Exact synchronization for finite-state sources

From MaRDI portal



Abstract: We analyze how an observer synchronizes to the internal state of a finite-state information source, using the epsilon-machine causal representation. Here, we treat the case of exact synchronization, when it is possible for the observer to synchronize completely after a finite number of observations. The more difficult case of strictly asymptotic synchronization is treated in a sequel. In both cases, we find that an observer, on average, will synchronize to the source state exponentially fast and that, as a result, the average accuracy in an observer's predictions of the source output approaches its optimal level exponentially fast as well. Additionally, we show here how to analytically calculate the synchronization rate for exact epsilon-machines and provide an efficient polynomial-time algorithm to test epsilon-machines for exactness.


The synchronization process for a class of finite-state hidden Markov models, the \(\epsilon\)-machines, is analyzed. An \(\epsilon\)-machine is exactly synchronizable if the observer synchronizes to almost every sequence generated by the machine in finite time. One proves that an observer synchronizes exactly to the internal state of any \(\epsilon \)-machine exponentially fast. The synchronization rate can be computed analytically. The observers average uncertainty in the machine state as well as the average uncertainty in predictions of the future symbols decay exponentially fast. At the end of the article, necessary and sufficient conditions for exactness of an \(\epsilon\)-machine together with a test-algorithm are provided.











This page was built for publication: Exact synchronization for finite-state sources

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q658481)