Joint diagonalization: Is non-orthogonal always preferable to orthogonal?

This paper deals with the efficiency of orthogonal versus non-orthogonal joint diagonalization techniques for blind source separation. The question is: are non-orthogonal algorithms always preferable to orthogonal ones in practical noisy conditions? The two techniques are compared in terms of separation performance through numerical simulations with different levels of additive noise and numbers of samples. Both fourth-order and second-order blind source separation contexts are investigated.