Universal Class-Based Block Source Coding for Discrete Memoryless Sources

We introduce a general class-based block source coding framework which, for the case of discrete memoryless sources, is specialized to four different constructions with varying levels of underlying structure in the class definitions. For all the constructions, we show that the matched random coding error exponent is achieved universally with a corresponding universal decoder. Increasing the structure in the classes results in a simpler corresponding universal decoder.