Summary: Error Exponents in a Source Coding Problem of Korner *
School of Electrical Engineering
Ithaca, NY 14853, USA
Consider a discrete memoryless stationary information source with nite alphabet.
Suppose that the letters emitted by the source are not all distinguishable. This might
be the case, for example, if the symbols were handwritten letters. Distinguishability be-
tween letters is a symmetric relation and thus determines a graph on the source alphabet.
Similarly, one talks of distinguishability between xed length strings from the alphabet.
Korner considered the problem of block coding for such a source with the requirement that
only indistinguishablestrings can be assigned the same codeword. He found a fundamental
quantity, the graph entropy, which gives the minimum rate at which coding can be done
with vanishingly small error probability. In this note we compute the error exponents for
this source coding problem.
* Research supported by NSF PYI award NCR 8857731, an IBM Faculty Development
Award and by BellCore Inc.
Error Exponents in a Source Coding Problem of Korner *