The performance of any intracardiac electrogram processing method is limited by the accuracy of its activation detection approach. The most common activation detection approaches in the literature aim to find the highest peak in the activation envelope disregarding the start and end points. However, the duration of the activation can be used to extract useful information such as wave collisions. In this work, we propose a novel orthogonal based approach for fast and accurate estimation of the start and end of the activations (activation envelope) in intracardiac recordings during atrial fibrillation. Wavelet decomposition of the signals was used to create a pool of basis functions for the proposed modeling method. The database included 24 recordings of approximate length of 6s obtained from atrial endocardium of 5 patients who underwent catheter ablation therapy. The start and end of activations in each electrogram was manually annotated by an expert electrophysiologist and the annotations were used as a gold standard to calculate the performance of our envelope detection method. The results show promising performance and excellent robustness to training data for our proposed method with respect to envelope estimation error.