On the capacity of data dependent autoregressive noise channels

Summary form only given. A magnetic recording channel can be very accurately modeled as a data dependent finite-state machine. Recently, Monte Carlo methods have been proposed to optimize Markov source inputs and compute tight capacity lower bounds. In this work, we fit data dependent autoregressive noise models (with window sizes I and D, and noise memory length L) to perpendicular recording channels and evaluate the capacities of the models. For each model size, we optimize the Markov source distribution using the iterative algorithm proposed by Kavcic (2001), and compute the information rate to lower bound the capacity.