Distinct eigenvalues of a normal matrix

Given a matrix $\displaystyle X\in\mathbf{R}^{3\times n}$ and a diagonal matrix $\displaystyle D$ with known positive diagonal entries, what conditions may I impose on the columns of $\displaystyle X$ such that $\displaystyle XDX^T$ has distinct eigenvalues?

Here are some things that I already know:

If $\displaystyle D=\text{diag}(a)$ and $\displaystyle a\in\mathbb{R}^n$, then

$\displaystyle XDX^T=\sum_{j=1}^na_jx_jx_j^T$.

Also, this seems to be the most promising lead I have:

$\displaystyle \lambda_i=\sum_{j=1}^na_j(x_j^Tv_i)^2$,

where $\displaystyle \lambda_i$ and $\displaystyle v_i$ is an eigenvalue/eigenvector of $\displaystyle XDX^T$.

Because $\displaystyle XDX^T$ is Hermitian, the eigenvectors define an orthonormal basis. Thus the change from the canonical basis to a basis of the eigenvectors is solely a rotation which does not change the geometry of the vectors $\displaystyle x_j$. Therefore, it seems that any geometrical properties that I might find before the change of basis might still hold, but I don't know how I can prove this.

Thanks