Using an eigenmode decomposition technique, we numerically determine the backreflection coefficient of the modes of air-core photonic bandgap fibers for flat terminations. This coefficient is found to be very small for the fundamental air-guided mode, of the order of 10-5 to 10-6, in contrast with the surface and bulk modes, which exhibit significantly higher reflections, by about three to four orders of magnitude. For the Crystal Fibre HC-1550-2 fiber, we find a reflection coefficient of 1.9×10-6 for an air termination, and approximately 3.3% for a silica termination. We also find that the Fresnel approximation is ill suited for the determination of the modal reflection coefficient, and instead propose a more accurate new formula based on an averaged modal index.
© 2007 Optical Society of America
Air-core photonic bandgap fibers (PBFs) [1–3] offer the enticing theoretical prospect of very low nonlinearities, reduced sensitivity to environmental factors such as temperature , potentially low backscattering and hence very low propagation losses, and they are finding interesting novel applications in various photonics fields, including fiber sensors and high-power laser delivery. As is the case for conventional solid-core fibers, the performance of a fiber optic system based on PBFs is expected to be adversely affected by the reflection that takes place at the end of an air-core fiber, whether the fiber is terminated in air (free end) or butt-coupled or spliced to another air-core or solid-core fiber. Backreflection from such terminations are notoriously detrimental to the performance of fiber interferometers, communication systems, and lasers and amplifiers, as they result in increased noise, instabilities, and other deleterious effects.  While the reflection at the end of a photonic crystal fiber has been studied,  evaluating the magnitude of the reflection at the end of an air-core fiber, such as the HC-1550-2,  has never been reported before and it is therefore becoming a pressing issue.
The backreflection from the end of a conventional fiber terminated by a semi-infinite medium of refractive index n 0 is characterized by a power reflection coefficient that is usually calculated by the well-known approximate Fresnel formula:
where neff is the effective index of the mode considered. The Fresnel approximation is equivalent to replacing the waveguide with a medium of uniform refractive index equal to the mode effective index, and to approximate the mode by a plane wave at normal incidence. As a result, the Fresnel approximation performs best for weakly-guiding waveguides with a small index contrast across the waveguide cross-section, for modes possessing a small numerical aperture (lower order modes), and for a strong refractive index contrast between the waveguide and the termination medium. Conversely, this approximation has proved to be inaccurate to evaluate the reflection coefficient of strongly guiding waveguides,  and it does not account for the spatial intensity distribution of the mode. Because PBFs are strongly guided waveguides, and because we are mostly concerned with the reflection of a leaky mode (the fundamental air-guided core mode) into air (low refractive index contrast between waveguide and termination medium), we expect the Fresnel approximation to be ineffective in this context, and that a more elaborate calculation will be required.
In this paper, we detail for the first time a computation procedure for the derivation of the reflection coefficient of a PBF mode into a medium of arbitrary refractive index, which we then apply to calculate numerically the reflection coefficient of the fundamental core mode, as well as representative surface and bulk modes, of an air-core fiber, and their dependencies on wavelength and fiber parameters. We then study the influence of PBF design parameters on the reflection coefficient, in order to design PBF components with very low reflections for improved performance in fiber devices. We also derive a simple formula based on the Fresnel approximation that provides accurate predictions of the backreflection from the end of a PBF, especially for fibers with a high air-filling ratio (above 90 %).
The configuration studied is represented in Fig. 1. It consists of a semi-infinite PBF, extending over the z < 0 region (region I). The z > 0 region (region II) is a semi-infinite medium of uniform index n 0, such as air. The PBF parameters are as follows: the cladding crystal periodicity is Λ, the core radius is R, and the cladding air hole radius is ρ. The transverse plane is hereafter defined as the interface plane terminating the PBF. We consider the reflection at this interface of a mode field propagating through the PBF in the +z direction.
To calculate the reflection coefficient, we use the formalism of eigenmode decomposition.[9,10] The incident and reflected fields (in region I) are decomposed into a sum of PBF modes and the transmitted field (in region II) into a sum of plane waves. The transverse incident electrical field is denoted EI(r) and can either be a core mode, a cladding mode, or a surface mode of the PBF, labeled E0(r). The reflected transverse electrical field, labeled ER(r), propagates in the -z direction and is written as a superposition of core and cladding mode fields of the PBF, which are collectively labeled Em(r).
The modal decomposition of the reflected transverse electrical field is therefore written as:
where the summation is carried over (NM+1) modes, and am is the amplitude of the signal reflected into mode m. Furthermore, only modes of the same symmetry class as the incident mode will be present in the decomposition of the reflected field. For most application, we are concerned with the amount of light backreflected into the incident mode, which is labeled as m = 0, since all other coefficients are much smaller (by many orders of magnitude). The intensity or power reflection coefficient into the fundamental mode is:
In the same fashion, the transverse transmitted electrical field propagating in region II in the positive z direction is denoted ET(r) and decomposed in a sum of plane waves propagating in free space. The transverse magnetic fields are decomposed in the same fashion. Using the continuity of the transverse electric and magnetic fields, we obtain the following relationships at the PBF interface plane:
As described in the appendix, we can then use the orthogonality of the transverse fields of the waveguide modes, and the orthogonality of the transverse fields of the plane waves, to transform these equations into a system of linear equations, which are then solved numerically to recover the reflection coefficients am.
The PBFs were modeled using the Stanford Photonic Bandgap Fiber (SPBF) code, a finite-difference mode solver developed for the purpose of modeling the modes of waveguides with complex refractive index profiles such as air-core fibers.  In the simulations carried out for this paper, the SPBF code was used with periodic boundary conditions, with a simulation step of Λ/50 and a grid size of 10Λ×10Λ. We used typically 260 PBF modes and between 2000 and 5000 plane waves for the computation of the reflection coefficients.
To compute the projection of the PBF modes onto plane waves, we used a two-dimensional discrete Fourier transform (2D DFT) of the computed fields. Figure 2 illustrates a typical matrix A. The non-zero coefficients are illustrated as blue dots, and the zero coefficients are left out (shown as white spaces). The upper left corner indicates the projection of PBF modes onto each other; the lower right corner shows the projection of plane waves onto each other. The remaining portions of the matrix show the projection of PBF modes onto plane waves propagating in air. The matrix size is 2082 × 2082 elements, of which nearly 950,000 are non-zero complex elements. This sheer number of non-zero elements makes it impossible to invert such a matrix accurately directly, so we resorted to an iterative method, described in Ref .
3. Variation of the fundamental mode reflection coefficient across the PBF bandgap
In the first part of this work, we modeled the PBF whose refractive index profile is illustrated in Fig. 3. The cladding holes' radius is ρ = 0.47Λ, and the core radius R = 0.8Λ. Figure 4 represents the calculated bandgap of this fiber, as well as the dispersion curves of all of its core modes. The bandgap extends from λ = 0.56Λ to λ = 0.64Λ. Due to the choice of core radius,  this fiber does not support surface modes, and it has only two core modes, the two degenerate, orthogonally polarized fundamental modes. Figure 5 shows the longitudinal Poynting vector profile of one of the fundamental modes at λ = 0.60Λ. The fundamental core modes have a very large fraction of their intensity (over 90 %) concentrated in air, which is the reason why a weak reflection at an air interface is expected.
The Fourier transform intensity of this same mode's transverse electric field, plotted in logarithmic scale in the wavevector domain, is shown in Fig. 6. The black circle shows the boundary for all wavevectors such that , i.e., all wavevectors admissible for plane waves propagating in air. Since the fundamental mode profile exhibits little variation across the bandgap, as the wavelength varies across the bandgap, the most significant change occurs in the value of k 0, and thus in the radius of this circle. As the wavelength increases, the radius of the circle decreases, and thus fewer plane waves can be excited past the PBF termination plane, yielding a decrease in transmitted field. We therefore expect the fundamental-mode reflection to increase with wavelength. At the same time, since the circle boundary lies in a region where the Fourier transform intensity of the electric field is very weak (roughly 0.1% of its maximum value), we expect a weak dependence on wavelength.
This is indeed found to be the case. Figure 7 shows the reflection coefficient of the x-polarized fundamental mode, defined as |a 0|2, when the fiber is terminated in air, calculated at several wavelengths across the bandgap. The reflection coefficient remains weak across the entire bandgap, its coefficient varying between ~0.6×10-5 at the short-wavelength end of the bandgap and ~1.3×10-5 at the long-wavelength end. We also observe a general trend towards an increase in the backreflection as the wavelength increases, which was predicted earlier on the basis of the smaller wavevectors associated with longer wavelengths: the mode energy is transmitted in air over a smaller range of wavevectors, and fewer plane waves can be excited, thus resulting in an increased reflection coefficient.
4. Reflection coefficient of various PBF modes
It was also interesting to determine the reflection coefficient for the other kinds of modes supported by air-core fibers, in particular higher order core modes, surface modes, and cladding modes. To do so, we retained the same cladding holes radius (ρ = 0.47Λ) but increased the fiber core radius to R = 2.0Λ, so that the fiber now supports a large number of core modes, as well as some surface modes. As is well known, the surface-mode energy is concentrated mostly in the silica at the core-cladding boundary. The core region of this fiber is shown in Fig. 8, together with the intensity contour lines of an exemplary surface mode.
Figure 9 shows the calculated reflection coefficients of selected modes of this fiber, namely the six surface modes closest in effective index to the fundamental modes, and the first six air-guided modes (the two LP0,1-like fundamental modes, labeled HE1,1 x,y, and the four LP1,1-like first-order modes, labeled TE0,1, HE2,1 x,y, and TM0,1, respectively). These coefficients were calculated at λ = 0.6Λ, which is about at the center of the bandgap, for a termination in air, and they represent reflection into the same backward-traveling mode as the incident mode. The reflection coefficient for the fundamental mode is 1.2×10-5, which is close to the value calculated for the previous (single-mode) fiber. The higher order core modes (LP11-like) have a significantly higher reflection coefficient, averaging 7×10-5. This is expected, since these modes have a higher fraction of their energy in the silica portions of the PBF, and hence are more strongly reflected. Note that the backreflection coefficients of the four LP11-like modes differ by as much as a factor of 3. This is due to the vectorial nature of the PBF waveguide, in which the strong index contrast between the air core and the silica cladding membranes lifts the near-degeneracy of these modes normally present in weakly-guiding fibers. The surface modes, which have an even higher fraction of their energy in silica, exhibit a much higher backreflection coefficient, in the range of 8×10-3. All of these core and surface modes have very close effective indices, in the vicinity of the refractive index of air, so the Fresnel approximation (Eq. 1) yields fairly similar reflection coefficients for all them (see Fig. 9). Clearly, and expectedly, the Fresnel approximation yields poor predictions.
The reflection coefficient was also calculated for multiple representative bulk modes (also for reflection into the backward-traveling fundamental mode of same polarization). All reflection coefficients were approximately equal to 1%. This is again entirely consistent with the field distributions of these bulk modes, which are strongly localized on the thin silica veins of the photonic-crystal cladding. This feature brings their reflection coefficient up to a value of the same magnitude as a reflection from silica into air (~3.37%).
5. Effect of PBF design parameters on the fundamental core mode reflection coefficient
Unlike in a conventional fiber, the magnitude of the backreflection at the end of a PBF is strongly influenced by the fiber parameters. These parameters can therefore be adjusted to minimize backreflection. One such parameter is the core radius. A proper choice of core radius can prevent the PBF from supporting surface modes.  At the same time, the absence of surface modes has been linked to a reduction in the fraction of fundamental mode power localized in the silica regions of the PBF.  We thus expect that the reflection of the fundamental mode will be higher for core radii that support surface modes. Since larger cores support fundamental modes that are more strongly confined in the air core (i.e., the fundamental mode intensity at the core-cladding interface is weaker), we also expect that as the core radius increases, the fundamental mode reflection should generally decrease.
To confirm these trends, we plot in Fig. 10 the calculated dependence of the fundamental mode reflection coefficient on the core radius, for the same cladding holes radius as used before (ρ = 0.47Λ) and at about the center of the bandgap (wavelength λ = 0.6Λ). As expected, as the core radius increases, the reflection tends to decrease, except for radii where the PBF supports surface modes, such as for radii between 1.05Λ and 1.3Λ, and between 1.45Λ and 1.7Λ. For those radii, the fundamental mode reflection coefficient increases very sharply and significantly, again because of the substantial increase in the fraction of fundamental mode energy in silica. This result points to another reason for avoiding surface modes: they increase backreflection of the fundamental (and other) core modes.
The minimum backreflection predicted for a single-mode PBF is determined to occur for a radius of R = 0.95 Λ, and is approximately equal to 2.9×10-6. Further reduction in the backreflection can be achieved by selecting a radius R = 1.9Λ, where the backreflection is predicted to be 4.2×10-7, but at the expense of the PBF no longer being single-mode.
6. Effect of the termination medium refractive index on the reflection coefficient
While the reflections of the core modes of a PBF terminated in air are found to be low, much stronger reflections are expected for a solid or liquid termination medium (such as silica or lithium niobate, for example), because the core modes then propagate into a medium with a much greater index than their own. Since the Fresnel approximation is more accurate for a large index contrast between waveguide and termination medium, it is also expected to give better estimates of the reflection coefficients.
To investigate the influence of the termination medium index, we show in Fig. 11 the reflection of the fundamental mode for a PBF with R = 0.8Λ and ρ = 0.47Λ, and a wavelength λ = 0.6Λ, as a function of the termination refractive index. The reflections predicted by the Fresnel approximation are shown as a red curve. While both methods are in agreement for large termination medium indices, there are as expected significant differences for a termination medium index n 0 close to the refractive index of air. When n 0 is close to 1, the Fresnel approximation ceases to be valid and yields inaccurate answers, as we have seen before, by as much as 3 orders of magnitude. One interesting feature of these curves is the location of their minima. The Fresnel approximation predicts a minimum reflection for a termination medium of refractive index equal to neff < 1, while the calculated reflection exhibits a minimum for a termination medium index of approximately 1.005.
For larger values of the termination medium index, the agreement between both approaches is much improved. Notably, the calculated reflection coefficient for a silica termination is found to be 3.13%, differing by 8% with the value of 3.37% expected for a plane wave at normal incidence to an air-to-silica interface. The Fresnel approximation predicts a reflection of 3.67% for the fundamental mode at the same air-silica interface, within 17% of the computed value. This reflection coefficient is also approximately the reflection we expect from the butt-coupling of this PBF to a conventional index-guiding silica fiber, such as an SMF28 fiber.
7. Simplified formulation
Since the Fresnel reflection coefficient is a poor approximation for the modal backreflection coefficient, we developed a new approximate formula for evaluating that quantity. This approximation is based on a Fresnel coefficient in which the mode effective index is replaced by an averaged modal index. We derived the formula for the averaged modal index in a heuristic fashion, to obtain the best match with our calculation of the modal reflection coefficient.
The averaged modal index is defined as the normalized overlap of the transverse electric field intensity with the refractive index distribution:
By definition, 1 ≤ nmodal ≤ nsilica, and the larger the fraction of energy the mode carries in air, the closer the averaged modal index is to 1. We show that the modal reflection coefficient can be well approximated by:
The physical interpretation of this approximation is that, from a reflection standpoint, the light effectively propagates as a plane wave under normal incidence, in a medium of refractive index nmodal. The use of the averaged modal index also allows to account for the occurrence of a minimum in reflection for a termination medium with a refractive index larger than 1.
Figure 12 shows a plot of the averaged modal index calculated from Eq. (9) for the fundamental mode as a function of core radius for a PBF with ρ = 0.47Λ, and at a wavelength of λ = 0.6Λ. The averaged modal index also shows sharp increases whenever the PBF supports surface modes, due to an enhanced overlap of the mode with the silica regions,  as does the fundamental mode reflection coefficient (see Fig. 10). This is in particular visible for core radii between 1.05Λ and 1.3Λ, and core radii between 1.45Λ and 1.7Λ. As the core radius increases, the averaged modal index outside of these bands decreases, which is again consistent with the predicted behavior of the reflection coefficient.
In Fig. 13, we have plotted a comparison between the fundamental mode reflection coefficient obtained by our complete eigenmode decomposition calculation (blue curve), and by the averaged modal index approximation (red curve). We note that while the trends of the reflection coefficient are well predicted by this new approximation, the actual values predicted differ by a factor of 2 to 6, depending of the core radius, and that the fit has a somewhat better quality when the PBF supports surface modes. We attribute this difference mainly to the angular spread of the energy of the fundamental mode, which is quite significant in these PBFs (nearly 10° for a core radius R = 0.8Λ).
It is therefore of interest to consider PBFs where the angular spread of the fundamental mode is much reduced, so as to have a mode behaving nearly like a plane wave under normal incidence at a PBF termination. A larger air-filling ratio than those considered heretofore (around 85 %) is required for that purpose, and we elected to model the HC-1550-2 fiber by Crystal Fibre,  which has an air-filling ratio of over 90%. Figure 14 shows a scanning electron microscope picture of the PBF structure, with the entire photonic crystal cladding visible (8 layers of holes). We modeled the PBF using the refractive index structure shown in part in Fig. 15, using 5 layers of holes and periodic boundary conditions.
The longitudinal Poynting vector component of the fundamental mode at a wavelength of λ = 0.4Λ is shown in Fig. 16. There is very strong confinement of the mode field into the air core, with a field intensity overlap with silica now less than 1%. The Fourier transform intensity of this same mode's transverse electric field, plotted in logarithmic scale in the wavevector domain, is shown in Fig. 17. The black circle shows the boundary for all wavevectors such that , i.e. all wavevectors admissible for plane waves propagating in air. The energy of the fundamental mode shows a much smaller angular spread, of the order of 3°. The calculated value for the fundamental mode reflection coefficient is 1.9×10-6 for an air-termination.
The averaged modal index approximation is compared to the calculations of the reflection coefficient in Fig. 17, for the PBF HC-1550-2 at a wavelength of λ = 0.4Λ, as a function of the termination medium index. The agreement is now excellent, the disagreement between the two approaches being at most 50% for all the points shown in the figure. Also, the averaged modal index model predicts that the minimum in the reflection occurs for a termination index of 1.0036, in good agreement with the prediction of the exact model (black dots), which falls somewhere between 1.003 and 1.004. The conclusion is that the averaged modal index approximation can be used for qualitative predictions when modeling a fiber with a low air-filling ratio (around 85%), and for quantitative predictions when modeling a fiber with an air-filling ratio of 90% or higher.
We have used an eigenmode decomposition method to compute the backreflection coefficient of various modes in PBFs, for terminations in a uniform medium, since the assumptions on which the commonly used Fresnel approximation relies cease to be valid for air-core fibers. The predicted values range from 10-5 to 10-6 for the fundamental mode, a few 10-3 for surface modes, and around 10-2 for bulk modes. Furthermore we observed strong variations of the fundamental mode reflection as a function of core radius, due to the presence of surface modes, which enhance the overlap of the fundamental mode with the silica. For the Crystal Fibre HC-1550-2 PBF we predicted a reflection coefficient of 1.9×10-6 for an air termination, and 3.3% for a silica termination, which would correspond to the splice reflection of a PBF with an index-guiding conventional silica optical fiber.
We also proposed a heuristic simple reflection formula, based on an averaged modal index, and allowing the reflection of a PBF mode to be predicted accurately, within less than an order of magnitude for PBFs with a low air-filling ratio (around 85%) and more accurately for an air-filling ratio of 90% and over.
Appendix: Eigenmode decomposition method
The transverse incident electrical field is denoted EI(r) and is assumed to be a mode (core or cladding mode) of the PBF, E0(r). The reflected transverse electrical field propagates in the negative z direction and is a superposition of PBF modes (core and cladding modes), and is denoted ER(r). The modal decomposition of the reflected transverse electrical field is of the form:
where a 0 is the amplitude reflection coefficient of the incident mode considered into itself. The modes present in the decomposition of the reflected field are of the same symmetry class as the fundamental mode. The intensity reflection coefficient is:
In the same fashion, the transverse transmitted electrical field propagating in region II in the positive z direction is denoted ET(r) and decomposed in a sum of plane waves propagating in free space:
where Ω is the set of plane waves used to decompose ET, s is a normalization constant, and E1W,k and E2W,k denote the two amplitudes along orthogonal polarizations of a plane wave propagating in air, with a wavevector k=[kx ky kz] satisfying:
The transverse magnetic fields are decomposed in the same fashion:
Using the continuity of the transverse electric and magnetic fields, we obtain the following relationships:
and we define a hermitian product over the function space by:
where the integral is calculated across the transverse plane.
We can then exploit the orthogonality of the transverse fields of the waveguide modes, and the orthogonality of the transverse fields of the plane waves, to derive the reflection coefficients am and transmission coefficients bk:
These linear equations are encoded in a sparse matrix A of size (1+NM+2NW)×(1+NM+2NW), containing 1+NM+2NW+4(1+NM)NW nonzero terms. The reflection coefficients ν are determined by solving an equation of the form:
which is solved by performing an incomplete LU decomposition of A, which is used as a preconditioner to lower the conditioning number of A, and we thus solve:
which allows for faster and more accurate solving of the reflection coefficient vector through the numerical method of the stabilized biconjugate gradients method BICGSTAB, with an approximate solution ν0 with an error measured by:
thus ensuring the very high accuracy of the solution within a few iterations of BICGSTAB (fewer than 10 usually), in a few seconds of calculation.
This work was supported by Litton Systems, Inc., a wholly owned subsidiary of Northrop Grumman Corporation.
References and links
1. J. Broeng, D. Mogilevstev, S. E. Barkou, and A. Bjarklev, “Photonic crystal fibers: A new class of optical waveguides,” Opt. Fiber Technol. 5, 305–330 (1999). [CrossRef]
2. J. C. Knight, T. A. Birks, R. F. Cregan, P. St. J. Russell, and J. P. Sandro, “Photonic crystals as optical fibers-physics and applications,” Opt. Mater. 11, 143–151 (1999). [CrossRef]
3. R. S. Windeler, J. L. Wagener, and D. J. Giovanni, “Silica-air microstructured fibers: Properties and applications,” Optical Fiber Communications Conference, San Diego, CA, USA (1999).
4. V. Dangui, H. K. Kim, M. J. F. Digonnet, and G. S. Kino, “Phase sensitivity to temperature of the fundamental mode in air-guiding photonic-bandgap fibers,” Opt. Express 13, 6669–6684 (2005). [CrossRef] [PubMed]
5. D. M. Dagenais, K. P. Koo, and F. Bucholtz, “Fiber interferometry limitations due to parasitic optical cavities,” Lasers and Electro-Optics Society Conference, San Jose, CA USA (1993).
6. J. Corbett, A. Dabirian, T. Butterley, N. A. Mortensen, and J. R. Allington-Smith, “The coupling performance of photonic crystal fibres in fibre stellar interferometry,” Mon. Not. Roy. Astron. Soc. 368, 203–210 (2006).
7. Crystal Fibre website, http://www.crystal-fibre.com
8. D. Khalil, “Reflection at the end of strongly guiding dielectric waveguide,” Opto-Electronics and Communications Conference, Yokohama, Japan , 1, 352–353 (2002).
9. P. Gerard, P. Benech, D. Khalil, R. Rimet, and S. Tedjini, “Towards a full vectorial and modal technique for the analysis of integrated optics structures: The Radiation Spectrum Method (RSM),” Opt. Commun. 140, 128–145 (1997). [CrossRef]
10. D. Marcuse, Theory of dielectric optical waveguides, (Academic Press, 1974).
12. H. . van der Vorst, “BI-CGSTAB: A fast and smoothly converging variant of BI-CG for the solution of nonsymmetric linear systems,” SIAM J. Sci. Stat. Comput. 13, 631–644 (1992). [CrossRef]
13. H. K. Kim, J. Shin, S. H. Fan, M. J. F. Digonnet, and G. S. Kino, “Designing air-core photonic-bandgap fibers free of surface modes,” IEEE J. Quantum Eectron. 40, 551–556 (2004). [CrossRef]
14. D. C. Allan, N. F. Borrelli, M. T. Gallagher, D. Müller, C. M. Smith, N. Venkataraman, J. A. West, P. Zhang, and K. W. Koch, “Surface modes and loss in air-core photonic band-gap fibers,” in Photonic Crystal Materials and Devices, A. Adibi, A. Scherer, and S. Yu Lin, eds., Proc. SPIE 5000,161–174 (2003). [CrossRef]