In article J2F@news.cso.uiuc.edu, ja51359@uxa.cso.uiuc.edu (axelrod) writes:
>
> If I want to lower a sample's sampling rate, what kind of filter should I use?
> Is there a conventional digital filter design for this purpose specifically?
> It's an audio signal. How important is keeping phase linear in audio filters?
>
> Thanks
> Jeff
>
What you need is called a decimation filter. This can be a lowpass FIR designed at your initial sampling rate. If there is no noise above your band of interest, you don't even need a filter ! Just keep one out of every N sample, and you have the simplest decimation filter. Now, if there happens to be some noise (as is usually the case) above the band of interest, you would probably like to avoid having some of it aliased back in your signal. Just filter out as much as you can without distorting the au
dio signal before keeping one out of every N samples out of the filter.
If computing speed is of interest to you, be aware that interpolation-by-2 and decimation-by-2 filters can often be designed as what is called a half-band filter. These filters have the nicest property: every second coefficient is 0. See the advantage ? At the expense of a sligthly different data structure, you get the result in (almost) half the time. If you want to design one, here is how: I use Park-McClellan (equiripple) FIR type of filter design. Everything you specify should be symmetric about
Fs/4, including passband and stopband ripple, as well as the transition band (ex: for Fs/2 = 0.5, PBC = 0.2, SBC = 0.3, PBR = 0.01, SB = 0.01 will give you a half-band filter).
From my own experience, phase is important for the quality of audio, but phase linearity is not. What really hurts an audio signal is phase discontinuity. However this is not an issue here since decimation filters can be FIR, thus linear.
---
Yves Jolly Email: yves@mars.dgrc.doc.ca
Communications Research Centre, Dept. of Communications
