The real-time and wireless mobile has become the trend of electrocardiogram (ECG) monitoring system for atrial fibrillation (AF). At present, the ECG with multi-leads (12 leads) is needed by most of AF signal extraction algorithms in order to extract effective AF waves. However, it is not very convenient for patients' movements in a multi-lead ECG monitoring system. Although the traditional template matching method is for single-lead ECG extraction, it is less robust than blind source extraction algorithm, and is affected severely by noise. In view of this,we put forth a new real-time algorithm for extracting AF from the singlelead ECG, using non-stationary heartbeat series during AF to extend dimension (segmentation), and then applying a blind source extraction algorithm to extract the effective AF signal. Experiment results showed that this method could be used to extract AF signal effectively from a single-lead ECG data. Therefore, it is suitable to apply this method to Wireless Monitoring System using single-lead ECG.