How can I segment a time series signal?

Hi,

Any idea how can I break a signal like on this picture?

http://www.inf.ufpr.br/aksantos/signal_example.gif

I have a sine wave segment, then noise, then another sine (different in amplitude and phase from the first one), another noise segment, and then the continuation of the first sine.

Thanks.

Andre

[Star Strider]

I’m not sure what you want to segment. The signal between the red lines appears to contain noise that the rest of the signal does not. Is the noise what you want to detect?

[Paul]

Im pretty sure he just wants to plot the red lines to indicate the segments. Andre you can use the command line for this. See:

doc line

[Image Analyst]

Make it easy for us to help you. Attach an m-file that will produce the signal you show.

Answer 1

[Image Analyst]

Do you have the Signal Processing Toolbox? The noisy segments can be detected by fitting the signal to a 5th order polynomial in a sliding window which is what the Savitzky-Golay filter does. Then subtract the sg-filtered signal from the original and threshold to find elements that are above some threshold - those will be the noisy regions.

You can find tall and short sine wave segments just by taking the RMS value of the signal. Tall segments will have a higher RMS and lower segments will have a lower RMS.

signalSquared = signal.^2
rms = conv(signalSquared, ones(1, windowWidth), 'same');

[Andre]

Nice,

thank you very much Image Analyst,

gotta one more question about this.

Do you have any idea how can i get the differences between the phases of the two signals (first SIN, and the SIN after the first noise) ?

Andre

[Image Analyst]

You could find the peaks, perhaps with findpeaks(), and then look at where the "after" peaks are with respect to where the "should" be.