Study of finitistic channel models

The modeling of discrete-time stationary communication channels with memory was investigated. The approach was to develop distance measures to quantify the degree to which one channel or model approximate another and then to characterize the class of channels that can be well approximated by various types of models. A variety of channel distances were investigated. The results provide characterizations of the classes of channels approximable by finite memory models, primitive models, and finite state indecomposable models. The concept of channel entropy, which is the minimum amount of randomness needed to simulate a channel, was discovered. The results apply to both discrete and continuous alphabet channels.