Beyond Countable Alphabets: An Extension of the Information-Spectrum Approach

摘要

A general approach is established for deriving one-shot performance boundsfor information-theoretic problems on general alphabets beyond countablealphabets. It is mainly based on the quantization idea and a novel form of"likelihood ratio". As an example, one-shot lower and upper bounds for randomnumber generation from correlated sources on general alphabets are derived.