Strange result with Fourier transform after frequency removal

I have a line profile that Im trying to obtain a "width measurement" on. The problem is that its polluted with wiggles.

I was thinking taking the FT, removing the highest component and then inverse FT would be a useful method.

However, my code is giving strange results. My x and y data is also attached

%look at FT to try and remove periodic ringing
F = fftshift(fft2(ydata));    %Take the FT
sF = log(abs(F));             %frequency spectrum
mxsF=max(sF(:))               %Determine max contributor  
ind=find(sF==mxsF)            %Find its frequency (i.e.. x-axis)
%View frequency current spectrum
subplot(1,4,4)
plot(sF,'b.')
grid on
hold on
%suppress peak freq 
sF(ind)=0;
new=abs(fftshift(ifft(sF)));    %inverse transform
plot(sF,'r-')
hold off

figure
plot(x,new)

Answer 1

[Star Strider]

The easiest way is to lowpass-filter your signal:

[d,s,r] = xlsread('Jason temp2.xlsx');

t = d(:,1);
s = d(:,2);
L = length(s);

Ts = mean(diff(t));                                     % Sampling Interval
Fs = 1/Ts;                                              % Sampling Frequency
Fn = Fs/2;                                              % Nyquist Frequency

FTs = fft(s)*2/L;
Fv = linspace(0, 1, fix(L/2)+1)*Fn;                     % Frequency Vector (Hz)
Iv = 1:length(Fv);                                      % Index Vector

figure(1)
plot(Fv, abs(FTs(Iv)))
grid

Wp = 0.05/Fn;                                           % Design Lowpass Filter
Ws = 0.15/Fn;
Rp =  1;
Rs = 30;
[n,Wn] =  buttord(Wp,Ws,Rp,Rs);
[b,a] = butter(n,Wn);
[sos,g] = tf2sos(b,a);

figure(2)
freqz(sos)                                              % Filter Bode Plot

s_f = filtfilt(sos, g, s);                              % Filter Signal

figure(3)
plot(t, s_f)
grid

[Jason]

Thankyou

[Star Strider]

My pleasure!

Your Question was a fun way to start my day!

[Jason]

ahh. I don't have the buttord function as no signal processing toolbox. Is there a work around? thanks

[Star Strider]

My pleasure.

The best I can do in the absence of the Signal Processing Toolbox is Butterworth / Bessel / Chebyshev Filters from the University of York (U.K.).

The filter I used here has these coefficients:

b = [1.220454380697567e-03     4.881817522790266e-03     7.322726284185399e-03     4.881817522790266e-03     1.220454380697567e-03];

a = [1.000000000000000e+00    -2.897489668063421e+00     3.260788665023785e+00    -1.671326187500433e+00     3.275544606312290e-01];

Use those (and any you create from the York page) with the filter function. It’s not phase-neutral as is filtfilt, but it will likely work well enough.