Low-power perceptron model based ECG processor for premature ventricular contraction detection

This paper proposes an electrocardiogram (ECG) processor for premature ventricular contraction (PVC) detection in wearable monitoring. A novel feature called QRS areas ratio (QAR) is extracted from wavelet transform coefficients which can efficiently enhance accuracy of PVC beat classification. This feature along with other two beat interval features is introduced to a single neuron perceptron model based classifier. And the classifier achieves ultra-low complexity by linear processing and reduces power consumption by eliminating memory overhead. Finally, the proposed processor exhibits an average sensitivity of 98.7% and specificity of 98.9% in testing of MIT-BIH arrhythmia database. Implemented in 40 nm CMOS technology, it achieves 127 nW of power consumption at 0.5 V supply voltage.