0
$\begingroup$

This can be very general question, but I am trying to under the basics. I want to know the difference between the filters which require a physical model in order to perform the filtering and the ones which doesn't. I read someone one such type of filter is Luneberg. But I could not find anywhere about them. If anyone could explain difference between these two for understanding purpose, it'd be great.

PS- this is not a homework. I am just trying to understand.

$\endgroup$
  • $\begingroup$ Could you edit your question to be a little more specific? It's unclear what you're asking. $\endgroup$ – Izzo Nov 8 '16 at 15:09
  • $\begingroup$ @Teague, this is more of a Generic question. I want to know about the filters which needs a physical model in order to perform and the filters which doesn't. $\endgroup$ – Aashu10 Nov 9 '16 at 9:41
  • $\begingroup$ I'm not sure what you mean by physical model. Are you asking why some filters can only be implemented digitally, while other filters (analog) can be implemented with passive components (resistor, inductor, capacitor, mass, spring, damper, etc.)? $\endgroup$ – Izzo Nov 9 '16 at 14:32
  • $\begingroup$ No, I am talking about observers. Which need the information about the whole system in order to perform the operation, compared to filters like Butterworth for example? $\endgroup$ – Aashu10 Nov 9 '16 at 14:40
1
$\begingroup$

After reading your comments, the real question you're asking is:

"What's the difference between an estimator and filter?"

An estimator is a system that takes in inputs and attempts to estimate a signal (or state) given knowledge of the system. A simple example is system that takes in positional data and calculates the corresponding acceleration. In this case, your input data is position, the system is the second derivative with respect to time, and the output is acceleration. You've just estimated the acceleration by feeding it through a model.

A filter is just a system that takes an input and attempts to modify the frequency content of a signal. A Butterworth filter is a filter, not an estimator.

So the real question you have to ask your is: "What am I trying to do?"

A problem where you're trying to estimate something, you use an estimator (i.e. a system that produces an output using some physical model). A problem where you trying to change the frequency content of a signal, you use a filter. The filter doesn't care about the underlying process or signal, it's just setup to modify the frequency content of the signal.

Mathematically, a filter and estimator are the same thing (put signal in, get signal out), but they typically have different physical meanings and are used to achieve different things.

Edit: It also seems that your question may have been inspired by the 'Kalman Filter'. Even though this contains the word 'filter', its still considered more of an estimator. However, depending on how you set up the problem (i.e. how much uncertainty you have in your model and how much uncertainty you have in your feedback), the system will experience different frequency response characteristics.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.