Thanks for the help.
> Fred wrote:
> 1. You neglected to say what the important
> parameters are which you want
> to abstract from the collected data!
Sorry if I was not specific enough. I want to extract
the dominant frequency every 1/100 second (using FFT
in software). From the dominant frequencies series,
the software reconstructs the series of signals and
the pauses between them. The software has been tested
with test data, without hydrophones. The software also
emits similar signals underwater if the user gives it
the command to do so. A version of the software can be
downloaded for free from <a href="https://leafy.dev.java.net/";
rel="nofollow">https://leafy.dev.java.net/</a>
The current version of the software is limited to a
single dominant frequency value per frequency sample,
i.e., it assumes that the signal is a sine wave or
close to it, e.g., a whistle.
> 2. Some cetaceans and other sources go to at least
> 150 KKz.
Good point. If bioacousticians think that my system
should go to 150 kHz and the hardware to do so is not
more expensive then lets do it. As for the current
software, the upper frequency limit is flexible and
mainly dependent on the way the acquisition harware
communicates with the software. If the hardware can
give the software an array of voltage numbers every
1/100 sec then the acquisition software can handle
higher frequencies. The emission software may have
more difficulties in emitting whistles at higher
frequencies as it has to generate the voltage samples
at a faster rate.
> 3. You neglected to indicate the phase response you
> will need nor the
> error between the two channels in phase and
> amplitude.
This is because I do not know what I should require
for these parameters. I expect that *normal*
requirements will be sufficient. But this is a guess
on my part due to my lack of knowledge in this area.
You should tell me what I need.
> 4. The original purpose of fourier methods was to
> look at what happens
> inside a KNOWN boundary NOT I repeat NOT to
> determine the values of a
> boundary. There are many errors to be expected in
> the results of
> frequency determinations using any of the current
> FFT methods. For
> single input frequencies to a FFT frequency errors
> of 10% have occured.
> Phase errors also occur!
Your input here is extremely valuable. Can you please
give me some reference on this subject?
> 5. The current best A/D converter (which I would
> use) is only 14 bits.
> It is the ANALOG DEVICES AD10465 A DUAL CHANNEL 14
> BIY 65 MEGA-SAMPLES
> PER SECOND converter. I would use (if possible) IEEE
> 1394 VERSION 2.
> version 1 is not as fast as USB 2. There is an
> evaluation board
> available for the 10465 which I would recommend be
> incorporated into
> your front end.
I will look this up. Thanks.
>
> 6. For those interested in how to evaluate the
> output of an FFT (being
> used for frequency determination) routine let me
> know and I will post
> examples of pathologic cases.
That would be very useful. Please do.
Thanks again,
Serge
|