how to save frames of data into one struct?

I have the following code to read one frame data--> n points-->find 10 peaks(from n points). say I want to read 10000 frames of data, then save all the (1000*10)peaks into a struct. how do I modify the code?

a=arduino
n=20;
for ii=1:n
  v(ii)=readVoltage(a,'A0');
end
[pks,loc]=findpeaks(real(fft(v)),'npeaks',10);

Answer 1

[Walter Roberson]

repeat_count = 10000;
my_data(repeat_count) = struct('pks', [], 'loc', []);  %initialize for efficiency
n=20;
for count = 1 : repeat_count
   v = zeros(1, n);
   for ii=1:n
     v(ii) =readVoltage(a,'A0');
   end
   [my_data(count).pks, my_data(count).loc] = findpeaks(real(fft(v)),'npeaks',10);
end

Note: if you want 10 peaks out of 20 points. then you will need to select index 2 through 11 from the fft result.

Selecting peaks from the entire fft result is not the right thing to do because fft of real data returns complex conjugate data symmetric around the middle of the output.

Because peak selection requires that the point before and the point after be less than or equal to the current point, the maximum number of peaks that can be located is half of the input size -- and if you are going for that many you might as well just take every second point.

Effectively the maximum number of peaks you can find from real(fft(v)) is length(v)/4, and you should probably be aiming for less than that.

[Jun W]

1. so fft or real data is symmetrical->lenghth(v)/2, and because the points before and after should be less/equal->lenghth(v)/4. I get that.

2. does the repeat_count means 'how many times it should run the loop'? which means the total points read is repeat_count*n? if so, I get that too.

[Walter Roberson]

fft of real data is complex conjugate symmetric around the center, not directly symmetric. Because the values are complex, the total information in the first half is equal to the total information in the original signal.

Yes, the repeat count is the number of times to repeat the loop. You asked for 10000 frames of data, and you are doing a peak analysis for each frame.

[Jun W]

Thank you. Another question. I was trying to pass the generated struct into classification app. However the table is unsuitable(see image), is it because it is a 10*2 table with brackets? how do I make the 'pks' column into a 10*5 table?

[Walter Roberson]

Try

cell2mat(my_data1.pks)

[Jun W]

It worked! Thank you. But when I pass it to the classifier, there's always one row/column unsuitable, is it because it needs one row/column as label?

Also to get the first half of fft, instead of

[pks,loc]=findpeaks(real(fft(v)),'npeaks',10);

should the correct code be:

Y = fft(v);
P2 = abs(Y/length(v));
P1 = P2(1:length(v)/2+1);
[my_data(count).pks, my_data(count).loc] = findpeaks(P1,'npeaks',5);

but what is the length of the signal?

[Walter Roberson]

? length of the signal would be length(v) like you are using now, so I am not sure what you are asking about?

I suggest, by the way,

P1 = P2(1:ceil(end/2));