurement systems, the instrumentation amplifier is typically the first block of the analog front-end that processes the signals prior to analog filtering and the conversion with an analog-to-digital converter for further digital processing. Hence, being the first block in the system, the reduction of noise levels is of foremost importance during the design of instrumentation amplifiers, especially while reducing their power consumption for portable applications. Another trend is that gel-free electrodes (i.e., dry electrodes) are becoming increasingly popular in long-term EEG monitoring applications due to their practical advantages over wet electrodes. However, the use of dry electrodes is associated with high contact impedance at the skin interface. Therefore, a high input impedance is an important requirement for instrumentation amplifiers in order to acquire EEG signals with high fidelity. This research focuses on the investigation of a new chopper instrumentation amplifier (IA) architecture with a symmetric negative capacitance generation feedback (NCGFB) loop, which is primarily designed to be part of a low-power analog front-end chip for EEG signal monitoring. The purpose of the negative capacitance generation technique is to cancel the parasitic capacitances of the cable and dry electrodes at the IA input in order to boost the input impedance. The negative capacitance is generated through feedback loops containing 9-bit digitally programmable capacitor banks. For prototyping with automated measurements, these capacitor banks are controlled using two machine learning (ML) optimization algorithms, the genetic algorithm (GA) and the particle swarm (PSO) algorithm. These algorithms offer a significant time reduction compared to a calibration with an exhaustive search, reducing calibration time by a factor of over 10^6 (with four 9-bit digital control words) while conserving computational resources. Notably, this dissertation introduces the first use of ML-based techniques for the digital-assisted calibration of an IA's input impedance, marking an advancement in the field of IA design. The GA algorithm achieved an input impedance of 1.7 GΩ after four generations (iterations), while the PSO algorithm achieved 1.3 GΩ with five iterations. A chopping technique was developed to enhance the noise performance of the IA with input impedance boosting. The proposed IA is based on a capacitively-coupled IA (CCIA) architecture with circuit-level innovations for improved noise performance and higher input impedance. It has fully symmetric negative capacitance generation feedback (NCGFB) loops that enable the use of identical NCGFB capacitor banks for common-mode rejection ratio (CMRR) enhancement. The IA was designed and fabricated in 65-nm CMOS technology, and it consumes 2.46 μW from a 1.2V supply. Chip measurements showed that the IA has 44 dB gain, 40 Hz bandwidth, total harmonic distortion of -44.3 dB with a 35 mVpp sinusoidal output at 10 Hz, CMRR > 90.9 dB, a 92.3 dB power supply rejection ratio (PSRR), 0.54-μV integrated input-referred noise over a bandwidth of 0.1-40 Hz with a noise efficiency factor of 4.75, and an input impedance of 1.9 GΩ at 10 Hz even with an extra external capacitance of 100 pF at its inputs.--Author's abstract
Read more