The breakdown point of signal subspace estimation

The breakdown point of signal subspace methods, which is the SNR below which the algorithm's performance deteriorates dramatically, is intimately related to the breakdown point of PCA based signal subspace estimation. We shed new light on this breakdown point for a broad class of signal-plus-noise models, provide a transparent derivation that highlights its origin and verify the accuracy of the high-dimensional predictions with numerical simulations for a moderately sized system.