Hi, I need to estimate the amplitude of each harmonic of an harmonic sound. I need a method that is more robust than simple local maxima detection because I want to avoid the problem of noise (unwanted peaks) in the FFT. Can anyone suggest something useful? Thank you.
Sampling with high bandwidth high speed ADC.filter with simulated digital filter . Now a days lot of manufacturers are supplying assembled FPGA boards with sample software will solve the problem .Patches are available from supplier to to integrate with Matlab for development..Maxim-ic and Analog devices are the leaders in such FPGA chips with support.
In fact, you can use a modal identification algorithm on the FFT. You simply use it as if the sound was the dynamic response to a flat, unitary, input. The algorithm will give you back, not only the amplitude of the harmonics (peaks at resonances) but the phase for each one.
Dear Francesco,
One solution added to above solutions introduced by the colleagues is to use frequency synthesizer FS with a step fine enough to resolve the harmonics with the required accuracy. This synthesizer drives a correlator composed of a multiplier and and narrow low pass filter. The low pass filter can be substituted by an integrate and dump circuit with an integration time a multiple of the period of the fundamental. In this way the effect of the thermal noise will be very reduced. The signal is input to the correlator and the FS is scanned and the peaks are detected. Whenever we have a peak the frequency of the input signal will be equal to that of the synthesizer.
Such instrument works a spectrum analyzer.
wish you success
The problem is that you posted wrong question. The fact is, that whenever you want to improve some signal, you must use some apriori knowledge, what is desired and what is not. In your question you mention that you would like to eliminate unvanted peaks, however you do not state what make these peak unwanted. How can you distinguish the signal which causes unwanted peaks from the signal which causes wanted peaks?
Saying that the signal is harmonic (most probably you meant periodic, as harmonic signal has no higher harmonics) , is not enough. Simply by performing FFT of a signal, you implicitly assume that the signal outside FFT range is harmonic, that the same signal repeats over and over again and that the lowest frequency in FFT is the fundamental frequency of that signal. Thus all other frequencies are legal harmonics.
In short, to be able to eliminate unwanted peaks from the signal, you must now (or tell) something more about the signal in question. Is it really periodic? FFT over several "periods" of the signal can reveal this. If it is really periodic and you performm FFT over 4 periods of the signal only each 4th component of FFT should be nonzero. If you perform FFT over whole signal (it is impossible for real periodic signal) and the signal out of the range of FFT is zero, the frequencies in beetwen FFT frequency can be approximated by sinc function or obtained by zero padding and longer FFT.
If the signal contains frequencies outside of FFT raster, then you should use windowed FFT.
So, to be able to answer your question, you should tell more precisely what are your assumoptions and what is your aim.
Saso
- Badges
- Science topic
- Similar topics
- Engineering
- Telecommunications Engineering