Performance Comparisons of Frame Dropping Algorithms for DVS in Battery-Aware Low-Power Video Decoding

Dynamic voltage scaling (DVS) is an effective technique for low-power video decoding in battery-aware mobile devices. Once a frame is decoded, the processor waits until its deadline, when the frame must be played out. This waiting time is used to exploit the advantage of DVS. However, the predicted decoding time of a frame may exceed its deadline (or called expected decoding time, EDT). The decoding of a frame may be deferred or even the frame may be dropped when frames with large predicted decoding times occurred in a burst. This paper proposes two types of frame dropping algorithms to maintain the energy efficiency of DVS in MPEG decoding; meanwhile, to avoid consecutive frame dropping to preserve the quality of video playing. The difference between these two algorithms is determined by when and how to drop a frame. The first algorithm, called Late_dropping algorithm, selects a frame to drop when the deferred decoding time of a frame is larger than one EDT. And the second algorithm, called Early_dropping algorithm, selects a frame to drop when the deferred decoding time of a frame is larger than one half of an EDT. Performance comparisons between these two algorithms have been investigated, using a heavy burst interval in a sample video clip as an evaluation basis. Although Early_dropping algorithm produces almost twice more dropping situations than that of Late_dropping algorithm, it acquires better DVS application and thus saves more battery energy to about 20% than that of Late_dropping algorithm. Evaluation results also show that for both algorithms frame dropping will never be occurred in consecutive.