CUDA DSP Filter for ECG Signals

Real time processing is very important and critical for analysis of ECG signals. Prior to each processing, the signal needs to be filtered to enable feature extraction and further analysis. In case of building a data processing center that analyzes thousands of connected ECG sensors, one expects that the signal processing needs to be done very fast. In this paper, we focus on parallelizing the sequential DSP filter for processing of heart signals on GPU cores. We set a hypothesis that a GPU version is much faster than the CPU version. In this paper we have provided several experiments to test the validity of this hypothesis and to compare the performance of the parallelized GPU code with the sequential code. Assuming that the hypothesis is valid, we would also like to find what is the optimal size of the threads per block to obtain the maximum speedup. Our analysis shows that parallelized GPU code achieves linear speedups and is much more efficient than the classical single processor sequential processing.