# How to decide the filter length in ADNC using LMS Algo ?

Started by August 26, 2003
```Hi,

How to decide the FIR - filter length (no. of taps) required to design
the ADAPTIVE NOISE CANCELLER ?

I am using the LMS Algorithm..

```
```I guess you use the LMS algorithm to adapt to some system model. To
determine the length of the filter, you need to know the characteristics
of this system model. Typical impulse response length might be usable.

If you choose the length of the filter too low - you end up in an under
modeling situation, meaning the the filter will never be able to
represent the system. If you choose the length of the filter too high -
you will always be able to represent the system, however the converge
properties will be poor (due to over modeling). The trick is to find the
one in between.

Hope this helps...

/S&#2013266168;ren

Sandeep Chikkerur wrote:
> Hi,
>
> How to decide the FIR - filter length (no. of taps) required to design
> the ADAPTIVE NOISE CANCELLER ?
>
> I am using the LMS Algorithm..
>
> Thanks in advance...

```
```
Sandeep Chikkerur wrote:
>
> How to decide the FIR - filter length (no. of taps) required to design
> the ADAPTIVE NOISE CANCELLER ?
>
> I am using the LMS Algorithm..

You need to know the model of the system. The parameters of interest are
the flat delay and the impulse response length.
If you don't know the model of the system, you may try to do the
preliminary LMS algorithm with big number of taps just to estimate the
length of the impulse response, and then adjust the actual FIR filter to
this length. The preliminary LMS may run at the decimated rate for

DSP and Mixed Signal Design Consultant

http://www.abvolt.com
```
```Vladimir Vassilevsky <vlv@abvolt.com> wrote in message
news:<3F4B5E33.48616DA1@abvolt.com>...
> Sandeep Chikkerur wrote:
> >
> > How to decide the FIR - filter length (no. of taps) required to design
> > the ADAPTIVE NOISE CANCELLER ?
> >
> > I am using the LMS Algorithm..
>
> You need to know the model of the system. The parameters of interest are
> the flat delay and the impulse response length.
> If you don't know the model of the system, you may try to do the
> preliminary LMS algorithm with big number of taps just to estimate the
> length of the impulse response, and then adjust the actual FIR filter to
> this length. The preliminary LMS may run at the decimated rate for
> reduced computation load.
>
> Vladimir Vassilevsky, Ph.D.
>
> DSP and Mixed Signal Design Consultant
>
> http://www.abvolt.com

Do u mean, the no. of samples in the impulse reponse corresponds to
the FIR filter length for the unknown system's model ?
```
```
Sandeep Chikkerur wrote:
>
> Vladimir Vassilevsky <vlv@abvolt.com> wrote in message
news:<3F4B5E33.48616DA1@abvolt.com>...
> > Sandeep Chikkerur wrote:
> > >
> > > How to decide the FIR - filter length (no. of taps) required to design
> > > the ADAPTIVE NOISE CANCELLER ?
> > >
> > > I am using the LMS Algorithm..
> >
> > You need to know the model of the system. The parameters of interest are
> > the flat delay and the impulse response length.
> > If you don't know the model of the system, you may try to do the
> > preliminary LMS algorithm with big number of taps just to estimate the
> > length of the impulse response, and then adjust the actual FIR filter to
> > this length. The preliminary LMS may run at the decimated rate for
> > reduced computation load.
> >
> > Vladimir Vassilevsky, Ph.D.
> >
> > DSP and Mixed Signal Design Consultant
> >
> > http://www.abvolt.com
>
>
> Do u mean, the no. of samples in the impulse reponse corresponds to
> the FIR filter length for the unknown system's model ?

What you are looking for is the decay time constant of the impulse
response. The rule of thumb FIR length should be ~ 4...5 times of the
decay time constant.

DSP and Mixed Signal Design Consultant

http://www.abvolt.com
```
```>What you are looking for is the decay time constant of the impulse
>response. The rule of thumb FIR length should be ~ 4...5 times of the
>decay time constant.
>
>
>DSP and Mixed Signal Design Consultant
>
>http://www.abvolt.com
>

Is there any algorithm that support the rule?

This message was sent using the Comp.DSP web interface on
www.DSPRelated.com
```
```sephinroth wrote:
>>What you are looking for is the decay time constant of the impulse
>>response. The rule of thumb FIR length should be ~ 4...5 times of the
>>decay time constant.
>>
>>
>>DSP and Mixed Signal Design Consultant
>>
>>http://www.abvolt.com
>>
>
>
> Is there any algorithm that support the rule?
>
>
>
> This message was sent using the Comp.DSP web interface on
> www.DSPRelated.com

Support for any rule of thumb -- an interesting phrase, no? -- is
experience. Adding "nothing more" would belittle the value of
experience; very unwise.

It's easy to justify the rule as reasonable. An impulse decaying under
the influence of a single time constant reaches 5% after three time
constants, falls below 2% by four, reaches 2/3% at five and 1/4% at six.
The rule of thumb part comes from estimating how small is small enough
to be safely neglected.

Jerry
--
Engineering is the art of making what you want from things you can get.
&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;
```