Date: Wed, 28 Apr 1999 16:22:43 -0500 (EST) <br>
From: Michael Zoltowski <mikedz@ecn.purdue.edu> <br>
Subject: Re: Question on Homework 6 <br>
 <br>
> I noticed that the second filter (part B) calls for a square root raised <br>
> cosine filter of length 32. If I change N in PRNonuniform.m file from 17 to <br>
> 16 I no longer get perfect reconstruction. The same is true for the third <br>
> filter (Part C) which calls for length 48 and PRNonuniform.m file has N=25 <br>
> corresponding to a length of 50. The shorter filter lengths worked fine for <br>
> the uniform filter bank. It seems the nonuniform filter bank needs a <br>
> sharper raised cosine filter than the uniform filter bank. Since the goal <br>
> is perfect reconstruction should I leave the longer length filters? <br>
> <br>
  <br>
I am not sure what the problem is here, but as long as the variable <br>
I call "N" in PRNonuniform.m is an odd number, I achieve Perfect <br>
Reconstruction (PR).  In contrast, if I make "N" an even number, there  <br>
are some slight discrepancies between the original and reconstructed <br>
signal around pi/2.  Note that what I call "N" is not the length of <br>
h[n], but rather roughly half the length of h[n].  Whether I assign <br>
"N" to be even or odd, the length of resulting h[n] formed in <br>
PRNonuniform.m is even and symmetric. <br>
 <br>
Thus, I would recommend that for part (B), you assign "N" in PRNonuniform.m <br>
to be either 15 or 17.  For part (C), I recommend that you assign "N" <br>
in PRNonuniform.m to be either 23 or 25.  If I figure out, what the <br>
problem is, I'll let you know. <br>
 <br>
 - Michael D. Zoltowski <br>
 <br>