Rather than the broad mixture that is RQ - maybe try a tighter mixture e.g. a Gaussian centred on a particular lengthscale?