Two approaches to combined source-channel coding: A scientific competitionin estimating correlated parameters

After source coding of speech, audio, images and video signals in digital t ransmission, there is often some residual redundancy left due to complexity and delay constraints. A scientific competition compares two ways of explo iting this redundancy in order to improve the end-to-end quality of the tra nsmission over a noisy channel. One way is to use a new class of channel codes, which are called Source Opt imized Channel Codes (SOCCs). These codes are designed by an optimization p rocess, which takes into account the source and channel statistics as well as a quality measure. At the receiver the code redundancy is not exploited explicitly for error correction, but for the support of parameter estimatio n using Soft Bit Source Decoding (SBSD). Designing SOCCs with respect to th e parameter signal-to-noise ratio (SNR) is considered in detail. Another way is to use a channel coding scheme employing convolutional codes with Unequal Error Protection (UEP) achieved by puncturing. The influence of each bit of a quantized parameter on the signal quality after source dec oding is assessed. This approach is improved by exploiting two kinds of a p riori knowledge in the channel decoding process. These are the time correla tion and the correlation due to the distribution of a parameter. This metho d is called Source Controlled Channel Decoding (SCCD). Furthermore, we cons ider the bit-mapping and its effect on decoding. The mutual information is used to estimate the gain of decoding with a priori information.