I have an analog waveform defined in the time domain that I need to
detect and measure. I would like to design a matched filter to
optimize the signal to noise ratio of the detected waveform.

Although the waveform is a pulse, it is rounded on the top and can not
really be considered rectangular. The documentation I have been
finding seems to just talk about rectangular wave shapes for digital
bit detection. I am doing something different; I want to actually
measure the pulse area or at least have something proportional to it.

Can someone steer me in the direction of some software or technique
for designing these types of filters?

Thanks.  Al

On 28 May 2004 12:02:57 -0700, a...@etest.com (Al) wrote:

>I have an analog waveform defined in the time domain that I need to
>detect and measure. I would like to design a matched filter to
>optimize the signal to noise ratio of the detected waveform.
>
>Although the waveform is a pulse, it is rounded on the top and can not
>really be considered rectangular. The documentation I have been
>finding seems to just talk about rectangular wave shapes for digital
>bit detection. I am doing something different; I want to actually
>measure the pulse area or at least have something proportional to it.
>
>Can someone steer me in the direction of some software or technique
>for designing these types of filters?
>
>Thanks.  Al

I think most good comm books usually have a reasonable treatment of
matched filtering.   You might peruse a few and look in the ToC and
Index for matched filter or Nyquist filter.

Also, you can think of the receive filter as a correlator, and it
isn't hard to design a correlator for arbitrary waveforms (although,
obviously, some waveforms have better correlation properties than
others).

Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org

a...@etest.com (Al) wrote in message news:<e...@posting.google.com>...
> I have an analog waveform defined in the time domain that I need to
> detect and measure. I would like to design a matched filter to
> optimize the signal to noise ratio of the detected waveform.
> 
> Although the waveform is a pulse, it is rounded on the top and can not
> really be considered rectangular. The documentation I have been
> finding seems to just talk about rectangular wave shapes for digital
> bit detection. I am doing something different; I want to actually
> measure the pulse area or at least have something proportional to it.
> 
> Can someone steer me in the direction of some software or technique
> for designing these types of filters?
> 
> Thanks.  Al

It depends on circumstances, but a matched filter or a Wiener filter 
could perhaps do the job. There are pro's and con's with each class.

The problem with Wiener filters is that they require estimates for 
background noise characteristsics, and might not help at all if you 
get the initial parameters wrong. 

The matched filter, which is a FIR filter with impulse resonse like 
the time-reversed waveform you are looking for, does not require 
knowledge of anything else than what you are looking for. However, I 
would imagine the main problem with the matched filter would be 
hich cost, in that it might require a long FIR filter.

FWIW, 

Rune

On 29 May 2004 04:51:26 -0700, a...@tele.ntnu.no (Rune Allnor)
wrote:

>a...@etest.com (Al) wrote in message news:<e...@posting.google.com>...
>> I have an analog waveform defined in the time domain that I need to
>> detect and measure. I would like to design a matched filter to
>> optimize the signal to noise ratio of the detected waveform.
>> 
>> Although the waveform is a pulse, it is rounded on the top and can not
>> really be considered rectangular. The documentation I have been
>> finding seems to just talk about rectangular wave shapes for digital
>> bit detection. I am doing something different; I want to actually
>> measure the pulse area or at least have something proportional to it.
>> 
>> Can someone steer me in the direction of some software or technique
>> for designing these types of filters?
>> 
>> Thanks.  Al
>
>It depends on circumstances, but a matched filter or a Wiener filter 
>could perhaps do the job. There are pro's and con's with each class.
>
>The problem with Wiener filters is that they require estimates for 
>background noise characteristsics, and might not help at all if you 
>get the initial parameters wrong. 
>
>The matched filter, which is a FIR filter with impulse resonse like 
>the time-reversed waveform you are looking for, does not require 
>knowledge of anything else than what you are looking for. However, I 
>would imagine the main problem with the matched filter would be 
>hich cost, in that it might require a long FIR filter.
>
>FWIW, 
>
>Rune

Hi Rune,
  maybe Al could decimate his input time sequence 
by two, then the filter's length would be 
reduced by a factor of two.  If his input SNR is 
high enough, that might work OK.
I wonder if the signal Al's trying to detect 
is symmetrical in time (?).  If so, he could 
reduce the number of multipliers in his tapped-delay 
line FIR filter.

[-Rick-]