Stochastic context-free grammars and hidden Markov models for modeling of bursty channels

In order to design good error-control schemes for bursty channels and also to facilitate performance analysis, it is important to develop accurate and simple statistical channel error models for the channels of interest. We propose two novel generative methods to model the end-to-end error profile of radio channels described by long well-defined error bursts interleaved with long error-free intervals. The first method makes use of the power of stochastic context-free grammars to model palindromes. The second utilizes simple hidden Markov models with specific structures, which are suggested by the ideas presented in the first method. Both methods achieve much better performance than previously proposed approaches without introducing more complexity. Although the complexity of the second method is slightly greater than that of the first, its advantage is that it can be easily applied in decoding implementations specifically tailored to deal with bursty channels.