of the whole link. Another possible application is at the manufacturing stage, before cabling, to ensure that the bers have uniformly distributed PMD.

Transcription

1 Distributed PMD measurement with a Polarisation-OTDR in optical bers B. Huttner, B. Gisin and N. Gisin Group of Applied Physics University of Geneva, 1211 Geneva 4, Switzerland March 10, 1999 Abstract We present a new method for the measurement of distributed polarisation mode dispersion (PMD) in optical bers. This method uses a polarisation optical time-domain reectometer(p-otdr), and is based on a measurement of the degree of polarisation of the backscattered light as a function of distance in the ber. Both the average and the statistics of the degree of polarisation are used to estimate the two relevant parameters for measuring PMD, namely the beat length and the coupling length. At present, our P-OTDR gives qualitative results only. However, it enables to distinguish between high and low PMD sections in a long ber link. This should already have practical applications, in particular for the characterization of installed bers. 1 Introduction In modern optical communication systems, Polarization Mode Dispersion (PMD) [1] is increasingly becoming a serious limitation to the transmission rate. The problem is not so acute for new bers, which are now well specied for low values of the PMD, but rather for systems, which were installed several years ago. In these systems, the PMD of an optical link may reach very large values (several measurements of tens of picoseconds for links of several kms have been reported). In such a case, it is impossible to upgrade the transmission rate on this link without PMD compensation. At present however, there exist no commercially available PMD compensators, although various groups are working to develop them [2, 3, 4]. Therefore, when confronted with a link with high PMD, a telecom operator has only two choices: either he does not use this particular ber, or he has to unearth the whole length of the ber cable and replace it. Neither solution is really satisfactory, all the more so since, in general, the high PMD of the link is only caused by one or a few bad sections. This can be traced back to the fabrication process of old bers, where PMD was not even measured. Sections with very high PMD were therefore unknowingly introduced in cables. For this reason, there is currently a strong incentive to develop an instrument to measure, at least qualitatively the distributed PMD on installed optical ber cables. This is precisely the aim of our polarisation optical time-domain reectometer (P-OTDR). This instrument should enable to identify the worst PMD sections of a link, and only replace them. Alternatively, identifying sections with very high PMD, caused for example by very high birefringence, may enable to take into account in a better way the PMD statistics Bruno.Huttner@physics.unige.ch 1

2 of the whole link. Another possible application is at the manufacturing stage, before cabling, to ensure that the bers have uniformly distributed PMD. This improved quality control at the production level could prevent the costly errors of the past to recur. The idea of a P-OTDR was rst suggested by Rogers [5], almost twenty years ago. At this time, the main concern was not so much PMD, which was not seen as a signicant limitation for optical communications, but rather distributed measurements of physical parameters, such as temperature, strain, etc..., capable of aecting the polarisation state of light propagating down the ber. This idea was soon applied to the measurement of bend-induced birefringence in single-mode bers [6], and more recently to the measurement of twist-induced birefringence in short lengths of bers [7]. It can also be used to extract the intrinsic local birefringence in a single-mode ber. We have used a dierent instrument with a much higher spatial resolution, the coherent optical frequency domain reectometer, to measure the local birefringence and estimate the PMD [8]. These methods are interesting for the characterisation of short trunks of bers during the manufacturing process, to optimize the cabling process by imposing some xed twist, or to assess the response of a ber to external constraints. However, they cannot provide an estimate of the distributed birefringence properties on long sections (tens of kms). A slightly dierent setup was used to measure directly the dierential group delay (DGD) along a ber, using a frequency tunable laser [9, 10]. The methods used are adaptations of the xed analyser method [9], and of the Jones matrix eigenanalysis (JME) method [10] for measuring PMD, taking into account the fact that the light propagates through the ber twice [11]. To date, this method was only applied to bers with rather low PMD (in [9], the largest value was 0.21 ps= p km, and in [10] it was less than 0.1 ps= p km). To our knowledge, the case of long ber cables, made of several sections of dierent bers, including sections of high PMD, has not been fully addressed. The reason is that the usual way of analysing the birefringence properties of the ber by looking at the evolution of the state of polarisation (SOP) is not well adapted to this case. This is precisely the aim of this work. We will rst review the principle of the P-OTDR in Section 2, emphasizing that a P-OTDR does not measure directly the PMD, but two physical parameters, the beat length and the couling length, from which one can infer the value of the PMD. In Section 3, we will describe the limitations of the standard method, explaining why the SOP is not in our opinion the best parameter for an estimation of PMD of long bers, especially with high PMD. We will show that a more relevant parameter is the degree of polarisation (DOP) of the backscattered pulse. However, one has to dierentiate between the depolarisation caused by the nite linewidth of the OTDR laser, which is deleterious and should be avoided, and the temporal depolarisation, directly caused by the local birefringence properties, which is the basis of our measurement method. In Section 4, we will present experimental results, obtained from two series of ber links. The rst one is only composed of rather good bers, with low PMD, while the second one comprises also bad sections, with very high PMD values. In both cases, we will show that, with a good degree of condence, we can nd out the best and worst sections within the link. However, in our opinion, a truly quantitative measurement of distributed PMD along long ber links is at best a dicult task with standard telecom equipment. 2 Principle of P-OTDR 2.1 Basics The basic setup for a P-OTDR is shown in Fig. 1. A pulsed laser source emits short polarised pulses, which are sent to the ber under test (FUT) through a ber coupler. Rayleigh 2

3 backscattering, modeled as a distributed reection [12], sends some backreected light through a polarisation analyser (PA), which extracts the three Stokes components, s 1, s 2 and s 3, as a function of the time (the origin being the pulse entering the device). This enables to obtain the full SOP of the backscattered light as a function of the distance of the scattering point. The best way to represent the evolution of the SOP is to use the Poincare sphere, the SOP being a point moving on the sphere. 2.2 What do we measure In a section of birefringent ber, the SOP of the light transmitted through the ber rotates around the birefringence axis. The velocity of this rotation is given by the local birefringence,, which in turns denes the beat length of the section, L b by: L b = c ; (1) where is the wavelength of the light. L b is the periodicity of the rotation about the axis. The evolution of the birefringence axis gives the other important parameter, the coupling length, h. Loosely speaking, the coupling length is the distance after which the birefringence axis has moved signicantly. A more precise denition involves the spatial correlations of the birefringence vector: (z) ~ = (z)^u, where is the birefringence and ^u is the birefringence axis on the Poincare sphere: Z 1 L ~(z) (z L ~ + )dz = 2 exp 0,2 jj h ; (2) where is the RMS average of the birefringence over the length L, and can be obtained by letting = 0 in Eq. 2. The exponential decay of the correlation in Eq. 2 has been well veried experimentally. In principle, by following the evolution of the SOP along the ber, one can derive both the beat length (how fast the SOP moves) and the coupling length (how fast the rotation axis moves). However, practically speaking, this is not so easy. By plotting the SOP for example on the Poincare sphere, and looking at the displacement as a function of distance, one can infer L b. To derive h however, one needs to nd the rotation axis at all points, which involves dierentiation of the curve. Therefore, any noise or inaccuracy on the SOP will be strongly enhanced. This is the reason why this direct method has only been applied to bers with known birefringence axis, to derive L b. The above analysis shows that the physical local parameters, accessible through this type of measurement, are the beat length and the coupling length. The PMD itself is not a local parameter, but is a global quantity, which needs to be calculated from these parameters. For a section of length l, which is long with respect to both beat length and coupling length, the PMD is given by [13]: PMD = p lh : (3) L b c For a link made of a concatenation of N sections i, the global PMD is: PMD tot = vu u tx N i=1 PMD 2 i : (4) Eq. 3, although well-known, is quite interesting with respect to the P-OTDR. It shows that the two parameters, L b and h, appear in opposite ways in the denition of PMD. A large L b means high PMD, while a large h means low PMD. It is thus crucial to be able to estimate 3

4 both parameters separately. For standard bers, including old ones with high PMD, the range of variation of these parameters is the following: 0:5 L b 20 m 5 h 500 m : (5) In general the main characteristic of older bers, with high PMD, is the longer coupling length, with respect to more recent bers. The most recent bers have both a long beat length, and a short coupling length. An extra problem, not mentioned yet, comes from the use of backscattered light. This is better explained on an example. If we take a ber with purely circular birefringence, it is easily shown that the state of polarisation of the backscattered light is independent of the birefringence. The polarisation transformation on the way back compensates the forward transformation. If the birefringence is purely linear, then the periodicity of the rotation is L b. For a birefringence 2 axis, which is neither circular nor linear, the polarisation rotates around the axis with two periods, L b and L b. The Fourier transform of a trace, giving the two components can still be 2 used to extract the beat length. A careful analysis of the backscattering technique based on the evolution of the SOP is presented in [10]. This work also explains how to measure directly the distributed PMD by an adaptation of the JME method. 2.3 Depolarisation in wavelength Depolarisation in wavelength is a common problem, both for the standard method using the SOP, and for the new method using the DOP we are suggesting. It is caused by the nite linewidth of the laser of the OTDR. Indeed, when the PMD becomes large, the various frequency components of the pulsed laser experience dierent polarisation transformations. The pulse therefore becomes depolarised. This depolarisation is a cumulative eect, increasing along the ber. Therefore, after, say one section with high PMD, the pulse may be already depolarised, and no information about subsequent sections can be obtained [14]. This is a very serious limitation with usual OTDR lasers, which have a broad linewidth to prevent coherent eects from causing ripples in the Rayleigh backscattering. The requirement for a P-OTDR laser is largely the opposite: the linewidth needs to be small enough to prevent this depolarisation. In our P-OTDR, we do not use a standard OTDR laser, but had it replaced by a DBR laser with a sharper linewidth, about 0.2 nm. This corresponds to a coherence time of about 6 ps. This is enough to prevent depolarisation in wavelength, at least for PMD up to about 10 ps. Therefore, in the following, we shall assume that the laser linewidth is small enough to prevent this eect. 3 Using the DOP 3.1 Limitations of the standard method The usual way to obtain the PMD is to analyse the variation of the SOP along the ber. In principle, this is certainly the most precise way to measure the two important parameters, L b and h. However, in practice, we are limited by the length of the pulse, L p of the P-OTDR. The problems comes from the fact that the backscattered light, arriving at the detector at a given time, is in fact coming from several points in the ber, situated within half of the pulse length (because of the return trip). If L p L b, the polarisation of the light coming from these points is not constant. The measured SOP is thus only an average value, and the variations of the SOP as a function of distance are smoothened by this averaging. They do not directly reect 4

5 the local birefringence. The problem is also acute for methods measuring directly the PMD. For example, the xed analyser method is based on counting the number of extrema in the intensity going through a xed polariser. If the SOP varies within the pulse length, the visibility ofthe extrema decreases, due to the constant depolarised component. Therefore the method is only valid for L p L b. The shortest pulses available with standard OTDR equipment are in the tens of ns, which correspond to a length of a few meters. Therefore, the minimum value of L b, for which the method using the SOP is usable is several meters. Using the maximum value of the coupling length in Eq. 5 (500 m), this gives a maximum PMD of 0.7 ps= p km. This shows that the SOP method is limited to rather good bers, but may fail for bad ones, where L b may even be below one meter. Since the main incentive for the P-OTDR is to be able to analyse bad ber links, this is a serious limitation. Another problem linked to the use of such short pulses is the total intensity received at the detector. This is especially dicult for long links, where the attenuation along the line is already signicant. In fact, we have found out that the backscattered intensity we got from the shortest pulses we could get with our OTDR laser (30 ns) was too low after several kms. The noise on the measurement was too high to get any signicant result. One solution is to boost the intensity with an optical amplier, but this adds costs, and may also bring nonlinear eects, which may be hard to analyse. Therefore, we suggest to transform this limitation into an advantage, and use the DOP, which characterizes the change of polarisation within the pulse, to estimate PMD. 3.2 Temporal depolarisation The variation of the SOP within the optical pulse brings about depolarisation, as mentionned in section 3.1. It is created by the local birefringence at the point of retroreection. Therefore, the DOP of the pulse as a function of length gives information about the local birefringence. Let us emphasize that this type of depolarisation, on which we base our measurement method, is totally dierent from the depolarisation in wavelength discussed in 2.3. Depolarisation in wavelength is a cumulative eect, caused by the accumulated PMD between the begining of the ber and the reexion point. It can be prevented by using a laser with sharp enough linewidth. In contrast, the temporal depolarisation is a purely local eect, created by the birefringence at the reexion point. It could be reduced by using shorter pulses, as discussed above, or can be used to measure local birefringence, as we suggest here. The analysis of the DOP as a function of distance enables to dene three characteristic domains, depending on the relative values of the pulse length, L p, the beat length, L b, and the coupling length, h. The rst domain is dened by L p L b. In this case, the whole pulse has the same polarisation, and so the DOP is close to 1. This remains correct, regardless of the value of h (since there is nearly no polarisation change within the pulse, variations in the axis of birefringence does not modify the DOP). The DOP of the backscattered light is also close to 1. This is characteristic of low PMD bers, as already mentioned above. In the other extreme, if L p L b, the polarisation experiences several rotations along the pulse, and the DOP is reduced. To analyse this further, let us separate two domains, with respect to the coupling length. If L p h, the axis of birefringence of the ber also changes rapidly within the pulse. Therefore, the polarisation along the pulse is completely random, and the DOP around the reexion point is nought. When we consider the backreection, it is known that light, which was totally depolarised during the propagation experiences partial repolarisation on the way back, such that the mean DOP is 1 [12]. We conrmed this experimentally, by sending very 3 long pulses (1000 ns). This case corresponds to medium PMD's. The last case corresponds to h L p L b, i.e. the birefringence axis does not change along 5

6 the pulse, but the polarisation rotates quickly around the axis. The expected value of the DOP depends on the angle between the polarisation and the birefringence axis (since the polarisation rotates around the birefringence axis, this angle remains constant). The value of the DOP of the pulse at the reexion therefore uctuates as a function of distance, with the displacement of the birefringence axis, i.e. on a length scale of h. This remains correct for the backscattered light. This last case, with long coupling lengths, corresponds to high PMD's. 4 Experiments 4.1 Experimental setup Our P-OTDR prototype is based on a standard OTDR system (made by EXFO), with an added polarisation control system as explained in Fig. 1. In order to have intensities as high as possible, the coupler was replaced by a circulator (thus gaining 3 db). To prevent the depolarisation in wavelength, mentioned in Section 2, a special laser was adapted, with linewidth of about 0.2 nm, fty times smaller than the linewidth of the usual OTDR laser. The peak intensity of the laser was about 40 mw. In order to get enough backscattered intensity, we used 100 ns pulses. Shorter pulses would give very noisy curves, which could not be analysed. We also tested longer pulses, and found out that the DOP would go down to 1 as expected. This means complete 3 depolarisation on the way down the ber, and partial repolarisation on the way back. Our polarisation analyser was simply a rotating quarter waveplate followed by a rotating polariser, both controlled manually. One measurement therefore involved three scans, for the three Stokes components (in fact, in order to reduce incertainties, we performed six measurements, checking consistency between them). The software of the instrument was modied, to remove all the avering procedure needed to smoothen the OTDR curves. The results are presented in the next two sections. 4.2 Low PMD sections We analysed a 27-km long ber link, made of 16 sections of various lengths. The result is presented in Fig 2. The PMD of each section was measured prior to this experiment, and the values are shown above the graph. In this acase, all sections were reasonably good, with the highest PMD values about 0.5 ps, for sections of a few km's. Let us emphasize that the lowest value of the PMD, ps, is only the lowest measurable PMD with the instrument used. We believe that some of the sections have indeed much lower values. The gray curve, oscillating very rapidly, is the raw data, with no averaging. We see that the DOP uctuates wildly as a function of distance. To get usable results, we have to perform a rst average on the data. The one adopted was an average over 50 points, and the resulting curve is drawn in black. This curve shows clearly the link between average DOP and PMD. The good bers having a high DOP, and the bad ones a low DOP, tending to about 1 as expected. Note that the correlation 3 is however not perfect, with some sections with low PMD appearing with low DOP (sections 7 and 8), that is a short beat length. One possible explanation is that both L b and h are small, so that the PMD is still small. 4.3 High PMD sections The main application of the P-OTDR is the analysis of the PMD of old installed bers links with high PMD, which cannot be used after an upgrade of the transmission rate for digital transmissions, or introduce large distortions in the analog case. The idea is to identify the worst 6

7 sections, and start by changing those rst. This may represent large savings, both in time and cost, with respect to the alternative ofchanging all the sections in the link. It is therefore most important to be able to identify specically the high PMD sections. As seen in Section 4.2, the average DOP for bers with PMD higher than about 0.5 ps= p km is already about the minimum value of 1. This method does not allow to dierentiate betweeen bad and worse 3 bers. To achieve this goal, one has to turn to the statistics of the DOP. We illustrate this with Fig 3, which represents the trace of a17km-long ber link made of 6 sections, including some very bad ones (PMD up to 5.83 ps for a 2 km-long section). We see that, except for the rst section, which hasalow PMD and a high value of the DOP, all other sections have alow DOP, about 0.4. The dierence between the sections is given by the DOP variations. As mentionned above, the uctuations in the DOP are characteristic of the coupling length h. If h is longer than the pulse, the birefringence axis is xed along the pulse, and the DOP, which is already the average of the polarisation rotating around the axis, remains constant for about h. To analyse this, we perform Fourier transforms of the signals for all 6 sections in g. 4. We clearly see the sections with highest PMD, or equivalently longest coupling length, correspond to the highest low frequency components of the Fourier spectrum. 5 Conclusion We have developed an new Polarisation-OTDR, based on the measurement of the DOP of backscattered light. Our instrument does not give a direct measurement of the PMD, but rather estimates separately the beat length and the coupling length. We have shown that both the average DOP and its statistics are required to access these two parameters. The instrument gives good qualitative predictions of the PMD of the various sections within a long optical link. In particular, it can pinpoint the sections with highest PMD. Our P-OTDR should be of use for telecom operators, who possess old ber links with high PMD, and would like to discover the worst sections of their link. It could also be used by ber manufacturers, to perform quality control of their bers in the production stage. Further studies are still needed to assess the use of our P-OTDR for quantitative measurements of distributed PMD. In particular, we plan to add an optical amplier to our setup. This should enable us to use shorter pulses of higher intensity. Acknowledgements We gratefully acknowledge nancial support from EXFO (Vanier, Canada). We would also like to thank Swisscom and Datwyller (Switzerland) for access to their test bers. This work was supported by the Swiss Oce Federal de l'education et de la Science, within the COST 265 European project. References [1] \Fiber optic test and measurement", D. Derickson, Ed. (Prentice Hall PTR, Upper Saddle River, NJ, 1998), pp [2] R. Noe, D. Sandel, M. Yoshida-Dierolf, S. Hinz, C. Glingener, C. Scheerer, A. Schopin and G. Fisher, \Fiber-based distributed PMD compensation at 20 Gb/s", presented at ECOC'98 (Sept. 98, Madrid, Spain). 7

8 [3] F. Heismann, D.A. Fishman and D.L. Wilson, \Automatic compensation of polarization mode dispersion in a 10 Gb/s transmission system", presented at ECOC'98 (Sept. 98, Madrid, Spain). [4] F. Roy, C.Francia, F. Bruyere and D. Penninckx, \A simple dynamic polarisation mode dispersion compensator" in Optical Fiber Communication Conference, OSA Technical digest (Optical Society of America, Washington DC, 1998), TuS4-1, p 275. [5] A.J. Rogers, \Polarization-optical time domain reectometry: a technique for the measurement of eld distributions", Appl. Opt. 20, pp (1981). [6] Byoung Yoon Kim and Sang Sam Choi, \Backscattering measurement of bending-induced birefringence in single mode bres", Electron. Letts. 17, pp (1981). [7] R.E. Schuh, E.S.R. Sikora, N.G. Walker, A.S. Siddiqui, L.M. Gleeson and D.H.O. Beddington, \Theoretical analysis and measurement of eects of bre twist on polarisation mode dispersion of optical bres", Electron. Letts. 31, pp (1995); J.G. Ellison and A.S. Siddiqui, \A fully polarimetric optical time domain reectometer", IEEE Photonics Techno. Letts. 10, pp (1998). [8] B. Huttner, J. Reecht, N. Gisin, R. Passy and J.P. von der Weid, \Local birefringence measurements in single-mode bers with coherent optical frequency-domain reectometry", IEEE Photonics Techno. Letts. 10, pp (1998). [9] H Sunnerud, B.-E. Olsson and P.A. Andrekson, \Technique for characterisation of polarisation mode dispersion accumulation along optical bres", Electron. Letts. 34, pp (1998). [10] F. Corsi, A. Galtarossa, L. Palmieri, \Polarisation mode dispersion characterisation of single-mode optical ber using backscattering technique", J. Lightwave Techno. 16, pp (1998). [11] There is a discrepancy between [9] and [10] in the ratio of PMD for backscattered light and forward propagating light. In [9] this ratio is p 2, while it is in [10]. In our opinion the 2 correct approach is in [10]. [12] M.O. van Deventer, \Polarization properties of Rayleigh backscattering in single-mode bers", J. Lightwave Techno. 11, pp (1993). [13] N. Gisin, J.P. von der Weid and J.P. Pellaux, \Polarisation mode dispersion of short and long single-mode bers", J. Lightwave Techno. 9, pp (1991). [14] E. Chausse, N. Gisin and Ch. Zimmer, \POTDR, depolarization and detection of sections with large PMD", presented at Optical Fiber Measurement Conference (OFMC) '95 (Sept. '95, Liege, Belgium). Figures 8

9 Figure 1: Typical polarisation-otdr setup The OTDR mainframe drives a pulsed laser, emitting short (between ten to several hundreds ns) and intense pulses. The pulses are sent trough a coupler towards the ber under test. backscattered light goes trough the polarisation analyser, onto a fast detector. This enables to obtain the three Stokes components, and thus the state of polarisation (SOP), as a function of the distance of the reection. Analysing the SOP, one can derive the distributed PMD of the ber under test (see text for details). 9

10 PMD of the section [ps] DOP Length [km] Figure 2: Trace of a ber link with low PMD sections For ber section with relatively low PMD (here the highest PMD is about 0.5 ps= p km), the average degree of polarisation (DOP) is a good indicator for PMD. High DOP means long beat length L b, corresponding to low PMD. The gray curve represents the raw data, with no averaging. The black curve is the average over 50 points (one point representing 3 m). For comparison, we add, above the trace, the value of the PMD, which was measured for each section with a standard interferometric measurement. The lowest PMD value ( ps) is not representative, but corresponds to the lowest range of the instrument. We see that for PMD above 0.5 ps= p km) the average DOP is already 0.4, close to the predicted minimum of

11 PMD of the section [ps] DOP Length [km] Figure 3: Trace of a ber link with high PMD sections The optical ber link is made of one low PMD section, followed by ve high PMD ones. The full trace corresponds to the DOP averaged over 20 points. The dashed horizontal lines are the average DOP's over each section. We see that the average DOP does not allow to dierentiate between the bad sections. However, the large uctuations in the DOP, created by long coupling lengths, are characteristic of sections with highest PMD. A more detailed analysis is shown in Fig. 4 11

12 PMD of the section: ps PMD of the section: 1.42 ps PMD of the section: 3.17 ps 0.04 Amplitude [a.u.] PMD of the section: 0.83 ps PMD of the section: 5.83 ps PMD of the section: 1.17 ps Frequency [Hz] Figure 4: Fourier transforms of the DOP for each section in Fig. 3 Except for the st section, which has a higher average DOP, and thus a lower PMD, the ve other sections have roughly the same average DOP. To dierentiate between them, we made a Fourier transform of the DOP. The sections with larger low frequency components correspond to longer coupling lengths (the axis of polarisation, and therefore the DOP remains constant over the coupling length), and thus to higher PMD. 12