1. Introduction

The ITU-T standard G.654 governs specifications for cut-off shifted optical fiber, which until now has almost exclusively been deployed in undersea optical communication systems [1]. The standard was very recently revised to include a new G.654.E optical fiber standard for terrestrial transmission system applications. The general aim of the new standard is to allow terrestrial systems to be constructed with optical fibers with larger effective areas to reduce nonlinear impairments and thus improve the system reach with new high data rate multilevel modulation formats. However, a trade-off in fiber design generally requires allowing higher fiber and cable cut-off wavelengths as a means of reducing bend loss sensitivity for fibers with larger effective area. Thus the maximum cable cut-off specification of 1530 nm for the new G.654.E standard will be significantly higher than that for conventional terrestrial G.652-compliant fiber.

While the cable cut-off wavelength is below the signal transmission C-band, the same may not be true for the optical supervisory channel, depending on the system vendor’s design. The OSC is an important element of terrestrial telecommunication networks. It is a low data-rate (<2.5 Gb/s) optical channel located outside of the main transmission band that communicates system and control information between optical amplifier sites. In many system designs, the wavelength of 1510 nm is often used for OSC transmission [2]. Since 1510 nm falls below the likely cable cut-off wavelength specification of the new G.654.E fiber standard, there has been concern that OSC transmission performance could be adversely affected by multipath interference. MPI can be generated during transmission mainly through mode coupling at the splice points which can result in multiple delayed copies of the signal interfering with the main signal at detection at the end of a span. In this paper we investigate the generation of MPI below cable cut-off in a typical G.654.B-compliant fiber under cabled conditions. While higher order modes (typically LP₁₁) may be excited below cable cut-off, the level of MPI produced depends largely on the coupling between LP₀₁ and LP₁₁ at splice points and the differential mode attenuation (DMA) between LP₀₁ and LP₁₁ in the cabled fiber and bend loss induced in splice trays. We showed earlier that OSCs operating at 1.25 Gb/s and 2.5 Gb/s suffered no penalty even under worst-case conditions at wavelengths at least 40 nm below the composite average cable cut-off of the constituent fibers [3]. Here, we extend our earlier MPI measurements and modeling with more cases of splice loss and higher order mode attenuation in splice trays. We also find that including a very small distributed mode coupling coefficient in the model seems to enhance the agreement of the model results to the experimental data.

2. Simple model of MPI generation

To predict MPI levels, we first develop the simplest possible model of the growth of MPI in a fiber span that assumes MPI is generated only at the splice points within the span. We make the conservative assumption that at wavelengths below the cable cut-off, any loss of the fundamental mode at a splice point results in the lost optical power being fully coupled from the LP₀₁ mode to the higher order LP₁₁ mode. The power coupling coefficient is given by $\epsilon =1-{10}^{\left(-L(dB)/10\right)}$ where L(dB) is the splice loss of the fundamental mode. By reciprocity, any optical power already in LP₁₁ will couple back to LP₀₁ with the same coefficient. In this way, MPI is generated as power couples both ways between the modes at splices and thus multiple crosstalk terms may build up with delays relative to the signal (due to different modal group velocities) during propagation through a span. The MPI level is defined as the ratio of total crosstalk power coupled back into the fundamental mode to average signal power as given in Eq. (1). The total crosstalk power P_xtalk,total is the sum of the discrete power crosstalk terms generated at each splice, and the total MPI at span end is governed by the splice losses, distance between splices, and DMA.

Results from the simple MPI model for spans similar to the experimental set-up to be described in the next section are shown in Fig. 1. Figure 1(a) shows how MPI can build up as a function of the distance into a 92.4 km span, with discrete splice points within the span located every 4.2 km. The MPI level is given for 4 different values of DMA, which represents by how much the LP₁₁ mode attenuation exceeds the fundamental mode LP₀₁ attenuation. The data in Fig. 1(a) assumes the average splice loss is 0.017 dB and the first and last splices of the span are higher at 0.1 dB because the larger A_eff G.654 fiber is spliced to smaller A_eff standard single-mode fiber. Figure 1(b) shows the total MPI at the end of the span as a function of DMA for three different splice loss values. The splice losses of 0.024 dB, 0.017 dB, and 0.014 dB correspond to measured average splice losses between like fibers with effective areas of 82 μm², 112 μm², and 150 μm², respectively [4,5]. The data in Fig. 1 assume that each splice is a straight splice with no fiber loops surrounding the splice.

 

Fig. 1 MPI model results. (a) MPI generation as a function of distance into a 92.4 km span with 4.2 km distance between splices. Average splice loss = 0.017 dB. (b) MPI at span end as a function of DMA.

Download Full Size | PDF

3. Experimental set-up and results

To investigate the generation of MPI below cable cut-off and the performance of OSC channels below cut-off, we cabled G.654.B-compliant fiber (same maximum cable cut-off specification of 1530 nm) in two gel-filled buffer tubes [3]. The optical fiber was Corning® Vascade EX2000 with nominal effective area of 112 μm². Both buffer tubes were 4.2 km long and were each filled with 11 used optical fibers along with 1 unused fiber. As illustrated in Fig. 2, the fibers in the buffer tubes were spliced to each other and between the two tubes to create a 92.4 km optical span that had 21 intra-span splices. The larger effective area G.654-compliant fibers were spliced to standard G.652 fiber jumpers with smaller effective area at each end of the span. Each buffer tube was loosely coiled with a minimum diameter of 1 m in an effort to approximate realistic deployed cable conditions as much as possible. The G.654 fibers used in the tubes were mainly comprised from two source reels. One of these fibers had a measured cable cut-off wavelength of 1520 nm and the other had a cable cut-off wavelength of 1480 nm. The cable cut-offs were measured with 2 m fiber lengths including 2x80mm diameter loops. While the span length of 92.4 km was fixed in this study, it is representative of many terrestrial spans. Furthermore, inspection of Fig. 1(a) reveals that MPI grows slowly with longer span lengths. For example, extending the span from 92.4 km to 121.8 km will increase the total MPI by < 2 dB even for the case with DMA = 0 dB/km.

 

Fig. 2 Experimental configuration of 92.4 km optical fiber span constructed for MPI measurements.

Download Full Size | PDF

For the first set of MPI measurements, we created the span under worst-case conditions in which all splices were done in a straight-through manner, meaning that the spliced fiber was passed straight through the splice trays without adding any approximately 80 mm diameter loops on either side of the splices such as is commonly done in splice trays in terrestrial systems. Such loops can act as a higher order mode filter to attenuate any excited higher order modes but were not employed in the first experiments. The individual splice losses were not measured or controlled, but the total span loss was consistent with average intra-span splice losses of ~0.017 dB and with G.654-G.652 fiber splices of about 0.1 dB.

We made MPI measurements over a range of wavelengths from 1400 nm to 1520 nm in the first splice condition. The measurements were made by analyzing the transmitted power fluctuations of a tunable CW external cavity laser with linewidth 100 kHz through the span as described elsewhere [6,7]. At each measurement wavelength, detected power samples were collected every 20 ms, with a total of at least 200,000 samples. The laser polarization was continuously scrambled during the measurements to explore all possible input polarization states with a rate corresponding to about 0.03 steradians per measurement sample. The MPI was calculated assuming random relative polarizations between the main signal and crosstalk terms according to the relation

 

Fig. 3 MPI measured data through 92.4 km cabled span at wavelengths below cable cut-off.

Download Full Size | PDF

To investigate the effect of splice tray loops on MPI measurements as well as to check our MPI model, we next introduced 3x80 mm diameter fiber loops on both sides of each splice. Such loops are commonly found in terrestrial system splice trays at each splice point. We also made measurements of the actual LP₁₁ mode loss per 80 mm diameter turn for the constituent fibers to include in our model of the propagation down the fiber span. The LP₁₁ loss measurements were made with a cut-back approach designed to isolate this parameter [9]. The LP₁₁ bend loss data are shown in Fig. 4(a) for the two fibers that comprised the cabled span. These data show that the LP₁₁ bend loss is about 3 dB/turn even approximately 60 nm below the cable cut-off wavelength.

 

Fig. 4 (a) Measured LP₁₁ mode bend loss data per 80 mm diameter turn. (b) Experimental and model MPI data with and without 3x80mm loops around splices.

Download Full Size | PDF

In Fig. 4(b), the measured MPI values with straight-through splice conditions are again shown for reference, along with new measurements when 3x80 mm loops are put on each side of the splices in the splice trays. Also included are two sets of model predictions for the MPI with the loops present. The model predictions were produced as follows. First, we used the simple model described earlier in which mode coupling occurs only at the splice points with coupling strength governed by the splice loss. Given knowledge of the distance between splices and the measured MPI values, it is possible to then estimate the DMA values at each wavelength. Applying those DMA values to a modified model with an assumed LP₁₁ bend loss calculated as the average of the two fibers at each wavelength, and 3 loops on each side of every splice, we obtain the MPI predictions corresponding to the data in Fig. 4(b) with parameter k = 0. The parameter k represents a distributed power mode coupling coefficient that describes the strength of mode coupling occurring in a uniform distributed manner during propagation in the fiber [10]. The value of k = 0 means that this type of coupling was ignored and coupling occurred only at the splice points according to the simplest model. The other model data in Fig. 4(b) corresponds to a case with a very small value of distributed mode coupling, k = 9x10⁻⁵ km⁻¹. This model was implemented by sub-dividing each fiber length between splice points into segments of 0.01 km, and applying the coupling coefficient k to both LP₀₁ and LP₁₁ modes at every point between splices. The step size used was sufficiently small to ensure accuracy of the calculations [10]. In this case, the implied DMA values at each wavelength need to be re-calculated such that the initial measured MPI values with no loops are obtained. Finally, with revised DMA values corresponding to the coupling coefficient k, we can then predict new MPI values in the presence of the 3x80 mm loops around the splices. We found that with the small distributed coefficient k = 9x10⁻⁵ km⁻¹, the model predictions fit the measured MPI data with loops slightly better than if k is assumed to be 0. This was especially true at the longer wavelength range from about 1450-1480 nm.

When the cabled span was first constructed we took no special care with regard to the splices and ended up with a total span loss consistent with the expected average intra-span splice loss of 0.017 dB and G.654-G.652 fiber splice losses of about 0.1 dB. To further study splice loss effect on MPI below cut-off and the MPI model accuracy, we then intentionally introduced higher splice losses in 10 out of the 21 intra-span splices. The average of these 10 higher splice losses was 0.07 dB, making the average overall intra-span splice loss about 0.042 dB. New MPI measurements were carried out over the same wavelength range. The measurement results are shown in Fig. 5(a) for the cases with no loops placed around the splices, and for 3x80 mm diameter loops around the splices. The model predictions based on the estimated DMA values and distributed coupling coefficient determined from the prior measurements are included for comparison and agree well with the experimental data.

 

Fig. 5 (a) MPI measurements and model predictions of new span configuration with higher splice losses. (b) Estimated DMA values below cable cut-off for 92.4 km cabled span.

Download Full Size | PDF

Several observations can be made from the data in Fig. 5(a). The first is that the higher splice losses introduced do increase the measured MPI by approximately 6-10 dB with no fiber loops in the splice trays. However, even in this condition the measured MPI was ≤ −32 dB for all wavelengths down to at least 1440 nm, which is 40 nm below the cable cut-off of the lowest cut-off constituent fiber in the buffer tubes. Therefore we would expect no transmission penalty for an OSC even down to 1440 nm. We also see that the placement of a few 80 mm diameter loops around the splices easily suppresses the MPI to levels ≤ −40 dB all the way down to the lowest measured wavelength at 1400 nm. And finally, we also note that the MPI model using the derived DMA values and small distributed coupling coefficient from the previous measurements gives accurate predictions of the new MPI measured results, increasing our confidence in the model validity. The DMA estimates obtained from the MPI model for the cabled span are shown in Fig. 5(b), along with an exponential function fit. The estimated DMA was >12 dB/km at 1480 nm. Recall that half of the fiber in the cable had a cable cut-off of 1480 nm while the other half had a cable cut-off of 1520 nm. These DMA values represent the average condition for the full span of fiber in the cable itself and do not account for the extra attenuation to LP₁₁ introduced by the 80 mm diameter fiber loops as shown in Fig. 4(a).

4. Summary and conclusions

We have examined in experiments and modeling the level of MPI generated at wavelengths below cable cut-off for a G.654.B-compliant fiber in a cabled fiber span. This work was initially geared toward understanding if optical supervisory channels operating below cable cut-off may be subject to MPI penalties. We found that the levels of MPI observed predict no penalty, which was confirmed by earlier transmission experiments. Here we more closely investigated the validity of our MPI model for such a span. We found that the simplest possible model with mode coupling generated only at the splice points is reasonably accurate, but that the overall fit of the model predictions to the measured data can be improved by the inclusion of a very small distributed coupling term. The model was compared with measured MPI data with good accuracy in several span conditions, with and without the inclusion of 80 mm diameter fiber loops in the splice trays, and with nominal and intentionally increased splice losses. Overall, the MPI level was measured to be ≤ −40 dB down to at least 1400 nm when 3 fiber loops were used in the splice trays, even with a significantly higher average intra-span splice loss of 0.042 dB than would be expected from normal splicing procedures.