I am trying to pick out a 75hz drive signal from a large open coil which picks
up a good magnitude of 60hz. I find that the derived amplitude number of the
75hz bin bounces quite a bit up to about 1-2%.
My current scenario is this:
I am sampling at 4080hz (I started with 4096 and found that the amplitude
peak/centering was at 4080hz) taking 2048 samples. With the samples, I am using
Welch's method reconstructing 9 1024 sections overlapping equally. Each
section has a Blackman Window applied. I then run a 512 point complex input FFT
(using all real data) on each section, convert the output data to complex data
and find magnitudes. Each section's magitude is squared, accumulated and
then averaged at the end to get the final number.
I have run tests using an oscillator, inputting 75hz only and can scale it to
match a DMM with good accuracy and little bounce (5th digit dithers).
I am almost positive that my results are due to spectral leakage. Is there any
further pre-processing that could be done to improve my results with mixed
frequencies? Is it possible to use an FIR filter before the FFT to reduce the
unwanted 60hz noise. Is zero padding a possibility, doubling the 1024 original
samples with 1024 zeros and doing a 1024 point FFT?