Optimal Unequal Channel Protection for Multiple-Description Concatenated Codes

Multiple-description product codes, consisting of an erasure correction row code (e.g. Reed-Solomon code) and an error correction/detection column code (e.g. convolutional codes+CRC) have recently been introduced to protect progressive transmission of multimedia contents. The significance of this approach is its compatibility with and full use of the progressive nature of the rate-distortion scalable source code, which is a much desired provision for quality of service. It provides a robust and scalable means of communication over packet-lossy networks and also extends itself nicely to hybrid scenarios consisting of a packet erasure wireline backbone and a wireless interface. This approach constructs a multiple-description, progressive code, with all descriptions having equal importance. In other words, the reconstructed signal when any K of the total N of these descriptions are received, would be the same, regardless of which packets and in which order are available at the receiver. Probably, this property of equal importance has led designers to adopt equal channel protections for all those equally important descriptions, even when facing varying channels. But this intuition is suboptimal. This paper will shed some light on the optimum channel protection strategy of the description packets for varying channels without feedback, where the channel statistics are only known to the transmitter. For a set of idealized assumptions the globally optimal channel protection assignment is found. For more practical scenarios, the approach provides a systematic means of finding a nearly optimal packet protection strategy.