Algorithm for curve smoothing

10 views (last 30 days)
Sordin on 13 May 2019
Commented: Sordin on 14 May 2019
I have written a simple code that performs a 3-point moving average smoothing algorithm. It is supposed to follow the same algorithm as Matlab's smooth(...) function as described here.
However, the performance of my code is very different from that of Matlab. Matlab's 3-point filter appears to perform a much more aggressive smoothing. Why is that?
Here is my code:
NewSignal = signal;
for i = 2 : length(signal)-1
NewSignal(i,:) = (signal(i,:)+signal(i-1,:)+signal(i+1,:))./3;;
end
And here is how I called Matlab's function:
signal = smooth(time, signal, 3, 'moving');
And here is a comparison of the results: As one can see, Matlab's function smooths the data a lot further. What is the reason for the discrepancy? How can I modify my code in order for it to perform more like the blue curve?
Any explanation would be greatly appreciated.
I am including the sample data which can be accessed through:
time = M(:,1);
waveform = M(:,2);
Sordin on 14 May 2019
Sorry, that was a typo. I did use '3' and I got the plot shown above.

darova on 14 May 2019
In this part
NewSignal = signal;
for i = 2 : length(signal)-1
NewSignal(i,:) = (NewSignal(i,:)+NewSignal(i-1,:)+NewSignal(i+1,:))./3;
% NewSignal(i,:) = (signal(i,:)+signal(i-1,:)+signal(i+1,:))./3; % it's different. try it
end
More generic version
NewSignal = signal;
n = 3;
for i = 1 : length(signal)-n+1
s = 0;
for j = 0:n-1
s = s + signal(i+j,:);
end
NewSignal(i,:) = s/n;
end
Sordin on 14 May 2019
Thanks a lot for attempting to generalize the code. But note that the span is always an odd number. So instead of 1, the start point of the loop has to be 2 for a 3-point filter, or 3 for a 5-point filter, etc.
Also, for some reason your code results in less smoothing than mine. In any case, our code is very different from the result returned by Matlab's smooth(...) function. It is as if it uses a much higher span than 3.