In this paper we propose a new sinusoidal model tracking algorithm that implements a non-time-progressive way of data processing. Sinusoidal partial parameters are estimated in the consecutive frames; however, the order of establishing individual connections between partials is determined by a greedy rule within the whole signal or within a specific time window. In this way, the strongest connections may be determined early, and subsequent predictions of each trajectory evolution are based on a more reliable partial evolution history, compared to a traditional progressive scheme. As a consequence, the proposed non-progressive tracking algorithm offers a statistically significant improvement of obtained trajectories in terms of better classic pattern recognition measures
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.