The coherence between incident and reflected acoustic waves, with wavenumber k, scattered from a random rough surface, with root-mean-square roughness height h, decreases exponentially as khcosθ increases. The incidence angle, θ, and the surface roughness properties are set by the geometry and characteristics of the environment while the wavenumber range is governed by the signal bandwidth. The frequency-difference autoproduct is a nonlinear field construction capable of recovering reflected-field coherence by downshifting recorded frequencies, effectively decreasing the wavenumber. However, due to the quadratic nature of the frequency-difference autoproduct, dependence on the surface correlation function is introduced. This effect, and other relevant features, appear in the analytic form of the frequency-difference autoproduct derived for Gaussian-distributed surface roughness and surface correlation function. Previous results from simulated isotropic Gaussian surfaces and simple laboratory experiments verify the validity of this form. This presentation provides new results for autoproduct-based coherence recovery in fields scattered from surfaces exhibiting non-Gaussian correlation functions. Coherence recovery is quantified by the coherent surface reflection coefficient (ensemble-average reflected-field amplitude divided by flat-surface reflected-field amplitude), and comparisons are made to the Gaussian-based theory. Experimental results may be provided. [Work supported by ONR and an NDSEG Fellowship.]