Find sudden jumps in signal

Hello,

I have a recovery signal of cells post photobleaching, and I am trying to find an algorithm on Matlab where it detects sudden changes in signal (primarily sudden increases). Is there a custom MATLAB function for this? I found something called 'findchagespts', but in this one you have to mention how many changes you are expecting. I am also looking for an algorithm where it gives some statistics value of how much the change is.

Answer 1

[Kaiguang Zhao]

Here are some extra thoughts. Not sure about to what extent these will address your problem, but at least, some aspects (e.g., abrupt changes and their signs) can be estimated from a Matlab tool called BEAST (Bayesian estimator of Abrupt Changes, Seasonality, and Trend) developed and maintained by me at https://www.mathworks.com/matlabcentral/fileexchange/72515-bayesian-changepoint-detection-time-series-decomposition?s_tid=FX_rc1_behav. The tool can be easily installed by running eval(webread('http://b.link/beast',weboptions('cert',''))) .

Below is a minimal example to explain:

eval(webread('http://b.link/beast',weboptions('cert','')))   % Install BEAST and some test datasets
load('Nile.mat')                  % Annual flow of the Nile river
o =  beast(Nile, 'season','none') % season='none' is used bcz Nile has no periodic/seasonal variation   
printbeast(o)
plotbeast(o)
o.trend.cp
o.trend.cpAbruptChange
o.trend.cpPr

Below is the plot of the detected sudden change(s). In this particular time series, only one sigficant changepoint is found. The Pr(tcp) curve shows the probability of changepoint occurence over time.

o.trend.cp gives the list of identified changepoints; o.trend.cpAbruptChange gives the associated delta_change corresponding to the changepoints in o.trend.cp; and o.trend.cpPr gives the probabilty of the detected changepoints being true ones.

uninstallbeast % uninstall BEAST

Answer 2

[Chris Turnes]

If you have access to R2017b, try taking a look at the ischange function. The name-value pair 'Threshold' will let you tune a sensitivity to detecting changepoints, rather then specifying an explicit number of changepoints.

[Kayihan Cömert]

With 'linear' method this finds all the changes regardless of if it increases or decreases. In his case and in my case, we are trying to find the just sudden increases or just sudden decreases.

Answer 3

[Hussam Ibrahim]

Thank you for your input. I think I have a license for the b version. I will install and run it to see if it makes a difference. Thank you.

[Chris Turnes]

There is also a similar option in findchangepts called 'MinThreshold'. The notes at the bottom of the documentation page discuss the exact mathematical meaning of this option.