I'm using tfest() to do some system ID on time-series data with mixed success. Take a look at this bode plot of the original data compared to the estimated two-pole, one-zero transfer fcn. There's obviously a much better way to fit this data.
I've tried using opts = tfestOptions('WeightingFilter',[5.28, 668]); to try to force the estimator to only look at the data frequency range, but that doesn't seem to change anything. Any ideas?