How to automate extraction of multiple parts of a signal?

Hi all,

I have a signal ('gas_lat_env_norm') collected over a 10 s period ('time'). I have identified timings of certain events ('l_heel_strike_uncroppedtime') and plotted them over the signal (events 1,2,3 red circles).

I want to:

• extract the signal between 1 and 2
• extract the signal between 2 and 3

I am able to do this in this way:

load 'gas_lat_env_norm' % load signal
load 'time' % load time 
load 'l_heel_strike_uncroppedtime' % load timings of events

% Extract the signal between event 1 and event 2

ind1=time>=l_heel_strike_uncroppedtime(1) & time<=l_heel_strike_uncroppedtime(2); % cycle 1
gas_lat_cycle1 = gas_lat_env_norm(ind1);

% Extract the signal between event 2 and event 3

ind2=time>=l_heel_strike_uncroppedtime(2) & time<=l_heel_strike_uncroppedtime(3); % cycle 2
gas_lat_cycle2 = gas_lat_env_norm(ind2); % signal 

% plot first extracted signal

figure('color','w');
plot(time(ind1),gas_lat_env_norm(ind1))
ylabel('% of max activity');
xlabel ('Time (s)');
box off;
axis tight;

% plot second extracted signal

figure('color','w');
plot(time(ind2),gas_lat_env_norm(ind2))
ylabel('% of max activity');
xlabel ('Time (s)');
box off;
axis tight;

% Make a mean of gas_lat_cycle1 and gas_lat_cycle2?
gas_lat_cycle = (gas_lat_cycle1 + gas_lat_cycle2)/2

Error: Matrix dimensions must agree.

However, the number of events may differ depending on a signal and therefor I would like to make this process automated and

• extract the signal betweene 3 and 4 and so on (4-5; 5-6....) (if such would(and will) be present)
• make the mean and standard devation of the mean of each extracted signal periods and save it as one variables

Could you please help?

Answer 1

[Image Analyst]

You can threshold at 10 or 15 or so, then look for starting and stopping indexes

inPeak = gas_lat_env_norm > 10;
startingIndexes = strfind(inPeak, [0, 1]);
endingIndexes = strfind(inPeak, [1, 0]);

I presume you know what to do from there.

It could be a little easier to extract certain peaks if you have the Image Processing Toolbox where you can label (identify) individual peaks and then use ismember() to get just the peak you want.

[Tomaszzz]

@Image Analyst

Thanks a lot of you answer.

I probably do not understand but it seems your lines basically do what my code already does, but diffrently.

I am able to extract signal between 1 and 2 and 2 and 3 but my script wrongly assumes (not in this example) there are only 3 events. I would like this process to be independent on how many events there is , something like:

for lenght of n of events

extract signal between 1 and 2

extract signal between 2 and 3

extract signal between 3 and 4

and so on

end

So I am afraid it migh be actually about "I presume you know what to do from there".

[Image Analyst]

There are lots of peaks all over that signal, especially in the valleys. I'm not sure how you define the peaks to take. Can we just threshold and look for runs of a certain length/time above the threshold?

Also, define more exactly what "extract" means to you. Is it like this:

numPeaks =  length(startingIndexes);
extractedSignals = cell(numPeaks, 1);
for k = 1 : numPeaks
    extractedSignals{k} = gas_lat_env_norm(startingIndexes(k) : endingIndexes(k));
end

Each extracted signal will be a different length. Or you can mask the signal by keeping the same length signal and just zeroing out elements outside the peak region.

numPeaks =  length(startingIndexes);
extractedSignals = zeros(numPeaks, length(gas_lat_env_norm)); % Initialize
for k = 1 : numPeaks
    indexRange = startingIndexes(k) : endingIndexes(k);
    % Assign values inside the peak region.  Outside the region will remain zeros.
    extractedSignals(k, indexRange) = gas_lat_env_norm(indexRange);
end

Then do whatever you want with the extracted signals. All of the above assumes that startingIndexes(1) comes before endingIndexes(1).

[John Navarro]

Just read the size and times of "l_heel_strike_uncroppedtime", loaded to the new vector and use it to define the starting and ending points, like Image Analysit propose.

In others words create a vector which size +1 bigger than the number of events and then load the times into that vector such as

a= [0, l_heel_strike_uncroppedtime(1), l_heel_strike_uncroppedtime (2) ... l_heel_strike_uncroppedtime (n)],

Using a for loop could help.

[Tomaszzz]

Thanks a lot. This will do the job.