Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!samsung!uunet!valid!caber!lou
From: lou@caber.valid.com (Louis K. Scheffer)
Newsgroups: comp.dsp
Subject: Re: 48k to 44.1k sample rate conversion
Message-ID: <461@valid.valid.com>
Date: 18 May 91 22:16:03 GMT
References: <5826@media-lab.media.mit.edu.MEDIA.MIT.EDU> <BWEEKS.91May13130405@fermat.UVic.CA>
Sender: news@valid.com
Lines: 31

bweeks@sirius.UVic.CA (Brent  Weeks) writes:

>In short, efficient implementation of Sample Rate Conversion involves
>filtering short windows (of length Q) of the input data using a
>polyphase filter (FIR with Q taps) operating at the output sample
>rate.  The polyphase filter has periodically time-varying
>coefficients.

>The sample rate converter should be quite do-able on a DSP Chip.  Use
>an up-conversion factor of 160 and a down-conversion factor of 147.
>(44.1 * 160/147 = 48)  The polyphase filter cycles through 160 states,
>then repeats.  The filter coefficients would be stored in 160*Q words.
>For each output sample, besides managing windowing of the input data,
>just Q multiply and accumulates need to be performed.  Typical values
>of Q range from 5 to 20.

   For audio work, you will need more coefficients.  You probably want
   delta(passband) = 0.01 (about .1 db ripple)
   delta(stopband) = 0.00001 (100 db of rejection)
   delta F = 2 KHz (flat to 20, down 100 db by 22 KHz)
   F = 7 MHz
   Looking at the chart, D(delta(s), Delta(p)) is about 4, so
   you need about F/deltaF * D = 14000 coefficents.  Hence Q is about
   14000/160 = 88 for good audio work.

   This is more than the examples in the paper because audio work
      a) tries to get real close to the sampling limit, leading to steep
	 transition bands.
      b) tries to get very flat response in the passband
      c) tries to get very good signal to noise. (100 db)
   All of these increase Q.