# Input parameters of Levinson-Durbin algorithm

Started by May 7, 2006
```Hello experties,
I'm using Durbin-Levinson algorithm to calculate the inverse impulse
response of a given impulse response h in the time domaine.
The algorithm requires three input parameters:
- Impulse response h
- The delay time of the inverse impulse response of h
- The length of the inverse impulse response

Is there any methode to estimate The delay time and the length of the
inverse impulse response which leads to an acceptable output(inverse
impulse response of h) of the algorithm?
ps:For some values of the delay and the length of inverse impulse
response, the output of the algorithm is false!
Thank you very much for any suggetion or idea!
-------
Stef
```
```stef wrote:
> Hello experties,
> I'm using Durbin-Levinson algorithm to calculate the inverse impulse
> response of a given impulse response h in the time domaine.
> The algorithm requires three input parameters:
> - Impulse response h
> - The delay time of the inverse impulse response of h
> - The length of the inverse impulse response
>
> Is there any methode to estimate The delay time and the length of the
> inverse impulse response which leads to an acceptable output(inverse
> impulse response of h) of the algorithm?

This seems to be a general implemenntation of the Levinson recursion?
For the "usual" AR estimation problem, the delay ought to be 1.

Apart from that, the impulse response h is given and you will have to
find what length is needed to get OK results.

> ps:For some values of the delay and the length of inverse impulse
> response, the output of the algorithm is false!

Sure. Data processing is a highly subjective dicipline. The outcome
of depends entirely on the choises you and I make as operators.
You choose one length fro the impulse response, I choose another.
The results for the inverse filter are different. You choose the AIC
order estimator, I choose the MDL order estimator. The results
are different.

It is a very common misconception, possibly due to all the
fancy maths floating around in data processing labs, that there
is one "right" way of doing things, and that it is possible to
process the data so as to obtain the "truth" about whatever
experiment or measurement. No such "right way" or "truth"
exists. Only alternatives and choises.

Rune

```