How to straighten a sinusoidal signal?

I have a signal that bouncy (the waves are all over the place), and I've been trying to filter out the bouncy signals without changing the amplitude of the whole signal. Basically, I'd like a sinusoidal signal that is somewhat straight (doesn't have to be perfect). The data from the signal was imported from an excel file.

I have attached an image of what one of my signals look like. The problem with that particular signal is that little dip at the end of the signal that is very uneven with the rest of the waves.

Thanks for your help!

[Adam Danz]

An image would be helpful (none attached). What does a straight sinusoid mean?

Are you trying to fit a sinusoid to noisy data?

[Chinwe Orie]

Thanks, no problem. I have attached an image of what one of my signals look like. The problem with that particular signal is that little dip at the end of the signal that is very uneven with the rest of the waves. So I'd like it to be somewhat close to being even with the rest of the signal.

[Greg Dionne]

Hi Chinwe,

Is your goal to measure the amplitude of the sinusoid?

[Chinwe Orie]

Yes. I planned to take the envelope peaks of the signal and then subtract the upper envelopes from the lower ones to get the amplitude, but at the end the signal is uneven so taking the envelopes wouldn't be accurate there.

[Greg Dionne]

There are a few ways to do this. Image Analyst's approach is good (remove the discontinuity, then filter). You can then view the spectrum (or spectrogram) to obtain the amplitude of the filtered signal.

Answer 1

[Image Analyst]

It looks like the signal is going along just fine until suddenly there is a big jump, and then it's fine again. So rather than filter the whole signal, I'd just identify the jump locations, determine the new offset, and subtract that from the signal. Here is the code. It's s fast and simple for loop.

% Initialization steps.
clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 20;

xy = xlsread('capture 2.xls');
x = xy(:, 1);
y = xy(:, 2);
subplot(3, 1, 1);
plot(x, y, 'b-');
grid on;
title('Original Signal', 'FontSize', fontSize);

subplot(3, 1, 2);
differences = [0; diff(y)];
plot(x, differences, 'b-');
grid on;
title('Difference Signal', 'FontSize', fontSize);

offset = 0;
for k = 1 : length(y)
  if abs(differences(k)) > 0.001
    offset = differences(k);
  end
  correctedSignal(k) = y(k) - offset;
end
subplot(3, 1, 3);
plot(x, correctedSignal, 'b-');
title('Corrected Signal', 'FontSize', fontSize);
grid on;
% Enlarge figure to full screen.
set(gcf, 'Units', 'Normalized', 'OuterPosition', [0, 0.04, 1, 0.96]);

I think it does a pretty good job, don't you?

[Chinwe Orie]

This is more than a good job. Works with some other signals I have to. Thanks!

Answer 2

[Image Analyst]

I's use conv() to scan the signal with a window that covers an integer number of periods. Then I'd subtract the mean from the original signal. Something like

windowWidth = 31;  % Whatever works.
kernel = ones(windowWidth, 1)/windowWidth;
slidingMean = conv(signal, kernel, 'same');
correctedSignal = signal - slidingMean;

See how that works.

[Chinwe Orie]

Thanks. It works somewhat. It's a bit better than what I had before, but doesn't really solve the problem. Here's a picture of it after I used your solution (used a WindowWidth of 100000)

[Image Analyst]

Attach your data if you want more help.

[Chinwe Orie]

Here it is. First column is the time and the second is the voltage signal.