Design optimization of a novel integrated bi-directional (BiDi) triplexer filter based on planar lightwave circuit (PLC) for fiber-to-the-premise (FTTP) applications is described. A multi-mode interference (MMI) device is used to filter the up-stream 1310nm signal from the down-stream 1490nm and 1555nm signals. An array waveguide grating (AWG) device performs the dense WDM function by further separating the two down-stream signals. The MMI and AWG are built on the same substrate with monolithic integration. The design is validated by simulation, which shows excellent performance in terms of filter spectral characteristics (e.g., bandwidth, cross-talk, etc.) as well as insertion loss.
© 2006 Optical Society of America
Bi-directional triplexer optical filters capable of separating the three designated wavelength channels, i.e., 1310nm for the up-stream signal, 1490nm and 1550nm for the down-stream signals, are key components for the optical network units (ONUs) in the passive optical networks (PON) for the fiber-to-the-premises (FTTP) applications. The existing triplexer optical transceivers for the FTTP PON networks are built on the traditional discrete optical filter devices based on thin film technologies to perform the triplexer functions . Such an approach has been quite successful in producing triplexer optical filters with high performance. On the other hand, however, the traditional discrete optical device technologies require complex and labor-intensive optical assemblies, which is not suitable for mass production and cost reduction. Mass production of low-cost optical components such as the triplexer optical transceivers used for the ONUs is critical for the wide-spread deployment of the FTTP technologies in the access networks .
Photonic integration based on the planar lightwave circuit (PLC) is a promising solution to overcome the shortcomings of the traditional discrete optical device technologies, as it leads to a much smaller chip count and therefore is more amenable to auto-assembly and mass production at lower cost . This is of particular importance for the FTTP application as the cost and volume of the components become key factors for commercial deployment.
Several schemes for the triplexer filters based on PLC have been reported –, in which the devices proposed by Li, et al  and by Lang et al  are similar in terms of underlying operation principle, despite the fact that one is based on waveguide diffraction gratings and the other on arrayed waveguide gratings (AWG). These two designs can successfully de/multiplex three up and down-stream channels. In practice, however, these structures are difficult to meet the bandwidth specifications required by the industrial standards for the passive optical networks (PON), which calls for Δλ=100nm, 20nm, and 10nm, for wavelengths λ=1310nm, 1490nm, and 1555nm, respectively . The triplexer filters proposed by Shen et al  and by Song et al , using deeply etched or UV exposed Bragg gratings to demultiplex up and down-stream channels and a coupler to further demuliplex the two downstream channels, have similar bandwidth problem, especially the long and weak UV formed gratings which yield narrow bandwidth intrinsically. Furthermore, the device reported in  uses λ=1490nm as up-stream channel and it appears to be difficult to change the up-stream channel to λ=1310nm as required by the ITU-T G.983 since the two down-stream channels at λ=1490nm and 1555nm are too close to separate by the half/full cycle idea.
Considering the stringent requirements of the ITU-T G.983, we have recently briefly reported a novel idea with combined coarse and dense WDM . In the design, monolithically integrated multi-mode interference (MMI) device and array-waveguide grating (AWG) are used to perform the coarse and dense WDM. Initial simulation result shows that it can achieve different bandwidths for different channels. In this paper, we give detail description of the design principle and design process as well as design optimization to reduce the insertion loss and improve the overall performance.
It should be acknowledged that another PLC based triplexer filter was reported after the submission of this paper. The new design cascades a number of asymmetric Mach-Zehnders and directional couplers. Similar to what we proposed in the paper, it yields different bandwidths for different channels as required by the ITU-T G.983.
2. Overall design
In view of the challenging bandwidth specifications for the triplexer filters as required by the standard PON networks, we proposed a “divide and conquer” strategy by separating the coarse and dense WDM functions. The implementation of such strategy is schematically shown in Fig. 1. It consists of a 2×2 MMI a 1×2 AWG, both of which have been investigated extensively in literature for a wide range of applications , . The MMI device functions as a coarse WDM filter to de/multiplex the up-stream signal at λ=1310nm and the down-stream signals at λ=1490nm and 1555nm. The AWG serves as a dense WDM filter and de/multiplexes the down-stream signals and routes them into two different photo detectors.
In addition to achieving different bandwidths, the bi-directional design can also effectively minimize the cross-talk between the down-stream and up-stream signals. Among the channel cross-talks, the up-stream laser signal received by the detectors is most harmful for the system performance. The system specification requires that ONU has the maximum mean launched power of 7dBm, the minimum sensitivity of -28dBm and the minimum extinction ratio of 10dB. Hence, the isolation in 1490nm against 1310nm should be higher than 45dB. Due to the low reflection of the MMI as well as the multiple filtering of both the MMIs and the AWG, the laser signal feed into the detectors is extremely weak.
3. Material and waveguide
To demonstrate the design concept, we choose the Hydex™, a high index contrast material developed at Little Optics , though the design can be realized on any other material systems suitable for planar waveguides. The core used in the design has 17% index contrast with respect to the SiO2 cladding. In the following, we will investigate the single mode condition and the bending loss. They are two most important issues in the design. The single mode operation is an essential condition for an AWG, and the bending loss determines the minimum bending radius and ultimately the device size.
3.1. Single mode condition
To maintain single mode, the high order modes has to be cutoff. For a channel waveguide, single mode condition can be judged in two ways: the first high order mode leaks and its effective index is lower than the refractive index of the cladding. Using the APSS , we calculate both the real effective index and the leaky loss, which related to the imaginary part ni of the complex effective index via
Shown in Fig. 2 are the leaky loss and the effective index of the first high-order mode, which has an anti-symmetric field pattern, as a function of the waveguide width and height. Because the single mode condition is critical for the AWG, the wavelength used in the calculation is 1480nm to ensure the waveguide is single mode for the shortest wavelength going through the AWG.
Judging from the effective index, the first higher order mode is cut off at about 1.25µm. From the leaky loss, it is also found that the mode becomes leaky when the width and height is smaller than 1.25µm. To be safe, 1µm will be used in following design and simulation.
3.2. Minimum bending radius
Waveguide bend is inevitable in most PLC design and bending loss is generated naturally due to the radiation at the bend. The device will be not functional if the bending loss is too high. Therefore, we need to determine the minimum bending radius for any waveguide design, which is also one of the main limiting factors for the maximum feature length of the overall devices.
Using the APSS, we could calculate the bending mode in the cylindrical coordinator system  and the leakage loss of the 1µm×1µm channel waveguide at different bending radii is shown in Fig. 3. The wavelength used in the calculation is 1560nm, which is the case with maximum bending loss in comparison with other wavelength channels as the mode is least confined at this wavelength channel. For this worst scenario, it is found that the waveguide can be bent as small as 50µm radius with negligible bending loss. Therefore, R>50µm will be used as one of the design rules.
4. Coarse WDM
A 2×2 MMI is chosen to perform the coarse WDM to separate the up- and down-stream signals. Its working mechanism is similar to, but more robust and simpler than a directional coupler. The coupling length of a directional coupler is determined by the widths of the two waveguides and the gap, particularly, between them, while only the width of is a factor for the MMI. Furthermore, same fabrication tolerance, say 100nm, maybe significant for the directional coupler, but negligible for the MMI since the width of the MMI is much larger than the widths and gap of the directional coupler.
However, a standard MMI without any optimization cannot give desired spectral response in terms of spectral flatness as shown in Fig. 4.
To realize flat channel characteristics, tapering the input/output ports is an effective approach, in addition to optimization of the MMI width and length. Shown in Fig. 5 is the spectral response of an optimized 2×2 MMI. The insertion losses at the central wavelength λ=1310nm and the 100nm span around the central wavelength, i.e., at λ=1260nm and 1360nm, are 0.15dB and 0.6dB, respectively. Therefore, the passing band is very wide and flat. The optimized MMI is 4.7µm wide, 195µm long, with 127.5µm long taper.
By examining the field pattern of laser input at λ=1310nm and the field pattern of fiber input at λ=1555nm, as shown in Fig. 6 and Fig. 7, we found that some light with power level of about -40dB as read from the Fig. 5 and carried mainly by leaky higher order modes, still goes into another port. This will not lead to any crosstalk for the end-to-end triplexr filter performance as these modes will leak out before reaching the output ports due to the single mode condition of the waveguide.
5. Dense WDM
Considering the relative wide bandwidth of the two down-stream signals and large wavelength spacing between them, as well the extra filtering of the MMI, we design an unconventional AWG with a very wide free spectral range (FSR), in the range of 200nm. Therefore, the path difference between two adjacent array waveguides, defined by
is extremely small, around 6µm only. As a result, the titled angle of the two star couplers cannot be too big and only a limited number of array waveguides can be fit in between.
Having tried a number of different combinations, we finally come up the following layout as shown in Fig. 8. In the final design, the star couplers are tilted at 30°, separated by 800µm, and only 30 array waveguides can be fit in between. The path difference between adjacent array waveguides is 5.87µm.
Because of the limited number of arrays, the insertion loss is expected to be high since only the part of the input light diffracted by the first star coupler can be captured by the arrays. To reduce the insertion loss of the first star coupler, we made the following optimizations:
1) Taper the input waveguide as shown in the inserted chart A to ensure most diffracted light is distributed in the central part in the array region, as even as possible. After a careful optimization with proper balance of the overall performance, we choose 4µm as the final taper width.
2) Taper the arrays to the maximum, as shown in the inserted chart B, in order to capture all diffracted light within the array region. Although perfect zero gap cannot be realized, it can be approached by better fabrication technology.
To reduce the insertion loss of the second star coupler, we need to suppress the first order diffraction and enhance the 0-th order diffraction so that more light could be focused on the output waveguides. For this purpose, we can taper the array waveguides by successively reducing the width of the waveguide as they approach the second star coupler as shown in the inserted chart C. Such an approach yields remarkable results due to the fact that the narrower waveguide provides weaker guidance and expand the beam size even bigger than wider waveguide. The dependence of the final result on the taper width has been investigated and shown in Fig. 9 are peak powers from the output waveguides as a function of taper width.
Although zero taper width gives the best results, we still prefer 3µm wide taper considering the fabrication tolerance issue. Also, the difference is not significant and it is only about 0.2dB.
To meet the different bandwidth requirements of the two down-stream channels, we design the two output waveguides with different widths as shown in the inserted chart D in Fig. 7. Based on the AWG operating principles, the star coupler acts as a Fourier lens and the spatial width at output plane is directly proportional to spectral bandwidth. In the final design, the widths of the two output waveguides are 6µm and 4µm, respectively. The waveguide spacing, which is also proportional to the spectral spacing, is 11.8µm.
It is unnecessary to taper the output waveguides since that all the received power by the output waveguide can be transferred to the detectors, regardless it is guided in fundamental mode or high order modes. Due to the superposition of multi-modes, the spectrum will be wide and flat as shown in Fig. 10, which is the simulation result of the AWG with the optimized parameters.
As indicated, the optimization can reduce the insertion loss. The best result we obtained so far is about 0.35dB for both output channels within the 20nm and 10nm wavelength spans. The peak power is about 0.25dB for both channels. Hence, the spectrum is flat and the variation is only about 0.1dB within the desired bandwidth. As a reference, the typical loss is about 7dB at the peak for a similar AWG without any optimization.
6. Circuit integration
By combining the optimized MMI and AWG, we could form an integrated circuit as shown in Fig. 11. The total size is within 1×2mm.
Performing circuit simulation the APSS, we obtained the end-to-end spectral response of the entire triplexer filter as shown in Fig. 12. In addition to desired spectra around 1550nm, it is also found that the power to detectors around 1310nm is about -15dB, which implies that the backward reflection at the chip-fiber interface could be very harmful and AR coating of the chip facet is necessary.
The key performance parameters related to the proposed and optimized designs are summarized in the Table 1 for the sake of clarity.
As mentioned earlier, the cross-talk from the laser to the detectors is one of our main concerns. To examine this issue, we have utilized the bi-directional beam propagation method in the APSS and will be able to calculate the reflection of MMI and, eventually, the transmitted power from the laser to the detectors as shown in Fig. 13. It is found that the cross-talk is about 45dB, which meets the system requirement.
7. Other design considerations
So far, we have demonstrated the basic working principle of the triplexer filter, as well as design process and design optimization. However, to design a practical filter, several other aspects, such as thermal sensitivity, sensitivity and tolerance of the key design parameters again fabrication uncertainties, polarization sensitivity, as well as coupling efficiency with fiber and laser diode, etc., must be considered as well. These issues will be addressed briefly here, with focus on the AWG since it is more sensitive than the MMI. Detail analysis goes beyond the scope of this paper and will be studied and reported in the future.
7.1 Thermal sensitivity
The recommended operation temperature for the triplexer filter is -40°C~+85 °C, which is ±62.5°C around room temperature. Such wide temperature range may shift the central wavelength and affect the filtering performance. Although an analysis of the exact impact of temperature for the given design requires delicate modeling and simulation, we may provide an estimate the effect based on simplified model.
According to the experiment, 7.93×10-6/K is the measured thermal optical coefficient of silica . The refractive index change caused by the temperature fluctuation of ±62.5°C is about δn=±0.0005. The central wavelength of the AWG, defined by
will shift by
Here Δ=5.87µm is the path difference, m=6 is the diffraction order. Therefore, the temperature fluctuation of ±62.5°C causes about δλc=±0.5nm wavelength shift. Considering the narrowest bandwidth is 10nm, 0.5nm shift maybe acceptable. We can also design the filter with even wider spectra, to cope with the shift.
7.2. Sensitivity and Tolerance
The fabrication tolerance for the AWG can also be estimated by above Eq. (4). With stepper optical lithography for Hydex waveguide, ±0.1µm tolerance can be achieved easily as it was reported very recently . Based on the modal calculation, the effective index change caused by ±0.1µm variation of the waveguide width and height is about ±0.0006. Translated into the wavelength shift, it is about ±0.6nm, which is not very significant compared with the thermal drift.
7.3 Polarization dependence
Polarization dependence is a common problem for most waveguide-based devices and there is no exception for the proposed triplexer filter. Although the arrayed waveguided and the connection waveguides are polarization independent square channel waveguide, the MMI coupler and the star couplers, however, break up the balance between vertical and horizontal confinement and become polarization dependent. Residue stress induced birefringence, which is very common for most materials, is not very significant for Hydex based on the recent report . Shown in Fig. 14 and Fig. 15 are polarization dependent outputs for the MMI and the AWG, respectively.
It is observed that polarization dependence is quite significant, especially for the AWG, for which the central wavelength shift between TE and TM is about 15nm and the amplitude difference is about 1.2dB. Since AWG is a very popular and practical device, there are many different ways to overcome the polarization dependence, such as birefringence compensation . In this sense, the problem of polarization sensitivity can be overcome in a similar fashion. We are now in the process of investigating several possible schemes in the context of monolithical integration. The work on polarization independence design will be subject of our future publication.
7.4. Coupling efficiency
Compact waveguide size due to high index contrast has definite advantage for large scale photonic integration. In the meanwhile, however, it also leads to difficulties in facilitating efficient coupling of light between the PLC chip and the optical fiber and the laser diode, due to the modal size mismatch. This is indeed a commonly encountered problem in photonic integration and has been studied extensively. A spot size converter (SSC) developed at Little Optics can reduce the coupling loss between Hydex™ waveguides and optical fiber to 0.75dB from typical 10~12 dB butt-to-butt coupling . Among various other types of SSC, the one with double co-cores is more suitable for chip-fiber coupling since the waveguide by the Hydex™ is very similar to a silicon wire . As for the chip-laser coupling, the SSC with abrupt junction may be more suitable since the laser modal size is small and flat in general . Further investigated is required to obtain a more definite answer to the question of the efficient off-chip coupling.
We have designed a compact, integrated bi-directional triplexer filter for the FTTP applications based on the planar lightwave circuit. The design idea, design procedure, and the design optimization have been described in details. Simulation results show that proposed design can achieve spectral requirements in terms of bandwidth and spectral isolations for the up- and down-stream wavelength channels defined for the PON systems in ITU-T G983 and G.984.
The This work is supported in part by Ontario Photonics Consortia through Ontario Research and Development Challenge Fund (ORDCF) and in part by Canada Institute for Photonics Innovation (CIPI). The author Chenglin Xu would like to thank Dr. Linping Shen at Apollo Inc. for his assistance on using Apollo’s design tool.
References and Links
1. M. N. Ranson, “Fiber-to-the-home in a competitive era,” Optical Fiber Communication Conference26–27 (1997). [CrossRef]
2. N. Frigo, K. Reichmann, and P. Lannone, “Whatever happened to fiber-to-the-home?” Optical Fiber Communication Conference280–281 (2003).
3. Meghan and Fuller, “Price pressure rules PON transceivers,” Lightwave 21, 24–26 (2004).
4. L. P. Shen, C. L. Xu, and W.-P. Huang, “Integrated grating-assisted coarse/dense WDM multiplexers,” in Optical components and Devices, Proc. SPIE5577, 698–706 (2004). [CrossRef]
5. Xun Li, Gui-Rong Zhou, Ning-Ning Feng, and Wei-Ping Huang, “A novel planar waveguide wavelength demultiplexer design for integrated optical triplexer transceiver,” IEEE Photon. Technol. Lett. 17, 1214–1216 (2005). [CrossRef]
6. C. Xu, L. Shen, D. Zhou, W.-P. Huang, and J. Hong “Novel design of bi-directional triplexer based on PLC,” Photonic application in devices and communication systems, Proc. SPIE5970, (2005).
7. J. H. Song, J. H. Lim, R. K. Kim, K. S. Lee, K. Y. Kim, J. Cho, D. Han, S. H. Jung, Y. Oh, and D. H. Jang, “Bragg grating-assisted WDM filter for integrated optical triplexer transceivers,” IEEE Photon. Technol. Lett. 17, 2607–2609 (2005). [CrossRef]
8. T. Lang, J. J. He, and S. He, “Cross-Order Arrayed Waveguide Grating Design for Triplexers in Fiber Access Networks,” IEEE Photon. Technol. Lett. 18, 232–234 (2006). [CrossRef]
9. ITU-T G.983.3, “A broadband optical access system with increased service capability by wavelength allocation.”
10. W. Chen, B. Little, W. Chen, S. Chu, J. Hryniewicz, D. Gill, O. King, F. Johnson, R. Davidson, K. Donovan, and J. Gibson, “Compact, low loss chip scale triplexer WDM filter,” Optical fiber communication conference, post deadline paper, (2006).
11. L. B. Soldano and C. M. Pennings, “Optical multimode interference devices based on self-imaging: principles and applications,” J. Lightwave Technol. 13, 615–627 (1995). [CrossRef]
12. M. K. Smit and C. van Dam, “PHASAR-Based WDM-Devices: Principles, Design and Applications,” IEEE J. Sel. Topics Quantum Electron. 2, 236–250 (1996). [CrossRef]
13. ITU-T G.984.2, “Gigabit-capable passive optical networks (GPON): Physical media dependent (PMD) layer specification.”
14. Brent E. Little, “A VLSI photonics platform,” Optical Fiber Communication Conference 444 (2003).
15. APSS, Apollo Photonics Solution Suite, Apollo Inc., Hamilton, Ontario Canada.
16. N.-N. Feng, G.-R. Zhou, C. Xu, and W.-P. Huang, “Computation of full-vector modes for bending waveguide using cylindrical perfectly matched layers,” IEEE J. Lightwave Technol. 20, 1976–1980 (2002). [CrossRef]
17. Y. Kokubun, S. Yoneda, and S. Matsuura, “Temperature-independent filter at 1.55µm wavelength using a silica-based athemal waveguide,” Electron. Lett. 34, 367–368 (1998). [CrossRef]
18. C. K. Nadler, E. K. Wildermuth, M. Lanker, W. Hunziker, and H. Melchior, “Polarization insensitive, low-loss, low-crosstalk wavelength multiplexer modules,” IEEE J. Sel. Topics in Quantum Electron. 5, 1407–1412 (1999). [CrossRef]
19. T. Tsuchizawa, K. Uamada, H. Fukuda, T. Watanabe, J. Takahashi, M. Takahashi, T. Shoji, E. Tamechika, S. Itabashi, and H. Morita, “Microphonics devices based on silicon microfabrication technology,” IEEE J. Sel. Topics in Quantum Electron. 11, 232–240 (2005). [CrossRef]
20. T. Nakamura, C.L. Xu, X. Li, and W.-P Huang, “A Spot-Size Transformer Based on Mode Interference,” IEEE Photon. Technol. Lett. 10, 325–327 (1998). [CrossRef]