New bounds on the capacity of certain infinite-dimensional additive non-Gaussian channels

An additive non-Gaussian noise channel with a generalized average input energy constraint is considered. New bounds on the channel capacity are found for the case that the divergence of the probability measure induced in function space by the noise process, with respect to the measure induced by the Gaussian process with the same covariance as that of the noise process, is finite. Upper and lower bounds that depend on the noise process only via the divergence are given for large signal-to-noise energy ratio S. It is also shown that the increase in the capacity of an infinite-dimensional non-Gaussian channel, relative to the infinite-dimensional Gaussian channel capacity S/2, could be significant