State-space model estimation

[NR70]

Hello:
I estimate different state-space models in Eviews and I get the estimation of the final values of the state variables with the s.e., z-stat, prob., just fine, as well as the filtered and smoothed state and signal series.
However, from time to time, in some cases, when I re-estimate exactly the same state-space model as before - without effecting any change of any kind either to the state-space specification, the initial state variance matrix, initial state values, or the estimation sample, observed series...etc - I get the message "WARNING: Singular covariance - coefficients are not unique", whereas the first few times I estimated the model I did not get this message. The estimated smoothed signal series are exactly the same as before (prior to getting the warning message when estimating the state space model).
Can you please let me know why for exactly the same state space model, initial conditions and observed series, there isn't sometimes the singular covariance matrix warning and sometimes it is there?
Thank you
[I use Eviews 7]

[EViews Glenn]

http://forums.eviews.com/viewtopic.php? ... ues#p18944

http://forums.eviews.com/viewtopic.php? ... ues#p16687