PROCESS AWARE ANALOG-CENTRICSINGLE LEAD ECG ACQUISITION ANDCLASSIFICATION CMOS FRONTEND

The primary objective of this research work is the development of a low power single-lead ECG analog front-end (AFE) architecture which includes acquisition, digitization, process aware efficient gain and frequency control mechanism and a low complexity classifier for the detecting asystole, extreme bardycardia and tachycardia. Recent research on ECG recording systems focuses on the design of a compact single-lead wearable/portable devices with ultra-low-power consumption and in-built hardware for diagnosis and prognosis. Since, the amplitude of the ECG signal varies from hundreds of µV to a few mV, and has a bandwidth of DC to 250 Hz, conventional front-ends use an instrument amplifier followed by a programmable gain amplifier (PGA) to amplify the input ECG signal appropriately. This work presents an mixed signal ECG fronted with an ultra-low power two-stage capacitive-coupled signal conditioning circuit (or an AFE), providing programmable amplification along with tunable 2nd order high pass and lowpass filter characteristics. In the contemporary state-of-the-art ECG recording systems, the gain of the amplifier is controlled by external digital control pins which are in turn dynamically controlled through a DSP. Therefore, an efficient automatic gain control mechanism with minimal area overhead and consuming power in the order of nano watts only. The AGC turns the subsequent ADC on only after output of the PGA (or input of the ADC) reaches a level for which the ADC achieves maximum signal-to-noise-ratio (SNR), hence saving considerable startup power and avoiding the use of DSP. Further, in any practical filter design, the low pass cut-off frequency is prone to deviate from its nominal value across process and temperature variations. Therefore, post-fabrication calibration is essential, before the signal is fed to an ADC, to minimize this deviation, prevent signal degradation due to aliasing of higher frequencies into the bandwidth for classification of ECG signals, to switch to low resolution processing, hence saving power and enhances battery lifetime. Another short-coming noticed in the literature published so far is that the classification algorithm is implemented in digital domain, which turns out to be a power hungry approach. Moreover, Although analog domain implementations of QRS complexes detection schemes have been reported, they employ an external micro-controller to determine the threshold voltage. In this regard, finally a power-efficient low complexity CMOS fully analog classifier architecture and a heart rate estimator is added to the above scheme. It reduces the overall system power consumption by reducing the computational burden on the DSP. The complete proposed scheme consists of (i) an ultra-low power QRS complex detection circuit using an autonomous dynamic threshold voltage, hence discarding the need of any external microcontroller/DSP and calibration (ii) a power efficient analog classifier for the detection of three critical alarm types viz. asystole, extreme bradycardia and tachycardia. Additionally, a heart rate estimator that provides the number of QRS complexes within a period of one minute for cardiac rhythm (CR) and heart rate variability (HRV) analysis. The complete proposed architecture is implemented in UMC 0.18 µm CMOS technology with 1.8 V supply. The functionality of each of the individual blocks are successfully validated using postextraction process corner simulations and through real ECG test signals taken from the PhysioNet database. The capacitive feedback amplifier, Σ∆ ADC, AGC and the AFT are fabricated, and the measurement results are discussed here. The analog classification scheme is successfully validated using embed NXP LPC1768 board, discrete peak detector prototype and FPGA software interface