Fixing data set with if/else, how do I index/indicate this properly?

I have a data set of seconds (plotted image below).

Rather than consistently increasing, they are dropping down and resetting. I am trying to write something that fixes this so the end result will be them spaced as they are now but in a consistently increasing line. I think I want to indicate that each time a value v is lower than the previous (v-1), add the difference between the two to v and keep going. I am new to Matlab and started to write something like this:

for v=1:length(Seconds)

      if (Seconds(v+1) <= Seconds(v))
          Seconds(v+1) = Sec

onds(v+1)+(Seconds(v)-Seconds(v+1));

      else
          (Seconds(v+1)) = Seconds(v+1);
      end
  end

How do I properly assign the new values? Thank you

[dpb]

yy=cumsum(y);

may be what you're looking for?

[L]

I'm not sure cumsum is what I need to use here. Rather than the cumulative sum, I need to adjust each value such that

if (Seconds(i+1) <= Seconds(i))
        Seconds(i+1) = Seconds(i)+(Seconds(i)-Seconds(i+1))

The values are not equally spaced and I do not want to add each of the previous values to every value. Please let me know if there is anything I can clarify.

[dpb]

Yeah, shorten the file to the first one (or at most two) cycles and show us what you think the answer would be to go with it...afaict, what you've written about is what cumsum does.

[L]

Thanks for the follow up. I've read more about cumsum and if I take, for example, this point marked in the image below, I can understand that I would need to add all the previous values to "stack" it. Because the first line of data is correct, though, I would need to exclude it from cumsum calculations until I need it for the start of the second line set of data, correct? (I want my data points to stay as they are until they drop back down to 0 for the first time)

Answer 1

[dpb]

OK, to increment every element from the previous by its absolute matnitude change is I guess what you're asking to do...

yy=cumsum([y(1);abs(diff(y(:)))]);

In this, then,

sum(diff(yy)<0) --> 0;

every point is the previous plus the change from the previous as if it were still growing rather than being cyclic; or, iow, as if you had rectified the signal.

NB: IF the data are a column you don't need the (:); I used it to ensure knew the orientation of the result from diff to do the concatenation with the first element.

ADDENDUM

W/ the revision of what the "reset" really is (I had presumed a sampled waveform, not the ramp)...

dy=diff(y);                          % get the differences 
ibrk=find(dy<0);                     % find the breakpoints
y(ibrk+1)=y(ibrk+1)+abs(dy(ibrk));   % fix up those points to match previous
dy(ibrk)=0;                          % now eliminate the reset offset
yy=cumsum([y(1); dy]);               % build the ramp

This ignores the ramp between the end of one cycle and beginning of next--those two points are same. If you don't want that plateau, you can estimate a slope and introduce an estimated dy at each break instead of setting to zero to cancel the reset.

[L]

Thank you. Growing rather than cyclic is exactly what I am trying to get. Here's a sample plot of the result. I think there may be an issue at the boundaries. For example, the first boundary was from 8.60928650000e+04 to 1.275610000e+02 in the original data set. In yy it is 8.609286500000000e+04 to 1.720581690000000e+05. I think at this boundary, for example, I would need to add 8.60928650000e+04 to 1.275610000e+02 and then sum the elements as you have written. Does this make sense?

[dpb]

Again, it would be much easier with a small subset of the data; I created something here to approximate what your figure shows and it doesn't show such jumps...I don't quite see where those must have come from with the code I gave; the difference should be just about the same as the previous.

Attached is plot of my test...multiplying to get big numbers doesn't do anything but change the scale of the axes; the relative sizes between points will still be the same.

I guess if the data aren't sampled uniformly, maybe there's the culprit but I can't tell what is really the underlying process just looking at the plot.

Attach a very short section via a .mat file; it only needs one segment breakpoint to see what's going on...

[L]

I think a simple cumsum(y) may be working.

[dpb]

Well, again, that depends on what your intent is...you claimed earlier you didn't want the sawtooth that it produces by reflecting the downside direction. :)

But, it was my firstest thought, yes! :)

ADDENDUM

Oh, the alternative you might consider would be cumtrapz

[L]

Small sample including 1 day break

[L]

Segment breakpoint at 194/195

[dpb]

Ah! That's a completely different form than I had presumed--that's a periodic reset ramp, not a sampled waveform...the problem w/ too many points on the plot not showing the actual point locations, drew wrong conclusion.

See ADDENDUM above...

[L]

Thanks, this was great. I really appreciate the help!