University of Padova

Research Unit Leader

Prof. Giancarlo Calvagno (calvagno@dei.unipd.it) (PA)

Other Participants

Dott. Andrea DE GIUSTI

Dott. Simone MILANI

Dott. Ottavio CAMPANA

Specific program:

Transport layer cross-packet error protection for multimedia data

During the last years, the telecommunications world has assisted to a relevant increment of multimedia services and applications, such as the widespread diffusion of digital terrestrial television, the introduction of video messaging and video communication services over mobile networks, and the availability of on-demand video contents on the Internet. However, the need for transmitting video information over heterogeneous networks has consequently increased the need for coding the video signal in such an efficient and robust way that makes it available to a wider and wider set of users with different transmission capability. To this purpose, in recent times new video coding standards were defined in order to support a new ubiquitous and user-transparent multimedia connectivity. Nevertheless, the communication facilities that allow user to transmit without limits of time and place often turn out to be inadequate to transmit digital video signals since they provide a limited bandwidth. In addition, the probability of loosing some information is high because of the time-varying nature of channels (e.g., radio channels) and the need for delay constraints required by some applications. For example, in a interactive video communication the sent information is discarded as obsolete whenever it is not received within a defined time interval. These reasons imply that data retransmission, even if partial, turns out to be hardly efficient or unacceptable in some cases because of the extra visualization delay that it implies.

In the overall scheme of the project, the main purpose of this Research Unit is to optimize the coding algorithms paying particular attention to cross-packet FEC coders. To be more specific, the available bandwidth is partitioned into two parts: the first part is used to represent the video source while the second part is used to code some redundant information that is used to estimate the lost data in case some information is missing.

Therefore, we need to adopt efficient source coding techniques in order to compress the video signal keeping the visual quality experienced by the user at the desired level. The amount of saved bandwidth can be reinvested in the channel coding operation increasing the robustness of the transmission.

On the other hand, it is necessary to increase the error recovering capability of the adopted FEC schemes in order to make the whole transmission scheme more and more robust to errors and losses.

Therefore, in the life span of this research project we plan to develop the activities described in the following.

The first activity of the research program concerns the optimization of the error protection techniques. In addition to the to the intrinsic protection provided by the MD scheme, each description of the transmitted information must be protected by means of cross-packet FEC codes. The packets produced by the source coder are included in a matrix of variable dimensions, whose rows (or columns) are adaptively coded. The performances of the coding scheme are highly affected by the adopted code and by the matrix dimension. In this project we propose to develop an adaptive algorithm capable to optimize the performance of the coding scheme on the base of the input sequence characteristics. In particular, we aim at studying the improvements obtainable by means of a joint source-channel coding optimization. Moreover, we propose to analyze packet-labeling algorithms which are able to characterize the importance of the transmitted information in the decoding process. This classification will be used at channel level by the Unequal Error Protection (UEP) algorithms developed by the Research Unit of Genova. The entire coding scheme will be tested on a channel simulator using error concealment algorithms at the receiver. The performances of the entire scheme can be improved using frame buffer refresh algorithms, whose parameter must be conveniently chosen depending on the channel and input sequence characteristics. At the end of this activity, we will compare the possible different configurations of the proposed cross-packet FEC coding scheme with a coding technique based on motion vector subsampling. Since motion vector of adjacent frames are correlated, it is possible to produce two different descriptions by means of a quincunx subsampling of the motion vector set. As opposed to other schemes based on a pre-processing of the video signal, this solution does not modify the coding performances of H.264/MPEG-4 AVC since it works on already coded data [Kim01]. To this purpose, we propose to develop a low computational cost implementation of such a scheme, so that it can be compared with the cross-packet FEC scheme.

Particular attention will be devoted to the H.264/MPEG-4 AVC coding standard, which have been recently improved and named FRExt. As for most of its predecessors, this standard defines a hybrid transform coder using motion compensation. With respect to the previous standards, the innovative characteristic is the introduction of a spatial prediction, which allows to increase the coding gain in such cases where the motion compensation is not effective (e.g., when coding reference frames or when scene changes occur). Moreover, the new standard uses an integer transform with low computational cost and a motion compensation using variable size blocks, which makes it very convenient for multimedia transmission over wireless networks. Recently the standard has been modified to improve its performance by increasing the number of possible coding modes available to the coder. Therefore, it is necessary to develop low computational complexity algorithms which allow to optimize the coding parameters choices to maximize the quality of the reconstructed images. To this purpose, we propose to investigate and develop low computational complexity algorithms for the optimization of the Rate-Distortion function and for the optimal adaptive choice of the transforms dimension. In addition, we will investigate a possible joint optimization with the multiple description scheme that will be developed by the Research Unit of Udine. In fact, the correlation introduced among the different descriptions implies a similar correlation among the respective coding parameters. The procedure used to determine the coding parameters for each description can influence the effectiveness of the lost information estimate and the computational complexity. It is therefore possible to jointly optimize such choice taking into account the characteristics of the information fed to each coder (i.e., the characteristics of the descriptions). One additional possible improvement can be obtained by considering also the transmission channel characteristics.

At the same time, we will investigate the possible improvements to the source coding algorithms, which can be possibly proposed for inclusion in the definition of the last-generation standards.

Some possible areas of investigation include:

- an improvement of the motion compensation based on the use of variable geometry blocks [Wan96];

- the parametric coding of the motion vectors;

- the development of a flexible arithmetic coding algorithm which allows for a graceful degradation of the reconstructed sequence video quality when a reduced bit-rate is available [Tau00].

The third research activity concerns the implementation of a set of software procedures that will demonstrate the coding techniques described above.