On Oct 24, 5:07 am, tontoko <
[email protected]> wrote:
> If we could eliminate the tank circuit from FM demodulator, we'll make
> the size of it much smaller. Here is a conceptual design of FM
> demodulator using the successive Fourier transform instead of tank
> circuit;http://hecoaustralia.fortunecity.com/c12508/fourier2.htm
Since you are effectively doing linear interpolation in the time
domain, you will need a lot of oversampling to remove the distortion.
You ought to try using a raised cosine kernal for the interpolation.
You can adjust the rolloff of the interpolating kernal (i.e., how much
of the frequency band do you want to have a flat response over?) to
get the distortion under control. Your link's plots show large amounts
of distortion. Since you are adding up the FT samples, you will need
to use the Heaviside shifting theorem in order to properly combine
them.
And by the way, Ratio detectors and Foster Seeley discriminator's are
actually (almost always) implemented in hardware - very seldom are
they done in software. Since the advent of DSP (techniques and
controllers), FM demodulation is done by other methods. I.e. a cross
correlation demodulator or simply by the difference of arctangents or
some approximation thereof. Many modern receivers provide the signal
as an analytic signal, so finding the instantaneous frequency is
rather easy.
You should find out about how these methods work and then compare your
approach to them to get a measure of your approach's efficacy.
Comparing a DSP based technique with an old hardware method is like
comparing a car to a horse, when everyone already has cars. If you
have a new type of car, you will need to compare it to other cars.
A common and important test for an FM demodulator is to see how well
it works in the presense of noise.
IHTH,
Clay