Forums

how to design window based half band fir filters

Started by bharat pathak January 16, 2008
Hello,

So far i have been designing half band filters using remez
function. Is it possible to design half band fir filter using 
window (like hamming, hanning, kaiser, blackmanharris etc)
based design method?

Regards
Bharat Pathak

Arithos Designs
www.arithos.com

On 16 Jan., 14:13, "bharat pathak" <bha...@arithos.com> wrote:
> Hello, > > So far i have been designing half band filters using remez > function. Is it possible to design half band fir filter using > window (like hamming, hanning, kaiser, blackmanharris etc) > based design method?
Sure. Just window the ideal fs/4 sinc-lowpass impulse response: h[k] = w(k/K) sinc(k/2), where k = -K, ..., 1, 0, 1, ... , K (for an odd number 2 K + 1 of coefficients). w(t) is the window function defined on [-1, 1]. Regards, Andor
"bharat pathak" <bharat@arithos.com> writes:

> Hello, > > So far i have been designing half band filters using remez > function. Is it possible to design half band fir filter using > window (like hamming, hanning, kaiser, blackmanharris etc) > based design method?
This probably isn't telling you anything, but you can window a sinc function representing the ideal (1/4)*Fs bandwidth. The frequency-domain result is the convolution of the rectangular pulse with the window function transform. So it is "possible," but I think that this is going to yield a much less ideal filter (e.g., in terms of stop-band attenuation) than remez. -- % Randy Yates % "And all that I can do %% Fuquay-Varina, NC % is say I'm sorry, %%% 919-577-9882 % that's the way it goes..." %%%% <yates@ieee.org> % Getting To The Point', *Balance of Power*, ELO http://www.digitalsignallabs.com
>So it is "possible," but I think that this is going to yield a much less >ideal filter (e.g., in terms of stop-band attenuation) than remez.
Thanks Randy, In some cases if i need only 50 db attenuation or so. i can use hamming window. the advantage is the slowly going down tail in the stop band region. Sometimes this is very advantageous in situations like image rejection filters, specially in communication applications. I might not require 50db attenuation near stop band frequency but the same might be required as I approach Fs/2. In such cases it is helpful to design filters with trailing response in stop band. Regards Bharat Pathak Arithos Designs www.arithos.com
"bharat pathak" <bharat@arithos.com> wrote in message 
news:qKydnTVOw_UxkxPanZ2dnUVZ_r-vnZ2d@giganews.com...
> >>So it is "possible," but I think that this is going to yield a much less >>ideal filter (e.g., in terms of stop-band attenuation) than remez. > > Thanks Randy, > > In some cases if i need only 50 db attenuation or so. i can use hamming > window. the advantage is the slowly going down tail in the stop band > region. Sometimes this is very advantageous in situations like image > rejection filters, specially in communication applications. I might > not require 50db attenuation near stop band frequency but the same > might be required as I approach Fs/2. In such cases it is helpful to > design filters with trailing response in stop band. > > Regards > Bharat Pathak > > Arithos Designs > www.arithos.com >
Well, you can use the Remez algorithm still. It's just that the common implementations like P-M and Matlab's remez use constant weights in each band. To get what you want you might simply use Matlab's firpm and make the stopband weight a function of frequency - so not constant. Then the stopband can be tailored. Another trick is to use Dirichlet functions as the kernel instead of cosines in a Remez implementation. But you won't likely find a canned implementation that will let you do it. By itself it would yield the same result. But, if you use superimposed triplets of Dirichlets weighted [1/2 1 1/2] as the basis set then the decay of the kernel is very rapid. The only difference between using the von Hann window in a windowing method and using the fft of the von Hann window as a basis function in a Remez algorithm is that you might have more direct control over the results in the latter. Fred