home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!parcplace!parcplace.com!khaw
- From: khaw@parcplace.com (Mike Khaw)
- Newsgroups: rec.audio
- Subject: Re: Where can I find the Nyquest theorem
- Message-ID: <khaw.714892058@parcplace.com>
- Date: 27 Aug 92 05:07:38 GMT
- References: <1992Aug25.200552.12909@fcom.cc.utah.edu> <1992Aug25.234452.4677@ptcburp.ptcbu.oz.au>
- Sender: news@parcplace.com
- Lines: 44
-
- In rec.audio you write:
-
- >dave@news.ccutah.edu (dave martin) writes:
-
- >>Where can I find the exact statement of the Nyquest theorem? I have seen
- >>some articles attributing certain properties of digital audio to the Nyquest
- >>theorem which happen to be false. I am trying to find out if the theorem
- >>is in error (unlikely) or if the author of the piece I read is wrong (more
- >>likely).
-
- >Here it is:
-
- >If a signal contains V discrete levels, and runs through a low-pass filter
- >of bandwidth H, the maximum data rate is
-
- > max = 2H log V bits/sec (the log is base 2)
-
- >In other words, the filtered signal can be reconstructed by making only
- >2H (exact) samples per second.
-
- All the statements I've seen of the Nyquist theorem state that the
- Nyquist rate is a *lower bound* sampling rate required to reconstruct
- a bandlimited sampled signal w/o aliasing, for example:
-
- The sampling theorem says that x(t) can be reconstructed
- perfectly if the sampling rate f-sub-s = 1/T-sub-s is
- GREATER THAN TWICE THE FREQUENCY BOUND B for the signal.
- The minimal sampling rate 2B is called the Nyquist rate...
- Schwartz&Shaw, Signal Processing, McGraw-Hill, 1975
-
- [emphasis mine. I would have preferred to quote a more commonly
- known text such as Oppenheim&Schafer, but their statement has a
- lot more math in it that I didn't want to try to type in on this
- ASCII-only machine.]
-
- Gedanken experiment: Suppose you sample a sinusoidal wave of frequency
- 2kHz (2k cycles per second, or 500 microsec/period) at 4kHz (every 250
- microsec). If you just happen to catch the zero-crossings you would get
- zero for every sample, which is not a very useful representation from
- which to conclude that the reconstruction should be a sinusoid rather
- than nothing.
- --
- Michael Khaw Domain=khaw@parcplace.com, UUCP=...!uunet!parcplace!khaw
- ParcPlace Systems, Sunnyvale, CA PRODUCT INFO: info@parcplace.com
-