| In order to combat channel failures in data communications, multiple description coding (MDC) produces two or more equally important bitstreams or descriptions, and transmits them independently over erasure channels. If only one description is correctly received, a coarse copy of the source is obtained. The more descriptions correctly received, the finer the accomplished quality. When all descriptions are correctly received, the transmitted signal can be completely reconstructed.;The proposed ERMDC encoder maximizes the Hamming distance between used codewords in MDC, so as to make as many errors as possible detectable at the decoder. In order to reduce the reconstruction distortion, the proposed ERMDC decoder can detect binary transmission errors and estimates their output values in two means: (i) one is MSE-optimal, but requires information about channel conditions; (ii) the other is suboptimal, but does not require channel conditions. The ERMDC achieves graceful performance degradation associated with BERs, and outperforms classic MDC when meeting with both packet losses and bit errors.;In order to avoid long time of design optimization, simplified index assignment (IA) algorithm for easy ERMDC encoder design is developed. This algorithm obtains "close-to-optimal" solutions as well as low computational complexity. Furthermore, this IA algorithm can be extended to embedded coding in progressive transmissions.;Moreover, we study performance of the ERMDC over Rayleigh fading channels by utilizing modulated signals as inputs. We also discuss usages of the ERMDC and its system-level performance over channels with both packet losses and bit errors. Experimental results show that, in general, the ERMDC system outperforms classic MDC systems.;In this work, we apply MDC to accommodate multimedia transmissions over hybrid wireline-wireless networks, which require low delay and high robustness against both packet losses and bit errors. In addition to the classical MDC channel model, i.e., on/off channels, we study channels that are also suffering from bit errors. Based on this channel model, we design what we call ERMDC or error resilient multiple description coding. |
