Atrial flutter is a life-threatening cardiac arrhythmia with an increasing prevalence during the last decennia. To characterise and diagnose atrial fibrillation based on the electrocardiographic waveforms, we propose an algorithm that extracts the f-wave from the observations. This algorithm is based on classical independent component analysis algorithms, but is applied in a transform domain of the original signal to assure that the source extraction contrast is discriminative for the source of interest. Results show that it is of high importance to choose the right domain before processing the signal. (4 pages)