Hi Hengamesh, this is a very informative tutorial,may be this can help you.

http://research.microsoft.com/en-us/um/people/zhang/INRIA/Publis/Tutorial-Estim/Main.html

Regards,

Divya

Ther are two levels of filtering. The first is to design analogue filters which filter any undesired noise from the the signal to be measured. This ensures that the digital signals obatined from the D/A converter may be correctly used to conduct any further digital filtering / processing using standard alghorithms etc.

As a miniumum you have to sample at least twice the maxiumum frequency of the highest frequency you will require and based on this you will then design the analgue filters to cut off or filter all frequencies higher than this to prevent aliasing.

Thus say the highest frequency you require is 1000Hz then the analogue filters should be designed to filter from 1000Hz upwards. This will ensure good signals for all frequencies < 1000Hz.

You may use high, band or low pass filters to remove or preprocess the anaolgue signals. The rule is measure junk signals obtain junk results.

see

http://en.wikipedia.org/wiki/Anti-aliasing_filter

http://www.maximintegrated.com/app-notes/index.mvp/id/928

Thank you. Actually I use numerical filtering methods for parameter estimation and I wanted to have literature about some more famous used numerical methods for machine parameters estimation.

It makes no difference as in the real world you still have to use analogue filters which should be included in your simulation before digital filtering is done as they introduce phase delays etc.

Thank you. I mean inserting laboratory measurement in a file and applying numercal methods to parameter estimate. I do not have some reference to use both filtering and the reason of that. I have something like this:

http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=744405

When you take any analogue measurements from the laboratory you will first have to filter the data using analogue filters to remove noise or unwanted higher frequencies. This is standard and any digital data acquisition system will have to do this. If not the data you store in your file might contain noise from higher frequencies which will appear as a low frequency (aliasing). You might then have problems when using the file data in any parameter estimation algorithm as the results are only as good as the data it is provided ie junk in - junk out. No matter how good the algorithm it will yield incorrect results.

Do some reading on digital signal processing and you will see this is one of the basic requirements.

Do you have an paper about your comment (analogue measurement then analoque filtering then digital filtering) Actually we just measure then we use smooth high frequency filtering because of more than 1 or 2 high frequencies in noise that is a kind of analogue filtering because we have more than 2 data not just 2 levels of data. It is said by a specialist that 2 analogue and digital filtering does not have necessity and is not good in this case but for me it is important to know if there is a reference.

Dear Hengameh,

There are advanced numerical techniques to do digital filtering to estimate parameters. But there are certain conditions (or necessities) for such an approach. You need to have input and output data (time domain samples) for the estimation. With this data, you can develop a dynamic model between the input-output pair. So, the filter effectively becomes an estimator. And whatever is left out of the estimator is usually noise. But there are certain tricks to it. One challenge is to determine the proper system model (Auto-regressive, moving average etc). Let me know if you are interested in such an approach.