I have problems to reconstruct a signal in time after an ideal filter

[bego]

hello,
my problem is: I have an ideal signal of 22050 samples. I make the complex_FFT and after that, I pass the spectrum by an ideal filter I have designed. This filter consist of taking the samples of the third octave band corresponding to the signal frecuency, and doing zero padding in both sides.

So, now I have an spectrum identical to the original, with the only diference that the zeros are ideal.

When I try to obtain the original signal using the inverse FFT, I find the problem that can be seen in the vi example I attach. It's somethig difficult to explain for me, so I have copy the graphs results in a vi so maybe somebody who found this problem before could tell me what the proble
m could be, and what the solution.

The signal given by the complex inverse FFT is more surprising indeed.

Thank you

[MDC]

To get a better idea of the exact processing, you might want to attach the processing VI itself. This would also allow others to experiment with different solutions.

I can bring in one detail to keep in mind when performing a single block frequency-domain filtering operation (a more complex method such as "overlap and add" should be used for continuous frequency-domain filtering). When you take a real FFT of your real signal, the complex result is symmetric about the Nyquist frequency for the real part, anti-symmetric for the imaginary part (Nyquist index is N/2). You must maintain this symmetric for all frequency domain data (signal FFT AND filter FFT) and operations so that the resulting inverse FFT is purely real. Other important indices to maintain are FFTre[0
] and FFTre[N/2] which are the "DC" and Nyquist frequency components. The imaginary parts FFTim[0] and FFTim[N/2] must be zero for a real signal.

I can help you more with more detailed processing information or the processing VIs themselves.