I’m not certain that I understand in detail what you’re doing.
Using datenum is likely the best option here. I experimented with caldiff (using synthetic data) and was not able to get a satisfactory result of the sort you probably need, although with your data it may work.
The diff function has the disadvantage of shortening the output by one element, since it takes successive differences. One alternative is to compute the numeric derivative with the gradient function, and to get the acceleration from the velocity and time vectors, use them together as:
dVdt = gradient(velocity) ./ gradient(time);
The calculation you choose entirely depends on the result you want.
