Estimating AR1 Signal (2)

Home

Let's continue the AR1 discussion. Underestimating p lead to this kind of problem:

The signal was buried in i.i.d Gaussian noise:

At this stage, the RE would be -0.25, so this is not a good reconsruction (RMSE is 0.57). Let's plot the first figure again, without the error limits:

This estimate (reconstruction) gives RE of 0.4 (RMSE 0.4). So, according to RE, heavy low-pass filtering improves the results. But this is very very dangerous.