Abstract
As mHealth (Mobile Health) thrives, advances in collection and analysis of vital signals on portable devices have become more and more important. In this thesis, a low latency and low end effect on-line segmental empirical mode decomposition (SegEMD) is proposed, which aims at the adaptive characteristics for nonlinear and non-stationary signal of the conventional empirical mode decomposition (EMD), as well as the notorious lengthy latency and highly demanded computing resources of it. The SegEMD is capable of processing continuous signals segment by segment with EMD, by reusing the previous slopes, the previous data and the estimation of signal characteristics in advance. Worst normalized mean squared error (NMSE) compared to the results carried out by the conventional EMD is less than 9%. For an 8-hour overnight electrocardiogram (ECG) signal, the processing time is twice the conventional EMD, but the memory requirement is reduced to below 1%. Compared with SEMD, the processing time is 83% less and the memory used is 63% less. Thus the proposed SegEMD is especially suitable for limited hardware, such as smartphones, tablets, and other mobile devices. SegEMD is also applied to bring out the ECG-derived respiratory (EDR) from ECG, which is able to reveals the status of respiration without the uncomfortable sensors in the traditional polysomnography (PSG). RenderScript, a novel application programming interface (API) of the Android OS is introduced to provide acceleration of the Android application (App). A 10-second, 1200 samples data sequence can be detrended and the EDR be extracted within 2 seconds in a Nexus 7 tablet. Such app can be installed on any Android devices with version 4.3 and later version, roviding possibilities to personal healthcare at home.