On causal source codes with side information

We study the effect of the introduction of side information into the causal source coding setting of Neuhoff and Gilbert. We find that the spirit of their result, namely, the sufficiency of time-sharing scalar quantizers (followed by appropriate lossless coding) for attaining optimum performance within the family of causal source codes, extends to many scenarios involving availability of side information (at both encoder and decoder, or only on one side). For example, in the case where side information is available at both encoder and decoder, we find that time-sharing side-information-dependent scalar quantizers (at most two for each side-information symbol) attains optimum performance. This remains true even when the reproduction sequence is allowed noncausal dependence on the side information and even for the case where the source and the side information, rather than consisting of independent and identically distributed (i.i.d.) pairs, form, respectively, the output of a memoryless channel and its stationary ergodic input.