MATLAB: filter nonperiodic noise in the signal

Question

I have gone through the below related posts of filtering noise from the signals

MATLAB, How to filter a discrete signal?

But the signal I want to filter contains noise which is not periodic as shown in the images below:

Can anyone suggest a filtering technique for the noise labeled with red colour in the image?

score 1 · Answer 1 · answered Sep 19 '13 at 10:00

1

You can do a non-decimated one dimensional wavelet transform on the image. Since the data is more or less homogeneous within a certain range, the outliers will be denoted by spikes in the wavelet coefficients. You can then use a threshold and remove those coefficients and perform inverse non-decimated wavelet transform.

You can check this paper also.

Do you mind sharing a copy of the data in .mat ? I would like to test it out in order to be sure.

answered Sep 19 '13 at 10:00

sm176357

86
4

Hello baptu88, how to share the .mat file here.Thanks. – user1016528 Sep 25 '13 at 11:17

score 1 · Answer 2 · edited Jun 18 '22 at 12:44

1

This type of noise shouldn't be difficult to remove.

One approach to try is to have a moving window computing the mean and variance. Which are then used to select a clipping value $c$.

As the window moves signal values that are higher or lower than $c$ are labeled as noise.

The method assumes a window clean from noise at the beginning of the signal.

A somewhat similar type of noise can be found with partial discharge data, this paper shows another approach, based on the correlation function, that can be adapted to your application

edited Jun 18 '22 at 12:44

lennon310

3,590
19
24
27

answered Sep 20 '13 at 04:05

Hasan

485
3
4

Hello Hasan,could you please explain it with an example. – user1016528 Sep 25 '13 at 11:10
1

take a window lets say of length 5 columns. compute the mean and the variance for the data within this window. any value higher than u + k $\sigma$ or lower than u - k $\sigma$ is considered noise. Next, add the 6th column to the window, drop the first, recompute noise values...keep iterating till the end of the signal. Obviously, you need to experiment to find the correct window length and k. – Hasan Sep 28 '13 at 19:36

score 0 · Answer 3 · answered Oct 30 '13 at 05:48

It looks like the noise, while not perfectly periodic, is concentrated about one specific frequency. I would guess that this noise is caused by electrical power frequency (usually 50 Hz or 60 Hz depending on your location). If you plot the spectrum of the signal you have, you should be able to find a strong component in this frequency range. You can then remove most of it with a simple notch filter.

MATLAB: filter nonperiodic noise in the signal

3 Answers3