OPTIMIZED AND PARALLELIZED PROCESSING ORDER FOR IMPROVED FREQUENCY SELECTIVE SIGNAL EXTRAPOLATION

摘要

In the recent years, multi-core processor designs have foundrntheir way into many computing devices. To exploit the capabilitiesrnof such devices in the best possible way, signal processingrnalgorithms have to be adapted to an operation in parallelrntasks. In this contribution an optimized processing orderrnis proposed for Frequency Selective Extrapolation, a powerfulrnsignal extrapolation algorithm. Using this optimizedrnorder, the extrapolation can be carried out in parallel. Thernalgorithm scales very good, resulting in an acceleration of arnfactor of up to 7.7 for an eight core computer. Additionally,rnthe optimized processing order aims at reducing the propagationrnof extrapolation errors over consecutive losses. Thus,rnin addition to the acceleration, a visually noticeable improvementrnin quality of up to 0.5 dB PSNR can be achieved.