Phase linking is a prominent methodology to esti-mate coherence and phase difference in interferometric synthetic-aperture radar. This method is driven by a maximum likelihoodestimation approach, which allows to fully exploit all the possibleinterferograms from a time series. Its performance is howeverknown to be affected by the accuracy of the covariance matrixestimation step, which usually requires to introduce additionalprior information on its structure when there is a small samplesupport (spatial window). Moreover, most phase linking algo-rithms are built upon the sample covariance matrix, due to theassumption of an underlying Gaussian distribution. In a scenariowhere SAR data is high resolution, or when the study area isspatially heterogeneous (e.g., urban area), this assumption canalso limit the accuracy of the covariance matrix estimation step.Considering the two aforementioned issues, we introduce alter-native statistical models, whose maximum likelihood estimatorsthen yield new phase linking algorithms. In order to be robust tonon-Gaussian data, we consider the use of a more general modelof scaled mixture of Gaussian. To address small sample supportissues, we also generalize this approach to a possibly low-rankstructured covariance matrix. A unified algorithm to performphase linking given these models is then derived and validatedby simulations and a real data case (Sentinel-1 data).