In your paper is solved jumps removing in a case of model:

x1 = const + trend + modulation + noise;

e.g. for example:

x1 = 300 + t + sin(t) + randn(size(t));

If yes, could send me this paper for info?

Constant term is not solved in your code, too.

In general I am looking for effective and reliable method to eliminate jumps in signal with constant offset + trend + noise + signal

]]>t = 0 : 0.1 : 10*pi;

x1 = t + sin(t) + randn(size(t));

x1(1,131:315) = x1(1,131:315) + 5;

x1(1,170:315) = x1(1,170:315) – 10;

just try the command:

findchangepts(x1,’Statistic’,’linear’,’MaxNumChanges’,2)

or

findchangepts(x1,’Statistic’,’linear’,’MinThreshold’,50)

You get nice trend segments

]]>In my case the signal has typically three signal components:

1. trend

2. global amplitude modulation

3. noise

Something like this, for example:

x1 = t + sin(t) + randn(size(t))

I am looking for any methods which will be able to remove jumps in this kind of signals.

Do you have any idea how to modified findchangepts approach in this case? I think the ‘linear’ statistics option should be used now instead of ‘mean’, but I am not sure.

Could you add final code with findchangepts function? ]]>