
328
J. Skowronski and L Dologlou
Since, in general, Sr is not an observation matrix, that is it does not exhibit the particular
Hankel structure, the reduced rank matrix Sr has to be projected onto the subspace of
Hankel matrices. One way of doing this, is to replace the elements of each antidiagonal by
their mean value. Obviously the projected matrix is not a reduced rank matrix anymore,
but it has been shown in [2] that it is closer to a reduced rank matrix than St. By repeating
this procedure, we obtain an assymptotically reduced rank observation matrix which lies
'close' to the original matix. ...