We describe a multidimensional maximum likelihood estimator for radio channel parameters. We also derive a data model to describe the complete data, that is virtually applicable to every antenna array geometry. The proposed iterative gradient based algorithm has been developed, since algorithms using component-wise updates of the likelihood function shows a slow convergence, if at least two propagation paths with nearly the same parameters exist in the measured radio propagation scenario, that means if super-resolution is necessary. The algorithm provides furthermore a variance estimate of the estimated parameters, since the Fisher-information matrix is calculated throughout the alcorithm.