New Ping-Pong Scheduling for Low-Latency EMD Engine Design in Hilbert–Huang Transform

Abstract

Hilbert-Huang transform (HHT) is a powerful tool to analyze biomedical signals such as electroencephalography and electrocardiography signals. Empirical mode decomposition (EMD) is the kernel operation in HHT. Because an EMD algorithm usually needs large memory with long latency in computation, EMD is generally implemented by offline computations. In this brief, we propose a low-cost EMD engine with novel ping-pong scheduling. The resulting chip implementation can reduce half latency of the conventional EMD. Compared with the data-reuse scheme, the latency is reduced to only 0.003%, and the energy consumption can be reduced to 1%. The proposed EMD engine design makes the sliding ensemble EMD feasible for real-time HHT applications.