adaptive savitzky golay to find optimum order and window size

Hi

I need to apply adaptive SG filter to my data to find the optimum order and window size. I used below code

to generate smoothed signal from the noise one , I need to find SG order and window size that preserve the

peaks in the signal and reduce the signal fluctuation between them .

for order =(0:1:9) ;
for framelen =(1:2:51);
x=-AV;
lx = 20;

sgf = sgolayfilt(x,order,framelen);

 plot(sgf);
 hold on;

 end
 end

Answer 1

[Image Analyst]

Find the peaks first with findpeaks(). Then replace those locations with the original signal in small windows around the peaks. So make a logical vector that's true where you want original spikes and false where you want the SG filtered signal, then do

sgf(peaksMask) = x(peaksMask);

[sarmad m]

could you please give me an example , it is not clear to me

[Jan]

@Image Analyst: This is a strange kind of filtering: Smoothing all data but keeping the peaks. The result will look like a perfect measurement. This is useful for image enhancing, but evil for the documentation of a scientific measurement. Perhaps the OP should search for something different.

@sarmad m: He means that you can filter the signal as you want and keep the original values for the peaks. Then you can freely choose the filter order without touching the peak levels.

How do you define "optimum" here? In which norm will you measure it? Is the goal to improve the optical impression, or do you want to reduce noise of a scientific measurement? For the later case it would be much better to determine the source of noise and create an exactly matching lowpass filter.

[sarmad m]

@ Jan Simon

I need to find the order and window size of SG filter that can preserve the peaks and reduce noise . Currently i'm changing the ( order and framelen ) of SG for each new data file . as you see in the image above I tried different order and window size but I don't know how to find out the best parameters

[Jan]

@sarmand m: All orders and window sizes do reduce the noise and preserve the peaks in a certain degree. You are looking for the "optimal" order and window size, but you did not specify how this is measured. How do you decide which filter parameters are "better"? There is no magic set of parameters which leave the peaks untouched and removes all noise. You cannot even know if the exact peak height is caused by noise.

Do you see the problem? To find the "best" parameters, you have to start with defining what "good" is mathematically.