Progressive Source Coding Combined with Regressive Channel Coding on Varying Channels

Joachim Hagenauer, Thomas Stockhammer, Christian Weiss, Anton Donner Institute for Communications Engineering (LNT) Munich University of Technology (TUM) e-mail: fhag,tom,chris,donnerg@lnt.ei.tum.de WWW: http://www.LNT.ei.tum.de Abstra t | A channel coding system for strongly varying channels (mobile radio, Internet) unknown to the encoder is presented. It matches progressively coded sources (e.g., video, images, audio, speech) with convolutional codes of very high memory applying regressive redundancy over the data frame of the source. The channel decoder with scalable complexity and delay employs modified sequential decoding. The decoder uses a new algorithm, the ’far end error decoder (FEED)’ which in a change of paradigm does not aim at low error rate, but rather makes the first error as far out as possible under the actual channel conditions. We determine the error-free region of the frame which means that for progressively coded sources we achieve the best reconstruction quality possible. The decoding method is self-adaptive to varying and unknown channel conditions (interference, fading, packet loss) and provides graceful degradation. Results with SPIHT coded images show performances better than previous known FEC schemes. Potentials for further improvements of this scheme are discussed.