This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2017.2678165, IEEE Photonics Journal
End-view Image Processing Based Angle Alignment Techniques for Specialty Optical Fibers Li Shen1†, Lin Gan1†, Zhuoran Dong1, Borui Li1, Deming Liu1, Songnian Fu1, Weijun Tong2, and Ming Tang1, Senior Member, IEEE 1Wuhan
National Lab for Optoelectronics (WNLO) & National Engineering Laboratory for Next Generation Internet Access System, School of Optics and Electronic Information, Huazhong University of Science and Technology, Wuhan 430074, China 2State Key Laboratory of Optical Fiber and Cable Manufacture Technology, Yangtze Optical Fiber and Cable Joint Stock Limited Company (YOFC). R&D Center, Wuhan 430073, China
Abstract: We proposed two end-view based digital image processing algorithms for high precision rotational angle alignment of specialty optical fibers. The first technique is based on the cross-correlation calculation between images of two fiber ends. Accurate alignment can be achieved between identical specialty optical fibers with arbitrary end face structures. The other one is based on the self-correlation method. Two fibers with different refractive index profiles in the fusion splicer can be rotated to pre-set angles independently with high precision. Accurate angle alignment of less than 0.5 degree’ uncertainty and low average splicing loss for 7-core multi-core fibers (MCFs) have been achieved in experiment. To speed up the alignment process, we further proposed two simplified algorithms for the abovementioned techniques. The time consumption can be reduced by about 50% with acceptable alignment accuracy. We confirmed that all the methods are suitable for other kinds of specialty optical fibers, such as polarization maintaining fibers (PMFs) and photonic crystal fibers (PCFs). Corresponding author: Ming Tang (email:
[email protected]). Index Terms: End-view, rotational angle alignment.
1. Introduction Specialty optical fibers, such as multi-core fibers (MCFs), polarization maintaining fibers (PMFs), and photonic crystal fibers (PCFs), are now applied in a variety of fields, such as optical communication, fiber sensing and fiber laser [1-6]. High quality fusion splicing for specialty optical fibers is of essential importance in real applications. Due to complex end-face structures embedded in MCFs, PMFs, PCFs and other kinds of specialty optical fibers, the fusion splicing is different from conventional single-mode fibers and additional rotational angle alignment with high precision is necessary [7]. Recently, a side-view based alignment method for MCF has been demonstrated [8,9]. The alignment angle is estimated by analyzing the side-view image from different angles. Since at least one turn rotation has to be made, it is time consuming. More importantly, with the increasingly complex structure of the fibers, the side-view features will become indistinct so that it may not be possible to achieve reliable angle alignment. On the other hand, with the rapid development of fusion splicing techniques, most of fiber processing platforms are equipped with end-view functions, such as Fujikura FSM-100P+, Vytran GPX3400, et al [7]. Applying advanced digital image processing with the end-view function will be promising to achieve high precision angle alignment for the fusion splicing. In this paper, we proposed and demonstrated two automatic alignment methods with high accuracy based on the end-view image processing techniques for MCFs. The time consumption has been further reduced using simplified algorithms with acceptable accuracy. Finally, we confirmed that our methods are suitable for other specialty optical fibers like PMFs and PCFs, etc.
2. End-view based alignment methods We developed two alignment methods based on the end-view technique: the cross-correlation method and the self-correlation method. Fig.1(a) shows the schematic of the cross-correlation method, which is applicable to fibers with identical end-face structures. We take the 7-core MCF as an example. Step 1, the end-view images of left and right side of fibers are acquired from the fusion splicer (Fujikura FSM-100P+). Step 2, after converting the RGB images to binary images, we can extract the edge of the cladding through Canny edge detection algorithm [10], then the center and radius of the edge circle can be fitted by Hough circle transform [11]. Step 3, converting the RGB images to grayscale, resize the grayscale images based on the parameters obtained in step 2 to locate the cladding in the center of the image. Step 4, rotate the right-side fiber image clockwise using bilinear interpolation method and calculate the correlation coefficient between the images of rotating fiber and fixed fiber. We assume that the high precision angle alignment can be realized when the coefficient reaches maximum. The correlation coefficient c(θ) at rotational angle θ is computed by [12]:
𝑐(𝜃) = Where
𝑛 ̅ ̅ ∑𝑚 𝑖=1 ∑𝑗=1(𝐴𝜃 (𝑖,𝑗)−𝐴)×(𝐵(𝑖,𝑗)−𝐵 ) 𝑛 𝑚 𝑛 ̅ 2 ̅ 2 √(∑𝑚 𝑖=1 ∑𝑗=1(𝐴𝜃 (𝑖,𝑗)−𝐴) )×(∑𝑖=1 ∑𝑗=1(𝐵(𝑖,𝑗)−𝐵 ) )
(1)
𝐴𝜃 (𝑖, 𝑗) and 𝐵(𝑖, 𝑗) are the gray values of raw 𝑖 column 𝑗 of the rotated right fiber image at rotational angle θ
1943-0655 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2017.2678165, IEEE Photonics Journal
and left fiber image, respectively. 𝐴̅ and 𝐵̅ are the average values of 𝐴𝜃 (𝑖, 𝑗) and 𝐵(𝑖, 𝑗), 𝑚 and 𝑛 are the number of pixel rows and columns. It is worth to note that the procedure for other kinds of specialty optical fibers will be the same.
Fig. 1. (a) Schematic of the cross-correlation alignment method for identical 7-core MCFs splicing. (b) Schematic of the self-correlation alignment method for two types of 7-core MCFs splicing.
The other method is called self-correlation method which realizes angle alignment by rotating the left and right fibers independently to arbitrarily defined positions. This method is able to conduct the fusion splicing between fibers with different end-face index profiles. Fig. 1(b) shows the schematic of the self-correlation method for two different types of 7-core MCFs. Step 1 and step 2 are similar with the cross-correlation method. Step 3, the end-view images are resized and then reversed upside down. Step 4, rotate the left and right fiber grayscale images clockwise individually and reverse them. For left or right fiber at different angles, self-correlation coefficients between the rotated images and their reverse are calculated individually using the same way in the cross-correlation method. However, the self-correlation method induces some ambiguities during angle positioning because any horizontally symmetrical image distribution produces correlation peak. As an example shown in Step 4 of Fig.1 (b), rotating images at two positions have the same peak value but different angles. In order to solve this issue, we conduct the spatial frequency analysis by Fourier transform of the grayscale distribution along the blue horizontal dividing line. It’s clear that position 1 has more high-frequency component than position 2. Finally, the two fibers can be aligned in the same position. There is a difference between the cross-correlation method and the self-correlation method. For the cross-correlation method, we need to calculate the correlation coefficient in a rotational angle range of 0-360 degrees to get the angle aligned; but for the self-correlation method, the calculating angle range can be reduced to 0-180 degrees for the left and right fiber. Because the coefficients of two angles with a 180 degrees’ difference are identical based on the calculating algorithm for the self-correlation method. Although geometric structures of two MCFs in Fig. 1(b) are quite similar, the dissimilar refractive index profiles may lead to a relatively low cross-correlation coefficient hence poor alignment accuracy. Self-correlation method in this case can avoid this problem. 1943-0655 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2017.2678165, IEEE Photonics Journal
3. End-view alignment results for MCFs To investigate the accuracy of our methods, we prepared the MCFs as shown in Fig.2. The MCFs have seven cores designed with G657.B3 profile. The cladding diameter and the core pitch are 150 μm and 42 μm, respectively.
Fig. 2. End-view image of the 7-core MCF.
Fig.3(a) shows the relationship between rotational angle and correlation coefficient using the cross-correlation method. There are 6 peaks with coefficients higher than 0.95, and the highest correlation coefficient is 0.981. The peaks are in a period distribution with peak angle values marked on the figure. We can use the angle uncertainty to evaluate the accuracy of the alignment method. The angle uncertainties are the difference between the theoretical angles and the calculated ones. The theoretical angles are obtained by using the first peak angle value as the datum and assuming all the angles are completely uniformly distributed. For the cross-correlation algorithm, angle uncertainties are less than 0.2 degree. The results of the self-correlation method are shown in Fig.3(b). The red and blue lines represent the results of left and right fibers, respectively. Because the MCFs has 6 axes of symmetry in a range of 180 degrees, there are 6 peaks in the relationship curves for each side of the fiber, and the highest correlation coefficients of left and right fibers are 0.986 and 0.979.
Fig. 3. (a) Calculated relationship between rotational angle and correlation coefficient using the cross-correlation method for a 7-core MCF. (b) Calculated relationship between rotational angle and correlation coefficient using the self-correlation method for the same 7-core MCF.
The results of the two methods match very well. For example, in the self-correlation method, left fiber rotates for 45.5 degrees clockwise and right fiber rotates for 59.0 degrees clockwise corresponding to the first peak. This operation is equivalent to rotate the right fiber for 13.5 (59.0-45.5=13.5) degrees clockwise, which is very close to the peak angle of 13.4 degrees in the cross-correlation method. This result indicates that both cross-correlation method and self-correlation method can achieve high accuracy.
Fig. 4.
Measured splice loss using the cross-correlation method for a 7-core MCFs.
Fig.4 shows the measured splicing loss of the MCFs using the cross-correlation method developed with C# platform and swing electrode system [13]. Four fusion splices were conducted, and the insertion losses were measured. The fiber cleaving angle for each splice is under 0.5 degree to ensure the splice quality, and the rotational angle uncertainty is no more than 0.5 degree. The insertion losses of each splice are marked by different colors and shapes. Core number 1-7 represent different cores, and core number 1 is the center core. The average losses for the central core and the outer cores are 0.058 dB and 1943-0655 2016 IEEE. Translations content mining are permitted for academic research Personal use is alsothe permitted, but republication/redistribution requires IEEE permission. See such 0.295 dB (c)respectively, andandthe crosstalk is usually below -60only. dB. Actually splicing loss relates to many other factors http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2017.2678165, IEEE Photonics Journal
as cleaving angle and splicing conditions. The ±1 um uncertainty of the outer core pitch may lead to a maximum loss of 0.7 dB [14]. Therefore, the results proved that our methods could align the rotational angle with high precision in practical applications.
4. Simplified method for MCFs Because an end-view image has tens of thousands of pixels, it requires a large number of image rotating and correlation operations in the cross-correlation and the self-correlation methods. We thus proposed a simplified method to speed up the alignment procedure. Fig.5 (a) shows the principle of the simplified method. We use the gray value distribution on the circumference of the grayscale image for calculation instead of using the whole image. The red and blue circle are circumferences 1 and 2. The radius of circumference 1 and circumference 2 are 33 pixels and 70 pixels, respectively. Circumference 1 is chosen to have the largest gray value difference, and the measured gray value distribution of the two circumferences are shown in Fig.5 (b). We can observe a periodic distribution in circumference 1, but there is no clear pattern in circumference 2. The circumference with greater gray value difference will have more regular gray value distribution and can be used for alignment. In the simplified method, we averaged the gray value of three circumferences with maximum gray value difference to guarantee the accuracy. The image rotating is replaced by the circle shift of gray value distribution and the corresponding coefficient calculation between two grayscale images is also substituted by two gray value distributions. After 100 times calculations, the averaged time consumptions from acquiring the end-view image to getting the precise alignment angle have been recorded. The time consumptions of the cross-correlation method and the self-correlation method are 6.86 seconds and 7.39 seconds, respectively. The time consumptions for the two simplified methods are 3.80 seconds and 2.80 seconds, with about 45% and 67% reduction, respectively.
Fig. 5. (a) Principle of the simplified method. (b) Measured gray value distribution on circumference 1 and 2.
Fig.6 (a) shows the results of the simplified cross-correlation method for the same 7-core MCF. It can be seen that the shape of the relationship curve between rotational angle and correlation coefficient is basically the same as the original method, and the highest coefficient is 0.956. We can compare the peak angles with the original cross-correlation method to assess the accuracy, the maximal angle difference among all the peak angles is no more than 0.1 degree. The negative correlation coefficient means the two images have negative liner correlation. Fig.6 (b) shows the results using the simplified self-correlation method. The highest correlation coefficient is 0.945 and the biggest peak angle difference with the original self-correlation method does not exceed 0.1 degree. The results prove that the simplified method can be accurate and efficient.
Fig. 6. (a)Calculated relationship between rotational angle and correlation coefficient using the simplified cross-correlation method for a 7-core MCF. (b) Calculated relationship between rotational angle and correlation coefficient using the simplified self-correlation method for a 7-core MCF.
5. End-view alignment method for other specialty optical fibers End-view based alignment methods developed in this work can also be used for other kinds of specialty optical fibers, such as PMFs and PCFs, etc. Fig.7 shows the end-view images of PMFs and PCFs. 1943-0655 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2017.2678165, IEEE Photonics Journal
Fig. 7. End-view images of the PMFs and PCFs.
Fig.8 (a) and (b) shows the results of PMF alignment using all of the methods mentioned above. For the cross-correlation method and the self-correlation method, the angle differences between the simplified method and the original method are 0.3 degree and 0.2 degree. The angle difference between cross-correlation method and self-correlation method is no more than 0.5 degree. Fig.8 (c) and (d) shows the corresponding results of PCF. The angle difference between the simplified method and the original method is no more than 0.4 degree, and the angle difference between the cross-correlation method and the self-correlation method is no more than 0.2 degree. It revealed that all the methods we proposed can be used for rotational angle alignment of PMF and PCF with high precision by using the end-view image processing technique. We also did an experiment on the PMF and PCF. For the PMF, the polarization extinction ratio after splicing is about 35 dB, which is the same with PMF without splicing. For the PCF, the average splice loss is 0.29 dB, the splice loss can be further reduced by optimizing the splicing parameters.
Fig. 8. (a)Calculated relationship between rotational angle and correlation coefficient using original and simplified cross-correlation method for a PMF. (b) Calculated relationship between rotational angle and correlation coefficient using original and simplified self-correlation method for a PMF. (c)Calculated relationship between rotational angle and correlation coefficient using original and simplified cross-correlation method for a PCF. (d) Calculated relationship between rotational angle and correlation coefficient using original and simplified self-correlation method for a PCF.
6. Conclusions Universal rotational angle alignment algorithms based on the fiber end-view image processing technique have been proposed and developed with high precision for high quality fusion splicing of MCF, PMF and PCF. The detailed digital image processing algorithms and experimental verifications are comprehensively presented and discussed. Simplified alignment versions are further conducted to achieve similar accuracy while with much less time consumption. With a high-resolution end-view imaging device and a fiber-rotating module, our method can realize efficient and precise alignment for a variety of fibers, which is of essential importance for high quality special fiber splicing. 1943-0655 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2017.2678165, IEEE Photonics Journal
Acknowledgements This research was supported by the National 863 High-tech R&D Program of China under Grant 2015AA016904, the National Natural Science Foundation of China under Grant 61331010, 61205063, 61290311, and the Program for New Century Excellent Talents in University (NCET-13-0235).
†These authors contributed equally to this work.
References [1] Z. Feng et al., “Multicore-fiber-enabled WSDM optical access network with centralized carrier delivery and RSOA-based adaptive modulation,” IEEE Photon. J., vol. 7, no. 4, Aug. 2015, Art. ID 7201309. [2] B. Li et al., “Experimental demonstration of large capacity WSDM optical access network with multicore fibers and advanced modulation formats,” Opt. Exp., vol. 23, no. 9, pp. 10997–11006, Apr. 2015. [3] L. Gan et al., “Spatial-Division multiplexed Mach–Zehnder interferometers in heterogeneous multicore fiber for multiparameter measurement,” IEEE Photon J, vol. 8, no. 1, pp. 1-8, Feb. 2016. [4] R. Wang et al., "Spatially arrayed long period gratings in multicore fiber by programmable electrical arc discharge," IEEE Photo J, vol. 9, no.1, pp.1-10, Feb. 2017. [5] D. Kim and J. Kang, “Sagnac loop interferometer based on polarization maintaining photonic crystal fiber with reduced temperature sensitivity,” Opt. Exp., vol. 12, no. 19, pp. 4490–4495, Sep. 2004. [6] J. Limpert et al., “Extended single-mode photonic crystal fiber lasers,” Opt. Exp., vol. 14, no. 7, pp. 2715–2720, Apr. 2006. [7] Y. Amma et al., “Fusion splice techniques for multicore fibers,” Optical Fiber Technology, 2016. [8] K. Saito et al., “Side-view based angle alignment technique for multi-core fiber,” 2016 Optical Fiber Communications Conference and Exhibition (OFC), Anaheim, CA, 2016, pp. 1-3. [9] K. Watanabe et al., “Study of fusion splice for single-mode multicore fiber,” 17th Microopics Conference (MOC), Sendai, 2011, pp. 1-2. [10] J. Canny, “A Computational Approach to Edge Detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-8, no. 6, pp. 679-698, Nov. 1986. [11] J. Illingworth and J. Kittler, “The Adaptive Hough Transform,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-9, no. 5, pp. 690-698, Sept. 1987. [12] L. Stefano et al., “Zncc-based template matching using bounded partial correlation,” Pattern Recognit. Lett., vol. 26, no. 14, pp. 2129–2134, Oct. 2005. [13] Y. Amma et al., “Low-loss fusion splice technique for multicore fiber with a large cladding diameter,” 2013 IEEE Photonics Society Summer Topical Meeting. Series, Waikoloa, HI, 2013, pp. 74-75. [14] D. Marcuse, “Loss analysis of single-mode fiber splices,” The Bell System Technical Journal, vol. 56, no. 5, pp. 703-718, May-June 1977.
1943-0655 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
