Create a notch filter for complex input data

I have a complex sampled signal where I want to filter out the signal at 244140 Hz with an Fs = 390625.

What I tried was:

% See the pic for what the signal looks like
d = fdesign.notch('N,F0,BW,Ap',200,0.625,0.045,1);
Hd = design(d);
y = filter(Hd,myComplexData);

But

1. The notch didn't show up where I thought it would
2. It is not complex

Now because it is not generating a filter with complex coefficients I understand why the notch is at where it is at.

My questions is: How do I create a complex coefficient filter?

[Star Strider]

What does a freqz plot of your filter look like?

[B P]

Here is the pic:

[B P]

This is what the output looks like after filtering

You can see that there are two spots where the filter is attenuating. I only want the filter to filter out around the 320 bin.

Answer 1

[Uladzimir]

 Maybe it will help you

F = 244140;
Fs = 390625;
NFFT = length(YourSignal);%or 2^n
YourSpectra = fft(YourSignal,NFFT); 
df = Fs/NFFT;% or, if your signal is complex, this must be a known parameter
Ns = round(F/df)+1; 
ComplexHarmonic = 2*YourSpectra(Ns);

[B P]

How does 2 x the frequency bin I am trying to filter help produce a complex valued filter?

[Uladzimir]

I just showed an idea with frequency transform. If you have already complex signal, 2x isn't needed, just in a case of fourier transform. You have a signal in a frequency domain. And need a value of one bin on the definite frequency. So, you just need to know the number of this bin.

Ns = round(F/df); 
ComplexValue = ComplexSignal(Ns);

Answer 2

[Billy Kihei]

Multiplication in frequency domain is convolution in time domain. (this works for real or complex)

1. Take x[t] -> fft -> Y[f]
2. zero out the values in Y[f] you want to get rid of -> Ymanuallyfiltered[f]
3. Ymanuallyfiltered[f]->ifft->x_filtered[t]
4. x_filtered[t]->fft->Yfiltered[f]

Note: This may make your code take much longer to run, but it's a quick and dirty way without designing a filter. It is important that you convert it back to time domain after your manually editing. Then re-fft it to get the natural frequency of the response after filtering. Hope this helps.