The third generation of mobile systems brought higher data rates that allow for provisioning of multimedia services containing also the video parts. The real-time services like video call, conferencing, and streaming are particularly challenging for mobile communication systems due to the wireless channel quality variations. The mechanism for video compression utilizes a hybrid of temporal and spatial prediction, transform coding and variable length coding. The combination of these methods provides high compression gain, but at the same time makes the encoded stream more prone to errors.
In this thesis, techniques for error resilient transmission of video streaming over wireless mobile networks are investigated. Focus is given to the recent H.264/AVC standard, although the majority of the proposed method apply to other video coding standards, too. First part is dedicated to exploiting the residual redundancy of the received video stream at the decoder. The redundancy is used for error localization within a damaged packet that would be discarded otherwise. Error detection using syntax analysis and detection of impairments in the picture domain do not require any additional rate while reducing the resulting distortion. As their detection performance is limited, alternative methods assisted by an encoder are further proposed. The experimental results show that particularly the resynchronization of the variable length code, facilitated by out-of-stream signalized side information, improves the performance considerably.
In the second part of the thesis, error concealment methods are reviewed, compared, improved and proposed. There is no single error concealment method performing best in all situations. Temporal interpolation typically performs better than spatial, unless scene change occurs. Therefore, an adaptive mechanism is proposed that chooses from several temporal and spatial methods according to the situation. A scene change detector based on dynamic thresholding, working independently of encoder settings, is also a part of the proposal. In spite of its low complexity, the proposed mechanism provides an impressive quality gain.
The final part handles the cross-layer design for UMTS (Universal Mobile Telecommunications System). It is shown that utilizing the radio link information at the application layer of the receiver can substantially improve the quality at practically no costs (no rate increase, no additional complexity). Moreover, it is shown, that UMTS radio link layer errors are predictable. Differentiation between the intervals with high and low probability of error allows for assignment of different priority level to the video packets of one user, based on semantic information or expected distortion. The scheduler disposing with such information considerably improves the resulting video quality.