how to resample non-linear breath-by-breath data (so inconsistant time)

Hi all,

I have a question regarding my data (example random numbers). If the subject had two breaths within a second there are two datapoints at the same time value. I think the biggest problem is the non-linearity of the data?

     x=[0 1 2 5 7 8 9 15 19 20 21 22 23 24 27 30 31 32 33 34 35 36 36 38]
     y=[1 2 5 3 4 8 6 5 8 5 8 6 85 85 66 44 55 66 88 77 99 88 55 66]

However, I need to resample to a 5 second time scale with it's resampled y data

x2=[0 5 10 15 20]

I wanted to use timeseries but this gave me the following error:

Warning: Cannot extrapolate

> In tsinterp>linearinter (line 145)
  In tsinterp>localInterpolate (line 240)
  In tsinterp (line 89)
  In tsarrayFcn (line 26)
  In tsdata.interpolation/interpolate (line 72)
  In timeseries/resample (line 106)

Answer 1

[Star Strider]

For your sample data, this works:

x=[0 1 2 3 4 4 5 6 7 7 8 9 10 11 12 13 14 14 15 16 17 18 19 20];
y=[1 2 5 3 4 8 6 5 8 5 8 6 85 85 66 44 55 66 88 77 99 88 55 66];
xi=[0 5 10 15 20];

dup = find(diff([eps x]) == 0);                                     % Find Duplicated ‘x’ Values
x(dup) = x(dup)+1E-8;                                               % Add ‘sqrt(eps)’ To Duplicated Values

yi = interp1(x, y, xi, 'linear', 'extrap');                         % Interploate

yi =
     1     6    85    88    66

If you have more than one repeated consecutive ‘x’ value, this gets a bit more involved, but the same approach applies.

[KFrenkie]

Ok!

But now I stumble on a second problem. The function does not interpolate on NaN values, while timeseries with resample does. Is this also possible?

Now it gives me the following:

Green: Raw dataset of the oxygen uptake with outliers deleted

red: The procedure you provided

I used the function from the file-exchange inpaint_nans once, is this applicable or do you advice another way?

Greetings

[Star Strider]

It can interpolate with NaN values. The trick is to create a continuous ‘xq’ (query) vector, then delete the entire row (or column) that contains the NaN values, and interpolate (and extrapolate if the NaN is at the end).

Example code:

M = [1:15;  2 8 7 6 NaN 8 4 9 7 2 NaN 7 1 3 5]';            % Original Data
Mn = M(~isnan(M(:,2)),:);                                   % Delete Rows With ‘NaN’
xq = 1:max(M(:,1));                                         % Query Vector

yq = interp1(Mn(:,1), Mn(:,2), xq, 'linear', 'extrap');

figure(1)
plot(M(:,1), M(:,2), '-bp', 'LineWidth',1.5)
hold on
plot(xq, yq, '--+r', 'LineWidth',1)
hold off
grid
legend('Original Data', 'Interpolated Data')

Answer 2

[KFrenkie]

Hmm.. Might have solved the problem. Can somebody comfirm my next try?

ts1 = timeseries(handles.OXY.VO2,handles.OXY.time);
time2=0:10:1110
res_ts=resample(ts1,time2,'zoh');

However, I don't know if "zero-order hold" is the right decision?