How to separate noise from a signal using Fast Faurier Trans

[IndiJones]

how to separate noise and signal

I have a 10 khz square wave signal filled with noise. I need to separate the noise from the signal by using FFT. For a pure 10 khz signal the FFT shows spikes of decreasing magnitude at 10, 30,50,70,.....khz. But the noisy signal shows random spikes of different amplitudes along with the 10 khz spike. Now the question is how I should determine the parameters in order to filter out the noise.

 

[MoonShine]

separating noise from signals

On 2001-08-14 05:19, IndiJones wrote:
I have a 10 khz square wave signal filled with noise. I need to separate the noise from the signal by using FFT. For a pure 10 khz signal the FFT shows spikes of decreasing magnitude at 10, 30,50,70,.....khz. But the noisy signal shows random spikes of different amplitudes along with the 10 khz spike. Now the question is how I should determine the parameters in order to filter out the noise.

How about using correlation functions. In radar, cross correlation and self correlation are used to extract the reflected signal amongst all of the noise. I hope this helps;its been a while for me working on this , but if needed I could look some of this up.

Regards,

MoonShine

 

[electronlover]

why use fft use to separate the noise

hehe, depends on were the noise lives. If the noise is only additive white noise, then you could use a limiter on your data before using the fft (for instance the data coming in should be between -1 volt and +1 volt, but because of noise is between -1+-.2 volt and +1 +-.2 volt, but after limiter is cleanly either -1 volt or +1 volt). Sadly, if your noise is in the zero crossings it is going to be a pain, as that is now phase noise (or frequency noise, its both). Where did the data come from? This would give a clue on how to best attack it. Hehe, if all you really wanted was square wave at 10khz, just regenerate it.

 

[sick_man]

stick it through a simple cap and res filter at the input first gate of a buffer chip like a 7407 this is sufficient at higher freq an 74ls07 and higher hc and then 74hct07
place a coupling cap at the v+ pin on chip
ground and link all gates in a line
with 22 ohm resistors

this will completly clean it
the r c filter should be parallel with a 10 nf to gnd at the pin
for that freq
i recomend a 1k and a 10 nf poly in parallel
10 khz with be 1k and 10nf for a series filter

 

[johnyaya]

odd harmonics for a square wave
must the solution be with software?

 

[IndiJones]

Thanks guys! But the problem is that I'm not using software for this. I'm trying to detect the noise by using oscilloscope.

 

[electronlover]

if your looking at it with an oscilloscope why did you mention an FFT? an fft is for discrete time stuff (i.e. sampled). If you want the real solution, its called.... pll! wahoo! you can use one and do clock recovery. Now your having fun. depends on how large the noise is relative to your signal. if noise is small, then just do what simbox said. if its large then you gotta do either a filter, then regenerate with a squaring circuit, or use a pll. have fun.

 

[johnyaya]

Tek o-scope that I use has an fft function built in, but I rarely use it cause it's not near as good as the spectrum analyzer. Can you get access to a spectrum analyzer?