Wireless digital transmission systems are vulnerable to burst errors which directly affect the underlying bitstream. If resources for channel coding are limited, the use of corrective channel coding is not always feasible. This is especially the case for systems which perform under very low delay conditions since channel coding normally would increase their overall latency. If the transmitted signal is an audio signal, errors in the bitstream will be perceived as artifacts. In practice the audio signal therefore is muted for the duration of the gap which only can reduce the artifacts and a perceived degradation of the signal remains.
Error concealment solves this problem by generating a replacement signal that not only fits into the gap but makes it unnoticeable. A real world scenario might require that gaps of up to 100 ms have to be concealed; however, the gap length is of unknown duration at the time of occurrence. In addition, live applications have critical demands for the overall delay of the concealment system that have to be met.
Our goal is an error concealment system which is able to perform under realtime conditions and at the same time does not add any additional delay. In this work we assume that the delay constraint is crucial. We focus our research on model based approaches which are able to perform sample based recursively using only past samples.
Two model based methods for signal extrapolation have been developed. Among these is the usage of adaptive Kalman filtering which can be processed in realtime without any delay. We also improved an existing approach based on linear prediction by including adaptive parameters and a variable concealment break-off.
