EVALUATING THE ROBUSTNESS OF COMMERCIAL VIDEO CODECS ON WIRELESS NETWORKS J. Silvestre, P. Manzoni, M.P. Malumbres, J. Pons e-mail:{jsilves,pmanzoni,mperez,jpons}@disca.upv.es Department Computer Engineering. Technical University of Valencia. Camino de vera s/n POB 46020 Spain Abstract Currently, the research and development of multimedia networking systems is getting an great impulse due to the demand of applications and services from the entertainment and audio-visual industry. Both, networks and media coding systems are tightly coupled with this kind of applications. In particular, we are working with video transmission over wireless networks, where the poor quality of transport service of these networks becomes the task of sending video a very exciting challenge. In this paper, we will study the robustness of commercial implementations of the most popular video coding techniques (i.e.: MPEG-4, H.263 and MJPEG) when video is sent across a wireless ad-hoc network. We will analyze the impact on the reconstructed video quality of the behaviour of wireless network. Experimental results show that those video codecs that do not employ temporal redundancy (MJPEG) reduction are significantly more robust that the others. However, MJPEG offers very poor video quality when it works at low to very-low bit rates. Finally, H.263 video codec offers the best results in terms of robustness, but if video quality is the main target, then MPEG-4 (DivX) would be the best candidate. Key Words Video Coding, Wireless Networks, Robust Video Transmission, Performance Evaluation. 1. Introduction The coding and transmission of compressed video streams over existing and future communications networks with non-guaranteed QoS presents many exciting challenges. Media-based error recovery techniques are necessary for a wide range of applications/environments: interactive video over the internet, personal video communications over wireless networks, and digital video broadcasting over satellite and cable networks, to name just a few. This kind of techniques has recently received a lot of attention from researchers in academia and industry. In a noisy or packet lossy environment, as wireless ad-hoc networks [1], video error resilience techniques are necessary due to the nature of compressed video bit streams. For example, standard-based compressed video bitstreams employ Variable Length Codes (VLCs) as means of entropy coding. A single bit error present in VLC coded video data can lead to a loss of synchronization between the encoder and decoder, resulting in the loss of many video blocks. Multiple bit errors, which are usually due to burst channel errors [2] or to packet loss [3,4], may lead to the loss of partial or complete video frames, causing error propagation in the temporal dimension. This propagation is a direct result of motion compensation, which is usually used to reduce video temporal redundancies. In such network environments, video codecs should be able to protect the video compressed stream before transmission and conceal network errors during the decoding process. The ability to efficiently perform these tasks determines the robustness degree of a video codec. This performance metric needs to be taken into account for choosing/designing video coding systems specially suited for wireless multimedia applications. In this paper, we analyze the behaviour of several commercial video codecs. In particular, we will study their robustness when transmitting a compressed video stream through a wireless channel. The obtained results will allow us to determine the importance of the robustness behavior with respect to compression rate and video quality when delivering video over error-prone networks. The organization of this paper is the following: In section 2 we will present proposed wireless channel model. In sections 3 and 4 we describe the simulation framework for evaluating the robustness of video codecs and the evaluation results. Finally, in section 5, some conclusions are drawn. 2. Wireless channel model The wireless data channel is modelled using Log-normal Shadowing model [5]. This model is an evolution of the Log-distance path loss model takes into consideration that the surrounding environmental arrangement may be very