Flexible Parallelized Empirical Mode Decomposition in CUDA for Hilbert Huang Transform

Abstract

Hilbert-Huang Transform (HHT) is a process of adaptive analysis applicable to non-linear and non-stationary data such as voice and biomedical signals. Empirical Mode Decomposition (EMD) is a key in HHT and decomposes data into multiple Intrinsic Mode Functions (IMFs). Traditionally, EMD is computed on all data points in a serial manner, thus making its execution time grows at least linearly with the data size. In this work, a 3-stage parallelized EMD algorithm working on a CUDA architecture is proposed to improve performance over traditional EMD. Moreover, additional merging cubic spline interpolation (MCSI) and GPU acceleration techniques are also incorporated for achieving high parallelism and high accuracy. Experimental result shows that our parallelized EMD in CUDA achieves 37.9x and 33.7X speedups with 0.0051% and 0.002% errors on voice and EEG datasets of 1-million points, respectively.