Nuclear norm minimization algorithms for subspace identification from non-uniformly spaced frequency data

The nuclear norm is an effective proxy for matrix rank in a range of minimization problems, including subspace identification. Nuclear norm-based methods are implemented via iterative optimization methods and in problems with very noisy data the quality of the nuclear norm-based estimate may warrant the additional computation cost. We present two methods (based on the dual accelerated gradient projection and the alternating direction method of multipliers) for nuclear norm based subspace identification in the case where the data is given as irregularly spaced frequency samples.