Polyphase Filtering Question

Started by westocl September 10, 2009
Im have a few question about polyphase filtering. 

Lets say you have a downsampling application. and in general with
polyphase filtiering yoou wish to exploit the efficiency assocationw it
downsampling before invoking the lowpass filter.

If you didnt do ployphase filtering, lets say you implimented a lowpass
that was to reject frequencies above .2Fs and then did the downsampling to
avoid aliasing.

If you were poly phase filtering, you have a bank of M filters, with
inputs delayed and downsampled accordingly.. now you wish to lowpass
filter. is the lowpass filter in the filter banks a .2Fs filter just at the
lower sampling rate?  


Meanwhile, in trying to answer that question, there lies a bigger one...
in a filter bank, where the input to each filter is a subsequence of a
larger sequence, wat is the relationship between each subsequence to the
original. in general each subsequence input to the filterbacnk will be a
badly aliased version of the origional sequence... how then do we get what
we want out of the filter?
On Thu, 10 Sep 2009 10:12:13 -0500, westocl wrote:

> Im have a few question about polyphase filtering. > > Lets say you have a downsampling application. and in general with > polyphase filtiering yoou wish to exploit the efficiency assocationw it > downsampling before invoking the lowpass filter. > > If you didnt do ployphase filtering, lets say you implimented a lowpass > that was to reject frequencies above .2Fs and then did the downsampling > to avoid aliasing. > > If you were poly phase filtering, you have a bank of M filters, with > inputs delayed and downsampled accordingly.. now you wish to lowpass > filter. is the lowpass filter in the filter banks a .2Fs filter just at > the lower sampling rate? > > > Meanwhile, in trying to answer that question, there lies a bigger one... > in a filter bank, where the input to each filter is a subsequence of a > larger sequence, wat is the relationship between each subsequence to the > original. in general each subsequence input to the filterbacnk will be a > badly aliased version of the origional sequence... how then do we get > what we want out of the filter?
You are confused. You don't downsample before the filter. Instead, you only run the filter for the output samples you want. Furthermore, if the ratio of the two sampling rates is harmonic (i.e. 2:1, 3:1, etc.) then you don't need to use polyphase filtering at all -- you can just run the one 'correct' filter. -- www.wescottdesign.com