Allocating Complexity Between Source and Channel Coding

Given an available number of arithmetic operations per data sample, the tandem connection of a transform source code and a Reed-Solomon channel code is optimized to minimize meansquared error when transmitting Gauss-Markov data across a binary symmetric channel. The goal is to learn how well-designed systems allocate complexity, i.e. arithmetic operations, between source and channel coding. The tandem system considered here is chosen to be as representative as possible. For the range of source and channel parameters tested, it is found that when at least moderate complexity is available, a well-designed system allocates 10 to 35% of its complexity to the source code.