Introducing new rows (or columns) into an existing matrix with interpolated values to fill technically non-existent gaps.

I have a dataset from a GPS unit that occasionally skips a second or two. I'd like to fill in the missing time gaps with the average of the previous and subsequent rows (well actually, that wouldn't work if there was a two second gap). Note: Each time I run this, the set of GPS data will be different and may or may not have these gaps in time.

For example, here is a snippet of what the matrix of interest looks like. The first column is a reference that mathes up with other data; After that it is YY MM DD HH MM SS. As you can see, the 49th second is missing.

GPS_time = [4791 20 11 8 23 32 47; 4792 20 11 8 23 32 48; 4794 20 11 8 23 32 50; 4795 20 11 8 23 32 51]

If you don't know but would like to help, I have two thoughts on how to approach this:

1. Create a complete reference from the first and last values of the reference column with no gaps to merge with the GPS_time matrix and using fillmissing to replace the missing values. I am trying to do this now, but am having trouble.

2. Give up and use the first and last rows of data from GPS_time to re-build the matrix from scratch. This doesn't sound fun because of the way time wraps around on itself before adding to the previous columns value.

I am using R2017a, btw.

Answer 1

[Walter Roberson]

Create a timetable() object and use retime() https://www.mathworks.com/help/releases/R2017a/matlab/ref/retime.html

[maffew]

Thanks for your input.

Instead of using timetable, it seems more managable to use datetime and then datevec to recreate the matrix with the missing values. This was I can easily append the reference values.

 t1 = datetime(2020,11,8,23,32,47);
 t2 = datetime(2020,11,8,23,49,10);
 t = t1:seconds(1):t2;
 t = t';
 t = datevec(t);

I would still like to know how to approach my original problem because it exists with data sets that are not time at all. But I still haven't been able to find a way to interpolate "missing" data. I think I would need to use my first approach to match a complete reference to force the dataset to have NaN values and then use fillmissing.