Optimization of Video Streaming over 3G Networks
VIDEO streaming over cellular networks has been made possible in the last years by better performing video codecs and wireless cellular networks oriented to data transmission. The interaction between two heterogeneous worlds, the telecommunication infrastructure and the coding video software, calls for advanced optimization mechanisms. The actors involved in the optimization process are the cellular system’s access network, UMTS and HSDPA, the wireless transmission channel and the fi nal user equipped with a mobile device capable of decoding video sequences. The knowledge and characterization of each of the building blocks allow the optimization of each element to the specifi c needs of the others. This doctoral thesis discusses three main contributions. In the fi rst part, the e ffects of transmission errors on video streams are analyzed. Incorrectly received video packets are usually discarded by the lower layers and not conveyed to the application. Part of the payload, however, still contains valid encoded information. Only the data stored after the error occurrence cannot be correctly interpreted, because of the variable length coding. Several error detection strategies at application layer are analyzed and compared. Finally, a proposal for a more efficient sorting of the encoded information is discussed. The second part of the thesis deals with the optimization of a speci c service: soccer video streaming. Soccer represents one of the most attractive contents, as nomadic users are interested on receiving this content live in their mobile device. The way human users evaluate the quality of this speci c video service is the key issue driving the study. By means of unsupervised segmentation, three regions of the frame have been identi fied: the fi eld, the players with the ball, and the audience. As the three regions have a diff erent impact in terms of subjective quality, their rate-distortion behavior has been optimized by assigning more data and, therefore, increasing the quality of the subjectively most important items. In the third part, two cross-layer mechanisms are presented. As fi rst, the application of SP and SI frames, specially encoded frames able to limit the temporal error propagation, has been discussed as an alternative to packet retransmission. The proposed optimization has been performed considering the measured error characteristics of the UMTS DCH. The second investigation is dedicated to the optimization of video streaming over HSDPA networks. Within a single video data stream, packets with di erent importance are discriminated. Exploiting secondary PDP contexts, several logical connection with diff erent quality settings can be established. By means of header ltering, each packet is transmitted utilizing the appropriate channel, privileging the most important payloads.