VLBI imaging Algorithms

Very long baseline interferometry (VLBI) is a radio-interferometric observing technique with unmatched spatial resolution. A (global) array of antennas records the signal observed from a single source that is unresolved by single dish observations. The correlation product of the signal recorded at each antenna pair at the same time (corrected for time lag) is proportional to the Fourier transform of the true sky brightness distribution with a spatial frequency determined by the baseline separating the two antennas. As the Earth rotates, the baselines rotate in Fourier domain (uv-domain) on typically elliptical tracks. However, due to the limited number of antennas in the array and limited observation time the coverage of measured Fourier coefficients (uv-coverage) is very sparse.

Various imaging approaches have been suggested in the past, starting from a greedy matching pursuit approach (CLEAN) that is the de-facto standard method by now, to current developments of Regularized Maximum Likelihood (RML) methods and Bayesian methods. My own research is based on the compressed sensing approach: We are modelling the image by a dictionary of wavelets that is fitted to the uv-coverage. This allows a better separation between covered image features and sidelobes introduced by gaps in uv-coverage and offers a pathway towards unsupervised VLBI imaging.

As a complementary approach we study VLBI imaging by evolutionary algorithms. These algorithms can explore the complete landscape of reasonable solutions and classify them. Moreover, we study several extensions to the classical imaging problem, i.e. polarimetry or dynamic reconstructions (movies).

Check out our VLBI software: