In general you need both kinds of functions $H_n^{(1)}(k\,r)\,e^{i\,n\,\phi} = (J_n(k\,r) + i\,Y_n(k\,r))\,e^{i\,n\,\phi}$ and $H_n^{(2)}(k\,r)\,e^{i\,n\,\phi} = (J_n(k\,r) - i\,Y_n(k\,r))\,e^{i\,n\,\phi}$ in any superposition, where $(r\,\phi)$ are the polar co-ordinates and $k$ the wavenumber. If the homogeneous region in question contains the point $r=0$ and you can be sure that there are no singularities at the origin (which is almost always, but see later on), then the superposition weights of the corresponding kinds of Hankel function must be the same: they must arise in pairs of the form $\alpha\,(H_n^{(1)}(k\,r) + H_n^{(2)}(k\,r)))\,e^{i\,n\,\phi} = \alpha\,J_n(k\,r) \,e^{i\,n\,\phi}$ because the Bessel functions of the second kind (Neumann functions) have branch points at the origin (and the spherical Hankel functions of order $n$ , which are the normal ones divided by $\sqrt{k\,r}$ have poles of order $|n|+1$).
If you're working with a field in a homogeneous cylindrical shell which doesn't include $r=0$ then arbitrary superpositions of all the functions are possible.
Now you can just as well say that your solutions are superpositions of the functions $J_n(k\,r)\,e^{i\,n\,\phi}$ and $Y_n(k\,r)\,e^{i\,n\,\phi}$, with the $Y_n$ terms absent in any homogeneous region including $r=0$ and everything will be equivalent. The reason that people work with Hankel functions instead is their neat asymptotic behavior:
\begin{align}
H_\alpha^{(1)}(z) &\sim \sqrt{\frac{2}{\pi z}}\exp\left(i\left(z-\frac{\alpha\pi}{2}-\frac{\pi}{4}\right)\right) &&\text{ for } -\pi<\arg z<2\pi \\
H_\alpha^{(2)}(z) &\sim \sqrt{\frac{2}{\pi z}}\exp\left(-i\left(z-\frac{\alpha\pi}{2}-\frac{\pi}{4}\right)\right) && \text{ for } -2\pi<\arg z<\pi
\end{align}
and so they decompose an EM field into inward propagating and outward propagating cylindrical waves, so they are especially useful for studying antenna problems with cylindrical symmetry, or the radiation shed from an cylindrically symmetric optical fiber, for example. If you know you've only got radiation going outwards, you use only the $H_n^{(1)}(k\,r)\,e^{i\,n\,\phi}$ functions in the superposition. If you know you've only got radiation coming inwards, you use only the $H_n^{(2)}(k\,r)\,e^{i\,n\,\phi}$ functions (see footnote).
But if you're interested in plane wave expansions, you include the point $r=0$ and there are no singularities in a plane wave, so that only the $J_n$ terms can be present. Indeed, the plane wave expansion follows from the generating function for the sequence $\{J_n\}_{n=-\infty}^\infty$ or the Jacobi-Anger expansion:
$$ e^{i z \cos \theta} = \sum\limits_{n=-\infty}^{\infty} i^n\, J_n(z)\, e^{i n \theta}$$
(the generating function formula is $e^{i \,z\,(t-t^{-1})}=\sum\limits_{n=-\infty}^\infty J_n(z)\,t^n$). Putting $z\,\cos \theta=k\,r\,\cos \theta=\vec{k}\cdot\vec{r}$ then yields the plane wave expansion:
$$e^{i\,\vec{k}\cdot\vec{r}}=\sum\limits_{n=-\infty}^{\infty} i^n\, J_n(k\,r)\, e^{i\, n\, \theta}$$
Footnote: Antenna Problems
Sometimes it is physically reasonable to have the singularity at $r=0$, at least at the scale you consider the problem. If, for example, you model an antenna by a thread current with constant phase along the thread's length, you think of the solution of the Helmholtz equation as $H_n^{(1)}(k\,r)\,e^{i\,n\,\phi} = (J_n(k\,r) + i\,Y_n(k\,r))\,e^{i\,n\,\phi}$ for a radiating antenna and $H_n^{(1)}(k\,r)\,e^{i\,n\,\phi} = (J_n(k\,r)- i\,Y_n(k\,r))\,e^{i\,n\,\phi}$ for a receiving antenna. The singularity arises from the idealization of a thread current, with zero diameter. This kind of calculation you would use to model the antenna's transmission and so forth. But if you need to design the conductor to carry the required current, or if you need to work out the radiation impedance of the system (whose reactive component diverges to infinity for a thread current), then you're going to break the problem up into two regions: one outside the wire modelling the radiation and containing only $H_n^{(1)}(k\,r)\,e^{i\,n\,\phi}$ terms (if a radiating antenna) and one inside the wire, where there are only functions of the kind $J_n$. Actually these latter have a very different, complex wavenumber to account for the conductor, but the singularities are still the same for complex arguments, so there are only $J_n$ terms present.
