On the Capacity of Data-Dependent Autoregressive Noise Channels

Data-dependent autoregressive noise channel models capture the nonlinearities and the first and second order noise statistics of magnetic recording channels. Recently, an iterative Monte Carlo algorithm to compute tight lower bounds on the capacity of intersymbol interference channels was proposed. This algorithm is readily applied to compute capacity lower bounds of data-dependent autoregressive noise channel models. In this work, we extract data-dependent autoregressive noise channel models from simulated micromagnetic perpendicular recording channel waveforms, and then optimize the source distribution to compute capacity lower bounds of the channel models. We note that the lower bound is a bound on the model. Finding an accurate model that fits more than just first and second order statistics is still an open problem. The method in this paper is an accurate estimate of the magnetic recording channel capacity only if the channel may be represented by a finite-state machine with up to 2 states (for computational complexity reasons), which applies only to magnetic recording at symbol densities less than 4 symbols per PW50.