Dear Anastasia. 

Damping is a sensitive parameter to estimate. There are several explanations for why unreasonable damping estimates are obtained by ERA/OKID or any other output-only system identification technique assuming white noise excitation. One possible explantation is wheather the dynamic model assumed in the technique is representative of the vibrations of the structure in question. If you find that the model is a good approximation, the remaining explanations to the damping ratio estimated could be related to the identification procedure i.e. what model order was chosen, how many time lags did you include for the construction of the block-Hankel matrices. 

Anela

You are right, I think it's a question of noise and model sufficiency. I tried various values for  'p'  (time lags) and kept the number of 'non-singular' values of the Hankel matrix quite large. I noticed that for values of p larger than 6 times the approximated order of the system the identification algorithm fails to provide meaningful results.

Thank you for your answer, Anastasia

Yes, for low signal-to-noise ratios the damping estimated will be unreliable. The relation between the number of time lags and the model order which you have observed is specific to the structure you are analyzing and the data acquisition, that is the natural frequency, damping ratio, record length and sampling frequency.  You may have observed that already. 

Good luck with your analysis. 

Anela

Yes sure, unfortunately I noticed there is no rule of thumb for the p value, although one would expect that the larger the p the more stable the result. Thanks again :)

I assume that the output correlation function matrix estimate that you have provided, is based on the unbiased estimator. In that case the variance error increases for increasing time lags. Hence the results are not more stable with increasing number of time lags included. 

No problem. Feel free to ask.