Moving average filter

Hi,

I am new in the dsp domain and I want to design a moving average filter
which separate the input signal in different signals based on different
frequency bands.
My input signal is sampled with 8000. I defined different frequency
bands and a center frequency. I would like to apply the filter to the
input signal and get the lowpass filter and after I obtained the first
signal to substract from the original signal and get the next lowpass
filter for the next band.
If anyone can help me please reply.

Thanks,
AK

"AK13" <arpi_ka@yahoo.com> wrote in message 
news:1137185197.471794.67000@g49g2000cwa.googlegroups.com...
> Hi,
>
> I am new in the dsp domain and I want to design a moving average filter
> which separate the input signal in different signals based on different
> frequency bands.
> My input signal is sampled with 8000. I defined different frequency
> bands and a center frequency. I would like to apply the filter to the
> input signal and get the lowpass filter and after I obtained the first
> signal to substract from the original signal and get the next lowpass
> filter for the next band.
> If anyone can help me please reply.
>
look for " half-band decimation" or similar to start with ; you should soon 
find lots of cross-references to different techniques and comparisons 
between them.

Best of Luck - Mike

AK13 wrote:

> Hi,
>
> I am new in the dsp domain and I want to design a moving average filter
> which separate the input signal in different signals based on different
> frequency bands.
> My input signal is sampled with 8000. I defined different frequency
> bands and a center frequency. I would like to apply the filter to the
> input signal and get the lowpass filter and after I obtained the first
> signal to substract from the original signal and get the next lowpass
> filter for the next band.
> If anyone can help me please reply.

What you describe is sensible - remember to use linear-phase lowpass
filters in order to implement the paralell subtraction scheme. Which
part do you need help with?

Andor wrote:

> ... paralell ...

That looks awful, sorry!

Andor wrote:
> AK13 wrote:
>
> > Hi,
> >
> > I am new in the dsp domain and I want to design a moving average filter
> > which separate the input signal in different signals based on different
> > frequency bands.
> > My input signal is sampled with 8000. I defined different frequency
> > bands and a center frequency. I would like to apply the filter to the
> > input signal and get the lowpass filter and after I obtained the first
> > signal to substract from the original signal and get the next lowpass
> > filter for the next band.
> > If anyone can help me please reply.
>
> What you describe is sensible - remember to use linear-phase lowpass
> filters in order to implement the paralell subtraction scheme. Which
> part do you need help with?

With both part. My problem is I don't know how to choose the right
parameters for the calculations.If I use filtering in the time domain
then how can specify frequency domain? How can I calculate from
frequency range the number of points for moving average which give me
the correct signal. My signal is encoded with ulaw. What is the
paralell substraction scheme?
What about the half-band decimation? 

Thanks,
AK

AK13 wrote:
> Andor wrote:
> > AK13 wrote:
> >
> > > Hi,
> > >
> > > I am new in the dsp domain and I want to design a moving average filter
> > > which separate the input signal in different signals based on different
> > > frequency bands.
> > > My input signal is sampled with 8000. I defined different frequency
> > > bands and a center frequency. I would like to apply the filter to the
> > > input signal and get the lowpass filter and after I obtained the first
> > > signal to substract from the original signal and get the next lowpass
> > > filter for the next band.
> > > If anyone can help me please reply.
> >
> > What you describe is sensible - remember to use linear-phase lowpass
> > filters in order to implement the paralell subtraction scheme. Which
> > part do you need help with?
>
> With both part. My problem is I don't know how to choose the right
> parameters for the calculations.If I use filtering in the time domain
> then how can specify frequency domain? How can I calculate from
> frequency range the number of points for moving average which give me
> the correct signal.

Well, you need to specify how many bands you want and at what center
frequencies the bands are placed (you have done that already). Now you
have to design the linear-phase FIR coefficients (which includes
determining how many coefficients you need to fullfill your filtering
requirements).

To do this you need some kind of filter coefficient computation
program. Scilab / Octave are available for free and should include what
you need. Or do you have access to Matlab? There are many commercial
filter design programs available, have a look in

http://www.dspguru.com/

> My signal is encoded with ulaw.

This is usually decoded by the hardware, don't worry about it.

> What is the paralell substraction scheme?

That's a way to implement a filter bank using several linear-phase
lowpass filters. Generate the highest band by subtracting the first
lowpass output from the original (delayed) signal, then generate the
second highest band by subtracting the the second lowpass filter output
from the first, etc. It's what you described above. If you draw a
schematic, you'll see a whole bunch of lowpass filters in parallel.

> What about the half-band decimation?

That depends on whether the bands are scaled by a factor of two.

Regards,
Andor

If I use the moving average filter and subtraction then can I obtain
the desired frequency bands? How I should choose the number of average.

Thanks,
AK

AK13 wrote:
> If I use the moving average filter and subtraction then can I obtain
> the desired frequency bands? How I should choose the number of average.
>
> Thanks,
> AK

Probably not. The moving average filter does not have very good out of
band rejection. Its frequency response looks like sinx/x which has an
out of band peak that is only 13 dB down. 

John

What is the best method and simple for reshaping the signal? If I make
the moving average and reshape the signal then I can approximate the
singal for the band and at the substraction I will get better result
then can I use this method for separate the bands?
For each band I need n samples in a period to detect a signal in this
band. How can I choose this sample times? (if for 200 hz band I need 4
samples in a period to detect the signal and for 2000 I need 40 samples
to detect the waveform then how do I choose this 4 samples? I take
every 10 samples?

Thanks.
AK

john wrote:
> AK13 wrote:
> > If I use the moving average filter and subtraction then can I obtain
> > the desired frequency bands? How I should choose the number of average.
> >
> > Thanks,
> > AK
>
> Probably not. The moving average filter does not have very good out of
> band rejection. Its frequency response looks like sinx/x which has an
> out of band peak that is only 13 dB down.

You are thinking of the FIR filter with impulse response equal to  [1 1
 ... 1], right? I was thinking that the OP thought of "moving average"
(MA) filters in the statistical processing terms, ie. MA = transversal
filters, AR = recursive filters. That is, moving average filters are
not constrained in their coefficients, just another term for FIR
filters. As such, it makes perfect sense to construct a filter bank
from FIR lowpass filters.

Regards,
Andor