What am I missing normalizing a FIR filter?

Fred,

It is small but it does make an importance difference in the filtered
signal.  If I don't subtract the mean the signal gets a DC component (which
generates no sound but it is a pain as the graph of it can shift off the
screen...I am doing this from a sound card on a PC in Windows). I don't know
what kind of effect choosing the limits of the filter to be zeros (or not)
has. But I guess it isn't much, because once I subtract the mean the zeros
at the terminous of the filter are no longer zero!

The filter looks pretty much like a sinc(x) function with a few extra
wiggles.

Brian


"Fred Marshall" <fmarshallx@remove_the_x.acm.org> wrote in message
news:gb2dncYkk_DMXJzd4p2dnA@centurytel.net...
>
> "Fred Marshall" <fmarshallx@remove_the_x.acm.org> wrote in message
> news:_L6dnSpEcJxclpzdRVn-hw@centurytel.net...
> >
>
> Brian,
>
> I want to modify this one thing that I said:
>
> > The continuous, infinite impulse response is then sampled and truncated.
> > Just to be sure the stopband at f=0 has zero response, the very small
mean
> > of the truncated filter is then subtracted from each filter coefficient
> > divided by the length of the filter -> i.e. subtracting mean/N.
>
> It would have been correct to say "subtracting the mean" if the objective
> was to set the response at f=0 identically to zero.
>
> However, I think this is an unusual thing to to and isn't guaranteed to
work
> in the general case.
> What if the filter unit sample response is 1 1 1 ? Then the mean is 1 and
> subtracting the mean yields all zero coefficients.  I rather think that
the
> response at f=0 is the same sort of thing as the response at any other
> stopband frequency.  So, you just accept it along with all the other
> non-zero responses in the stop band and don't subtract the mean.  In the
> case you have, it probably doesn't hurt either - the "correction" should
be
> pretty small.
>
> Fred
>
>