Embedded Real-Time Heart Variability Analysis.

Heart Variability Analysis (HRV) is not suitable for real-time processing on a resource-limited, single sensor network node, such as a Body Sensor Network (BSN) node, due to the high sampling rate (> 200H z) required to digitise ECG signals and the non-preemtable nature of operating systems such as tinyOS. Both reasons combined dictate that the processing of each sample needs to be completed withing the inter-sample period, typically 5 msec for ECG signals. This paper discusses a dual-layer real-time heart variability analysis algorithm. The top layer is invoked every time a sample arrives. This layer includes a real-time algorithm that delineates the significant part of the ECG signal, the QRS complex. The second layer, is near real-time and is invoked only when a potential QRS is detected, at a significantly lower rate that corresponds to the person heart rate. This layer is responsible for detecting R peaks, estimating the interval between two successive peaks and performs heart rate variability analysis in the frequency domain. Our system outperforms traditional ECG processing algorithms because the top layer completes well within the 5 msec sample inter-arrival period, ensuring that no samples are lost. The bottom layer can be delegated either to an underlying background task or a second processor. Because it is invoked less frequently than the top layer, it results in a lower interrupt rate, allowing for more flexible processing. [ABSTRACT FROM AUTHOR]