Find jumps in Signal take constant area and split signal to different mat files

Hey guys,

I have following Signal:

I need to split this signal into different mat files.

My problem i have the noise and the transient behaviour.

I need to save the constant "areas" into the different mat files.

In this case I have 7 to 8.

Thanks in advance.

[darova]

Kind of long code

clc, clear

noise = 0.1;
n = 10;        % number of areas
jump = 1;
cur_jump = 0;

t = linspace(0,30,500);
signal = round(t./10) + rand(1,length(t))*noise;
ds = diff(signal);

t = t(abs(ds)<noise);
s = signal(abs(ds)<noise);

ar = zeros(length(signal),n);
time = zeros(length(signal),n);

ns = 0;
while cur_jump < max(s)
    nj = 0;
    ns = ns + 1;
	for i = 1:length(s)
        if cur_jump-jump/2 < s(i) && s(i) < cur_jump+jump/2
            nj = nj + 1;
            ar(nj,ns) = s(i);
            time(nj,ns) = t(i);
        end
    end
    cur_jump = cur_jump + jump;
end

subplot(3,2,1)
plot(t,s)
hold on
for i = 1:ns
    subplot(3,2,i+1)
    plot(time(:,i), ar(:,i),'r')
end
hold off

[youjarr]

Hey and thanks for your help.

I tested your Code and i get this plot:

What are those lines? Is there a way to remove them and which part of the code do I need? I think this part here:

ds = diff(signal);

t = t(abs(ds)<noise);
s = signal(abs(ds)<noise);

ar = zeros(length(signal),n);
time = zeros(length(signal),n);

ns = 0;
while cur_jump < max(s)
    nj = 0;
    ns = ns + 1;
	for i = 1:length(s)
        if cur_jump-jump/2 < s(i) && s(i) < cur_jump+jump/2
            nj = nj + 1;
            ar(nj,ns) = s(i);
            time(nj,ns) = t(i);
        end
    end
    cur_jump = cur_jump + jump;
end

Everything above is creating the signal and after the plot.

[youjarr]

This is what is happening when I put my data in... .

It seems it get the area to early and ends the area to late

[darova]

Did you adaptate those variable for you model?

Those lines you asked are zeros element in ar and time. Number of current "area" less than number of elements in ar

[youjarr]

Yes it is mostly the same -> noise = 0.1 and jump = 1. I changed the 0 to NaNs now.

The green marked area is the area i want to be saved.

[darova]

can you attach your data?

[youjarr]

Yes. This is the Y-Axis.

Just Build a vector

t = 1:1:length(signal) 

should be the same effect.

Answer 1

[youjarr]

Maby an easier question.

How do I get the first value when it reached the next jump? I mean if it gets constant (green lines) that would be the best.

Answer 2

[darova]

I tried to smooth data. Then get every 600th point to compare with noise value and filter.

Then comparing points to find stable areas.

Hard to pick parameters for data (smoothing and filter). Maybe with user interaction it will be a little bit easier to draw those areas. See attached script

[youjarr]

Thank you very much for your effort.

I am going to test it later.

But can I adpat this code 1:1 to following data?

Or do I have to change smth. in the code?

[darova]

Hard to say. It depends on data, try to change smooth coefficient and filter point if needed.