A Successive Approximation Algorithm with Machine Learning for ECG Signals

This paper proposes a new approximation algorithm that digitizes the estimation error of second-order difference of signal samples rather than digitizing the individual samples or their first and second order differences. This new method allows the number of comparisons needed to convert a signal sample into digital numbers, which is usually fixed at N in a conventional successive-approximation-register (SAR) analog-to-digital converter (ADC), to fall between 2 and N for almost all the signal samples in an N-bit ADC. With the implementation of machine learning to estimate the second-order difference of samples, this method is tested on electrocardiogram (ECG) signals in MATLAB. The results indicate a reduction of up to 60.41% in comparison to the regular SAR ADC, and an 8.8% or more decrease in comparison to other state-of-the-art techniques. On top of that, it also reduces the digital to analog converter (DAC) switching energy as well as the energy consumption of the SAR ADC digital portion.