"Porky" writes:
"Arny Krueger" wrote in message
...
"Porky" wrote in message
The experiment I suggested will give the results I gave, and if it is
right at under the circumstances I suggested, it should be right
under all circumstances with the same conditions, right? In other
words, if it applies with a LF of .1 Hz or 1 Hz, it will still apply
at LF's 20Hz or 50Hz, is that not correct?
right. However, its a lot harder to properly measure doppler when the LF
tone has a very low frequency. To measure it with a FFT you must use a FFT
size that covers at least one cycle, and hopefully several cycles of the
process. If the LF tone is 0.1 Hz, this means an absolute minimum of 10
seconds of data, and ideally 30 or more. At 44,100 Hz sampling, this would
be a FFT composed of a minimum of 441,000 samples, and preferably several
million samples.
Consider the original example - the LF tone was 50 Hz. It had an 882
sample
period. Note how much overkill there was when analyzed using a 65k sample
FFT, or as I used a one million point FFT.
One of the problems with FFT analysis that we've all overlooked is that we
aren't really dealing with analog waveforms in our simulations, and we can
get erroneous results when using high FFT numbers because we start playing
in the digital "cracks", so to speak,
The waveform being analog or digital makes no difference as long as sufficient
bandwidth and dynamic range has been supplied by the A/D conversion. Rather, the
problem you are ignorantly referring to is that an FFT implicitly assumes the
input is periodic. If it isn't, you can get yourself befuddled. There is also
the problem with using the FFT to estimate the spectrum of a random signal -
it can be shown that there will be variance in the frequency estimates no
matter how many points are used in the FFT (see, for example, "Signal Processing:
Discrete Spectral Analysis, Detection, and Estimation," Mischa Schwartz and
Leonard Shaw).
--
Randy Yates
Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
, 919-472-1124