MathGroup Archive 1995

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Softball Fourier Question

  • To: mathgroup at christensen.cybernetics.net
  • Subject: [mg608] Re: [mg591] Softball Fourier Question
  • From: "Ed Boss" <boss at wln.com>
  • Date: Mon, 27 Mar 95 14:19:05 -1000

Brian writes:
>How do I factor in the sampling rate in order that the power spectrum I 

>get out is in Hertz?

The convention for scaling FFT output is to multiply by 1/(2 delta t) where
delta t is your sampling interval.  However, there are some pitfalls in
making spectral estimates based on a single FFT of your data.  You might
want to look into the Welch method which averages the results of subsets of
the data.  Matlab has a routine for implementing this method in its spectrum
command.  Also, see page 553 of Digital Signal Processing by Oppenheimer and
Schafer.  This is a pretty rigorous treatment of signal processing.  For an
introduction, you might want to read The Analysis of Time Series by C.
Chatfield.  Chapter 7 discusses spectral analysis.

Does anyone know if Mathematica has an implementation of the Welch method,
perhaps in one of the signal processing packages?

Good Luck,
Ed

----------------------------------------------------------------------

Ed Boss                                               .^ *
NOAA/Pacific Marine Environmental Laboratory         /  /
                                                    /   |
                                                   /     \
                                           ___.---~       `*------ ...


  • Prev by Date: Compact way to store BC's
  • Next by Date: Re: Change of variable in ODE's (fwd)
  • Previous by thread: Compact way to store BC's
  • Next by thread: RE: Softball Fourier Question