allanvar() surprising behavior for large Tau

I compared Matlab build-in allanvar() funtion to script provided in NXP AN5087 and can see significant difference for Allan deviation at large tau.

This happens for very large data count, e.g. >1e7.

AN5087 is much closer to expected behavior: slope -1/2 then slope +1/2 !

So I would not recomend using allanvar() for very large datafiles !

Answer 1

[Sean]

I think I'm observing the same problem. It is easy to reproduce using the following code

[sigma,tau] = allanvar(rand([1,1e8]));
figure;loglog(tau,sigma,'.:')