Estimating the information rate of noisy two-dimensional constrained channels

The problem of computing the information rate of noisy two-dimensional constrained source/channel models has been an unsolved problem. In this paper, we propose two Monte Carlo methods for this problem. The first method, which is exact in expectation, combines tree-based Gibbs sampling with importance sampling. The second method uses generalized belief propagation and is shown to yield a good approximation of the information rate.