On one hand, JavaMan is right: since the function $f$ is not in $L^1+L^2$ on $\mathbb R$, the classical
definitions of the transform do not apply. On the other, the Fourier transform is positive in the sense of distributions;
that is, it is a positive measure on $\mathbb R$. A nice measure, actually: smooth function plus a single
point mass at $0$. Indeed, we have a series
$$f(x)= \frac{1}{2}\frac{1}{1-2^{-1}e^{-x^2}}
=\frac12\left(\sum_{n=0}^\infty 2^{-n}e^{-nx^2}\right)
\tag1$$
which converges in almost every sense imaginable. The terms with $n\ge 1$ are Gaussians which
transform to other Gaussians (positive functions). The term with $n=0$ is $\frac12$ which is of course
not integrable, but its distributional Fourier transform is a positive point mass at $0$.