The continuous evolution of electrocardiography (ECG) recording has enabled the successful development of many significant applications of this vital signal for clinical diagnosis and monitoring. Recent trends in device miniaturization and wireless transmission have extended the uses of such a signal modality to telemedicine for home cares. However, it has posted new technical challenges for the scenarios of home users and associated business models. Among them, a smart algorithm for real time or early indication of critical cardiac conditions, e.g. ventricular fibrillation (VF), Ventricular Tachycardia (VT), are extremely important for sharing the work loads of remote side for proper responses and delivery of health care. It would be also rather critical to fit such a computation task into the wearable or mobile device for the requirement of low power consumption. In this paper, we report the development of a novel analysis algorithm based on Time-Delayed Phase Space Reconstruction (PSR) method to differentiate abnormal ECG segments from entire records. We used BIHMIT arrhythmia database and CU database to verify our original ideas. According to our test results, this algorithm successfully identified the three heart diseases of PVC (Premature Ventricular Contraction), VF (Ventricular Fibrillation) and VT (Ventricular Tachycardia) immediately. We calculated the statistic parameters to estimate the efficiency: the average of sensitivity is 98.7% and the specificity reaches 96.2%. We also implemented this algorithm for wearable applications in a single-chip micro controller (MSP430, TI) for arrhythmia ECG data. The total code size is about a few hundred bytes and the execute time meets the order of sub-second. This new algorithm provides a powerful real-time index for clinical diagnosis and long-term home-care applications.