How to avoid error calculation going to infinity when dividing by near-zero values?

Hello,

I am calculating the relative error between two signals using the next equation:

relative_error(i) = abs((measured(i)-estimated(i))./measured(i))*100;

The problem arises when measured is zero or near zero that relative_error explodes.

Do you know a more appropriated way of calculating an error between two signals that avoids this problem?

Thanks,

Cerilet

Answer 1

[Honglei Chen]

There are several ways to handle this, for example, you can add one more line to set all those points to a predefined value

relative_error(i) = abs((measured(i)-estimated(i))./measured(i))*100;
relative_error(measured==0) = 1;  % say predefined value is 1

Or just add an epsilon to avoid this issue

epsilon = 1e-8;
relative_error(i) = abs((measured(i)-estimated(i))./(measured(i)+epsilon))*100;

HTH