We propose a novel fiber design optimized for short-reach interconnects in consumer applications. A detailed analysis of the optical and mechanical properties of this fiber design is presented. Results are presented demonstrating (i) low bend loss and enhanced mechanical reliability in bends as small as 3 mm diameter; (ii) high power budget margin to enable relaxed mechanical tolerances on transmitter, receiver, and expanded-beam connectors for low-cost connectivity; and (iii) high bandwidth capability and system testing results at 10 Gb/s.
© 2012 OSA
Historically, optical communications have penetrated into various segments of communications networks driven primarily by a need to increase transmission speeds: from long-haul and submarine into metropolitan networks, and more recently in the “last mile” connection to individual houses with Fiber to the Home (FTTH) networks . Fiber is also common in Data Centers and Enterprise networks. It is becoming apparent that optical communication may find new opportunities in short-reach consumer electronics interconnects or home-networking .
Most often, new application spaces require a redesign of the fiber to achieve certain performance characteristics. For example, bend insensitive single-mode fibers were first introduced in FTTH applications in order to facilitate installations within apartment buildings [3, 4]. Routing the fiber cables throughout the building might introduce several bends which in turn can cause significant levels of bend induced attenuation in standard single-mode fibers. More recently, bend insensitive multimode fibers (50 µm core diameter) were developed for data center applications . Here a need for high density drives miniaturization of hardware and compact cable management systems, which in turn might result in numerous bends causing excessive loss in standard multimode fibers. Another requirement for both of these applications is a need to maintain backwards compatibility with industry standards given the vast installed base of fiber in these networks (ITU-T G.652 for FTTH applications and IEC 60793-2-10 for data centers). This backwards compatibility requirement imposes constraints in designing a new fiber profile designating that attributes have to be maintained within standard limits. For example, among other attributes single-mode fibers have to maintain cable cut-off wavelength no greater than 1260 nm imposing a limitation on how much bend-loss improvement is obtained in single-mode fibers ; For multimode fibers maintaining core diameter of about 50 μm and a numerical aperture (NA) of about 0.2 is also required. References [5, 6] discuss in detail the impact of these constraints on the design of the fiber index profile.
From the point of view of fiber design, consumer interconnects represent a new application space that has different requirements from traditional applications such as in data centers. First, backwards compatibility with standard fibers is not a strong requirement since optical fiber is not yet widely used in this space: this opens up new options in the fiber design. Link lengths are shorter: typically a few meters for the most common connections between computers and peripherals, and up to 20-30 meters for connections to devices such as a ceiling projector in a conference room. This alleviates the need for very high fiber bandwidth, although future upgradeability to higher data rates must be considered. The deployment conditions are more demanding in consumer interconnects: the fiber must withstand much smaller bend radii, since the cables will be handled and deployed by consumers rather than by trained professional technicians. This more demanding bend requirement creates a need to optimize the fiber for even lower bend loss, and the designer must also consider the mechanical reliability of the fiber under small-radius bend conditions. Finally, consumer applications require very low system cost, therefore the fiber must be designed to allow the use of inexpensive optical components, such as injection-molded plastic lenses and connectors, and also to allow relaxed mechanical tolerances in the assembly process to permit low-cost high-volume manufacturing.
In this paper, we investigate these requirements and propose a novel fiber design optimized for short-reach consumer interconnects. A detailed analysis of the optical and mechanical properties of this design is discussed. Results are presented demonstrating (i) low bend loss and enhanced mechanical reliability in bends as small as 3 mm diameter; (ii) high coupling efficiency with relaxed mechanical tolerances on transmitter, receiver, and expanded-beam connectors for low-cost connectivity; and (iii) high bandwidth capability to support 10 Gb/s over up to 100 meters. In section 2, we discuss the evolution of consumer connectivity and the potential advantages of using optical interconnects, and also lay out in more detail the requirements from a fiber design perspective. In section 3, we discuss the link budget and optimization of fiber core diameter and numerical aperture to allow relaxed tolerances. In section 4, we evaluate the mechanical reliability in tight bends. In section 5, we discuss the fiber profile design to achieve low bend loss and high bandwidth. Finally, in section 6 we present system performance testing results then conclude in section 7.
2. Evolution of consumer interconnects
Modern residential homes contain an ever increasing number of consumer electronic (CE) devices, which must be interconnected with each other and to the outside service providers that deliver entertainment content, telecom services and internet access. Today, broadband residential services are delivered to homes by coaxial cable, twisted pair, optical fiber, or broadband wireless. In most cases, the signal is terminated in a modem, and a “backbone” in-home network is used to distribute the signal from the modem throughout the house. There are several solutions for this backbone network such as wired Ethernet (IEEE 802.3), wireless (IEEE 802.11), power line communications (HomePlug), Multi-media over Coax Alliance (MoCA), and Home Phone-line Network Alliance (Home PNA). In addition to the backbone network, various interfaces exist today for a direct device-to-device communication. Examples are USB, HDMI, DisplayPort, FireWire, Thunderbolt and others. Figure 1 shows the evolution of transfer speed for these two types of networks in the home (backbone and device-to-device). A steady increase in data rates is observed in both, and it is interesting to note that device-to-device speeds are consistently significantly higher than the backbone speeds and therefore might be the first area where fiber is adopted in the future. Since 1995, the data rate for consumer protocols has increased by roughly two orders of magnitude reaching speeds around and above 10 Gb/s. USB 3.0 for example was released in 2009 at 5 Gb/s, HDMI spans a range up to 10.2 Gb/s, and in 2011 Thunderbolt was introduced with two channels each at 10 Gb/s.
With the increase in data rate, practical challenges arise for copper-based interconnects due to the higher cable losses at high frequency and crosstalk within the cable and electromagnetic interference with other devices: maximum transmission distance is shortened, cables and connectors become bulkier due to the additional wires used for parallel multiplexing, and power consumption increases due to equalization and re-timing chips. For example, a USB 3.0 copper cable has two additional differential twisted pairs compared to USB 2.0, and yet its reach was reduced from 5 meters to about 3 meters . Because of the additional wires the cable diameter increased to 5-7 mm depending on the construction. The connector plug also became wider to accommodate the extra wires taking up more space on the size of small devices. Thunderbolt  electrical cables are also limited to 3 meters in reach despite the use of equalization/re-timing chips embedded in the plug connectors (such implementation is referred to as an Active Electrical Cable, AEC). In comparison, a typical indoor two-fiber optical cable is approximately 3 mm in diameter, and can reach much longer transmission distances. Optical fiber is then being considered as a potential alternative to overcome these challenges, either in the form of active optical cables with electro-optical conversion embedded in the connector plug, or in the form of optical ports with the electro-optical conversion embedded in the device itself. Active optical cables are becoming popular for extended reach applications , and embedded optical ports are also being considered.
3. Link budget optimization
3.1 Full-link ray tracing simulation
Consumer devices must be manufactured in large volumes, with a broad supplier base and low cost. Therefore it is important to optimize the design of the optical link to allow relaxed mechanical tolerances on its components as well as relaxed requirements on alignment accuracy in assembly. Obviously misalignments can cause excessive optical loss thus compromising the link performance: one must then find an optimum design that allows relaxed tolerances but still maintain low link loss. In doing this, it is beneficial to choose the core diameter and NA of the fiber that minimize the total link loss in the presence of misalignments. In particular, a fiber with larger core and/or higher NA is less sensitive to misalignments at the transmitter and at in-line connectors, but an excessively large core or NA can impair the coupling efficiency at the receiver. For example, the use of Polymer Clad Silica (PCS) fibers with 200 µm core size and NA around 0.37 have been explored for low-cost links however with relatively low speed link. J. M. Trewhella et al  presented an analysis using 200 µm core size fibers to enable optical subassemblies and ferrules fabricated with plastic molding technology. T. Kibler and co-authors  presented a comprehensive analysis of a low-cost optical link for automotive applications with simple opto-electronics module and plastic ferrules based on 200 µm core size PCS as well as 1000 µm core size polymer optical fiber. Here, we study a high-speed link operating at 10 Gb/s. In this section we discuss our simulations of a typical link consisting of a transmitter, a receiver, and two in-line connectors: for a given set of alignment tolerances of the system, this simulation model enables us to estimate the optimum fiber core size and NA that minimize the sensitivity to misalignments.
We consider graded-index fibers, with core refractive index given by , where n1 is the maximum refractive index at the center of the core, α is the core curvature parameter and is typically close to 2, a is the core radius, and the core delta is with ncl being the cladding refractive index. With these definitions the NA is given by. As a first approximation, one could study the coupling loss individually at each junction (i.e. VCSEL-fiber, in-line connector(s), fiber-photodiode); however, when using multimode fibers, the misalignment at each junction also affects the modal power distribution in the fiber, and consequently the loss, at downstream junctions. To properly account for this dependence we created a ray-tracing model that includes all elements of the optical link as shown in the schematic of Fig. 2 . The link contains three lengths of multimode fiber with identical core diameter and NA. The optical transmitter in the simulation model uses an 850-nm Vertical-Cavity Surface-Emitting Laser (VCSEL) whose output is coupled into the fiber by a “photonic turn” element made of polymer material with refractive index 1.6395 at 850 nm. The aspheric lens pair in the photonic turn element has a magnification of approximately 1.4x. A similar photonic turn element is used in the receiver, providing approximately 1.4x demagnification from the fiber to the photodiode. The photodiode has a 60 µm diameter aperture, which is typical of devices designed for 10 Gb/s operation. The link includes two in-line Expanded-Beam (EB) connectors: the aspheric lenses in the EB connectors are also made of polymer material and produce an essentially collimated beam, approximately 300 μm in diameter. Compared to direct fiber-to-fiber connectors, the EB connectors greatly reduce sensitivity to lateral misalignment as well as sensitivity to dust contamination . The relatively large sizes of the fiber core and of the free-space beams allow using ray-tracing to calculate the coupling efficiency at each interface, as well as to predict the optical intensity distribution in the fiber core.
3.2 Experimental validation of the simulation model
To verify the accuracy of our simulation model we built a laboratory testbed that replicates the optical link in Fig. 2 and uses micro-positioning stages to introduce controlled alignment errors at the junctions. In this way we can measure the loss as a function of misalignments and compare it with the simulation results. We found that to improve the accuracy of the model we need to account for the specific radiant intensity profile of the VCSEL, which is shown in Fig. 3(a) . The measured radiant intensity profiles of two different VCSELs are plotted in Fig. 3(a) to show the significant variation among different manufacturers; our simulations and experiments presented in the following are based on the VCSEL with higher divergence (solid line in Fig. 3(a)) which represents the more conservative case because it generally results in higher coupling losses compared to a lower-divergence source. Our simulation also considers the detailed geometry of the active area of the photodiode: as shown in Fig. 3(b) the active area is circular, but partially obscured by the electrode. Figure 4 shows a comparison of simulation results and experimental data of the coupling efficiency sensitivity to lateral misalignment at either the transmitter or the receiver, showing that the agreement of simulation and experiment is within approximately 0.5 dB. To measure the transmitter-side sensitivity, the optical path includes the VCSEL, the photonic turn element and the first length of fiber, after which we measure the fiber-coupled power using a wide-area power meter. For the receiver measurement we added the photonic turn element at the output of the fiber, and we obtained the received optical power by measuring the photocurrent of the 60-µm diameter photodiode. In order to validate our simulation with fibers of various core diameters, we fabricated fibers with core diameter of 62.5, 80, and 100 μm. Figure 4(a) shows the comparison between the simulation and the experimental results of transmitter-side sensitivity for all three fibers. Figure 4(b) shows the comparison of receiver-side sensitivity for the 62.5-µm and 100-µm core diameter fibers. It is important to mention that the data of Fig. 4 was obtained by carefully aligning all the elements to the prescribed position, and only varying the lateral off-set of VCSEL or photodiode.
As we mentioned, with multimode fibers the coupling efficiency at each junction depends on the modal power distribution in the incoming fiber, which may depend on the misalignment at the upstream junctions. Therefore it is important that the ray tracing model accurately predicts not only the total power, but also the intensity distribution within the fiber at each junction. To confirm the accuracy of our model in this regard, we performed measurements of the intensity distribution using the apparatus shown in Fig. 5 . The VCSEL is mounted on a micro-positioning stage to introduce lateral off-set along x and y axes as indicated in the Fig. 5(a). At the output of the fiber, we projected an image of the fiber endface onto a CCD camera to measure the near-field intensity (shown in the inset). From the intensity profile we calculated the encircled flux curve, which gives the fraction of the total power within a circle of a given radius, ranging from zero up to the fiber core boundary. The experimental results for a fiber with 80 µm core diameter and NA = 0.29 for various VCSEL lateral off-sets are shown in Fig. 5(b), and compared with the ray-tracing model results. The agreement is better than 5% in the range studied. Having successfully validated our ray-tracing model in these representative test cases, we then proceeded to simulate the coupling efficiency for a range of core diameters and numerical apertures, which we discuss in the next section.
3.3 Coupling efficiency optimization
Various sources of misalignment may be present in manufacturing, and our goal is to design the link to be as robust (i.e., efficient) as possible. We give here a few examples of potential sources of misalignment. The passive optical components (photonic turn elements and expanded beam connectors) are fabricated based on low-cost plastic injection molding technology with a certain mechanical tolerance specification. Tight tolerance comes at the expense of yields and tooling cost and is therefore not desirable. Temperature and mechanical stresses also induce deformation from the prescribed geometries and operating distances. Passive-alignment “pick and place” processes are typically adopted to position the various components (VCSEL, photodiode, and photonic turn element), and once again relaxing the requirements on alignment accuracy enables a high throughput process and low cost pick and place machines. Fiber termination and assembly also involves positioning errors and a high volume, low cost manufacturing process that will benefit from a fiber design that is more tolerant to misalignments.
Using our simulation model we first investigated how the link efficiency in the presence of misalignments depends on the fiber core diameter and NA. In this first study, for simplicity, we assume all misalignments to be in the same plane, with the fixed values listed in Table 1 , and with their relative directions combined in such a way as to produce the worst case interactions (i.e. with either positive or negative off-sets relative to the nominal position combined as to maximize the coupling loss). With this combination of fixed off-sets, we calculated the total link loss for a range of fiber core diameters and NAs. This result is shown in Fig. 6 as a contour map showing the total link loss (including reflection losses) as a function of fiber core diameter and NA. A marked improvement of link loss occurs when the NA increases up to ~0.3 and the core diameter increases up to ~80 µm. Further increase in the core diameter or NA provide only a marginal advantage because the coupling improvements at the transmitter and in-line connectors become marginally better, while at the same time the coupling degradation when focusing the light onto the photodiode increases. The total link loss is approximately 6.2 dB for a fiber with 80 µm core diameter and NA = 0.3, compared to 11.5 dB for a standard multimode fiber with 50 µm core diameter and NA = 0.2. It is interesting to note that while in a perfectly aligned system it is sufficient to have the fiber numerical aperture approximately equal to the source numerical aperture (as transformed by the lenses in the photonic turn element), the presence of misalignments requires higher numerical aperture to maintain efficient coupling into the fiber. This can be understood considering that the fiber has a graded-index profile and as the incident beam is shifted towards the edge of the core due to misalignments; the local index contrast relative to the cladding is reduced compared to a launch at the center of the core.
The analysis presented above approximates a worst-case combination of a limited set of misalignments; to obtain a more realistic prediction one needs to consider how the various misalignments interact to affect the total link loss. We also expanded Table 1 to include 56 tolerance values corresponding to all dimensional parameters (such as lens curvature, conic constant, etc…) and possible misalignments (for example, longitudinal fiber to lens and lens to lens distances). To this aim, we created a Monte Carlo simulation that considers the statistical distribution of the misalignments and calculates the probability of the link loss exceeding a specified value. We assume that each variable representing a misalignment has a Gaussian distribution, with the 3-σ values given by Table 1 and all the other additional parameters. These values represent what we believe can be achieved with low-cost and high-throughput manufacturing processes involving the steps outlined above (plastic injection molding, pick and place positioning machines, fiber termination and assembly, temperature and mechanical stresses). The statistical simulation model is “three dimensional” in that it considers the misalignment in both planes, for both lateral misalignments (off-set) and angular misalignments (tilt). The result of the Monte Carlo simulation is shown in Fig. 6(b) as a cumulative distribution plot. We can see that 99.9% of the links (equivalent to a 1,000 ppm failure rate) has total link loss lower than 4 dB for a fiber with 80 µm core diameter and NA = 0.29, while the corresponding loss is 7.3 dB for a fiber with 62.5 µm core diameter and NA = 0.27. This result confirms the marked improvement gained by increasing the core size and numerical aperture. Based on this analysis, a fiber with 80 µm core diameter and NA = 0.29 is an appropriate choice for the link that we considered.
4. Mechanical reliability in small bend radius
As discussed in the introduction, bend insensitive fibers were first introduced for applications in FTTH, especially for installations within apartment buildings . In that application, installations are performed by trained professionals and routing the fiber cables is relatively controlled to bend diameters in the 10 to 20 mm range. For consumer interconnects, it is not expected that consumers will be trained to not severely bend the cable. Therefore the “tight bend” requirement is significantly more challenging (i.e. well below 10-20 mm diameter). For example, a temporary cable pinch is likely to occur sporadically - where a cable is completely folded onto itself for a short period of time (few seconds to minutes). In this condition, the optical fiber will experience a bend diameter comparable to the cable diameter itself (e.g. ~3 mm, depending on the cable design). One can also expect the fiber to be routed on desks or behind TV-sets, or even within devices for internal interconnects. In this later case, less aggressive bends are expected but with longer lifetime requirements (for example, 5-7 years is a typical design lifetime for consumer devices). Two implications to fiber design arise: (i) the fiber reliability must be evaluated to ensure survival in small bend diameter, and (ii) the fiber design should have negligible bend loss to ensure an operating link even while the cable is bent. For this study, we define a target of 7-years lifetime at 10 mm diameter bend for permanent deployment (such as fiber routing), and a minimum of 1-hour survival time at the most aggressive 3 mm diameter; simulating temporary bend such as an eventual cable pinch. In terms of bend loss, it is desirable that the bend loss does not exceed 1-2 dB in the most aggressive scenario (3 mm diameter pinch). These requirements have been set based on discussions with various manufacturers of consumer electronics devices. In this section we focus on the mechanical reliability of the fiber in bend conditions, and in the next section on designing the profile to achieve low bend loss.
To evaluate the reliability in tight bends, we simulated the cable pinch test by deploying the fiber in a two-point bend configuration  (2PB, inset of Fig. 7 ) and measuring the lifetime at various bend radii. The impact of different cable designs is also important, but is not discussed here. The lifetime of glass optical fiber is determined by fatigue growth of micro-flaws present on the fiber surface under a certain level of applied stress [13, 14]. Using the empirical power law for the dynamics of crack growth [15, 16], we can estimate the time to failure tf under an applied bend stress σa approximately by14]. In the dynamic fatigue test, the fiber is subjected to a stress that increases linearly with time at a rate, from zero up to the point of failure (σf is then the median failure strength obtained from dynamic fatigue at the corresponding rate). In the same dynamic fatigue testing, by scanning a wide range of stress rate it is possible to determine the crack growth resistance parameter n (or simply called fatigue parameter). The maximum applied stress σa in a 2PB configuration is [17, 18]19, 20]. Change in n has little impact on long term reliability at larger bend radii, however, for fiber experiencing transient very small (≤ 3 mm radius) bends, the increased fatigue resistance may substantially extend the lifetime of the fiber from minutes to days. For this application we study the lifetime under bend for a fiber with reduced glass diameter of 100 µm (instead of typical 125 µm) and an enhanced coating composition with a relatively high modulus value. Figure 7a shows a strength measurement at rate of 87 kpsi/s with median failure strength of 820 kpsi at 35°C and 90% relative humidity (the data is plotted in a typical Weibull fashion). As mentioned above, the fatigue parameter was measured using the same configuration and we obtained slightly higher values at 35°C and 90% relative humidity. Using Eqs. (1) and (2) and the measured parameters for failure strength and fatigue, we can obtain the lifetime prediction shown in Fig. 7(b) (blue curve). Also, shown is the lifetime for a reference fiber with 125 µm glass diameter and fatigue parameter of ~20 (red curve). At our target 3 mm bend diameter, we can see an increase of approximately 4 orders of magnitude in the lifetime, which significantly surpasses our minimum of 1 hour for transient bends. To confirm the predictions by Eq. (1), we directly measured the time to failure of a fiber statically deployed in 2PB. The results are also shown in Fig. 7(b), and are in good agreement with predicted lifetime.
5. Bend loss and bandwidth considerations
The requirements for consumer applications drive fiber design trade-offs, and for a multimode fiber, the design parameters that need to be optimized include the core delta, core diameter, core profile shape (step or graded index) and the fiber cladding diameter. In section 4, we demonstrated that a reduced clad fiber of 100 µm glass diameter meets the reliability requirements at tight bends, while in Section 3, our analysis indicates an optimum core size of 80 µm with an NA of 0.29 enables optimum coupling efficiency and relaxed alignment tolerances. The need for high data transfer rates requires the fiber to have a graded index core, since a step index core cannot achieve sufficiently high bandwidth and support >10 Gb/s over distances up to 100 meters. Finally, although the resulting cladding width is only 10 µm, we also found through modeling and experimentation that the addition of a low index trench in this narrow cladding  is beneficial for achieving both high bandwidth and low bend performance. In Fig. 8 , we show a schematic of the refractive index profile (a) and the respective measured bend loss (b) for three designs. Design A represents a regular 50 µm core diameter with parabolic profile and ~1% delta, design B adds low index trench at the cladding (delta around −0.4%) and in design C we further increase the core to 80 µm and delta to ~2%. By tuning the width and depth of the low-index trench in design C we were able to achieve loss in the order of 1 dB in a 3 mm diameter bend (this example has a trench with about −0.5% delta and about 5 µm width).
Bandwidth of multimode fibers depends not only on the refractive index profile design and manufacturing precision but also on the transmitter launch conditions, which determines the coupling coefficients on the various fiber modes. From a profile perspective, it is well known that the profile alpha parameter must be approximately 2.1 to achieve high bandwidth at 850 nm. The interplay with the low-index trench parameters (width, depth and position) must also be considered . We manufactured a number of fibers with design C of Fig. 8, and measured the Differential Mode Delay (DMD) profile with the source scanned across the core of the fiber in 0.2 μm steps. An example is shown in Fig. 9 with a well controlled DMD profile. The launch condition of the transmitter with relaxed alignment tolerances considered in this paper was already characterized in Fig. 5b. Using these results in combination with the measured DMD from Fig. 9, we calculated the expected Effective Modal Bandwidth  and obtained 1477, 1533, 1649, and 1580 MHz·km for the launch conditions in Fig. 5(b) corresponding to 0, 10, 15 and 20 µm off-sets, respectively. The exact reach that this bandwidth can support depends on a careful analysis of the overall system design (bit rate, jitter, laser linewidth, connector losses, allocated margin for dispersion penalty, rise and fall times of input pulses, detector bandwidth etc…). Assuming a requirement of 1 Hz/bps at the maximum reach, for a bit rate of 10 Gbps and distance of 50 meters the fiber bandwidth required would be 1 Hz/bps · 10 Gbps 50 m = 500 MHz·km. The results demonstrated in Fig. 9 are significantly above this threshold.
Another interesting aspect of the bend-insensitive design is that a fiber with moderate or high bandwidth (i.e., with relatively flat DMD profile) does not experience a bandwidth reduction when a tight bend is applied. In practice, there is often a slight increase in bandwidth due to outer modes being stripped out. To illustrate this we have measured DMD profile with a 3 mm bend applied at the launch and compared the results obtained with a straight fiber. The minimum Effective Modal Bandwidth for the launch conditions discussed here (0, 10, 15 and 20 µm off-sets) was 1355 MHz·km and a change of less than 1% was observed when the 3 mm bend was applied (which we consider within the measurement noise of our system).
6. Data transmission results
We performed digital data transmission experiments at 10 Gb/s using the setup shown in Fig. 10(a) ; the optical link contains the same components described in Section 3. The transmitter (“Tx” in Fig. 10(a)) consists of a 10 Gb/s, 850-nm VCSEL and the photonic turn element. A Bit Error Rate Tester (BERT) provides the modulation current to the VCSEL, and is connected through a bias-T in order to add the bias current generated by a DC current source. It should be noted that, unlike commercial VCSEL-based transmitters, our setup does not contain active electronics to shape the electrical pulse driving the VCSEL. This method of operating the VCSEL is not ideal in that there is no pre-emphasis applied to the drive current, but it has the benefit of allowing full control of both the modulation amplitude and the DC bias of the VCSEL, a feature which we exploited in our experiment as discussed later. The receiver (“Rx” in in Fig. 10(a)) consists of the photonic turn element and a 10 Gb/s photodiode with integrated transimpedance amplifier and limiting amplifier. The output of the limiting amplifier is connected to the error-detector input of the BERT. In general, one measures the Bit Error Ratio (BER) as a function of the received optical power by using a variable optical attenuator (VOA) to produce a controlled reduction of the optical power at the receiver. To properly perform such a measurement on multimode fibers, the VOA needs to be designed in a way that avoids introducing mode-dependent loss and mode-coupling. At the time of our experiments, such a “mode-neutral” VOA was not yet available for fibers with core diameter larger than 62.5 µm. To overcome this limitation, in our measurements we produced an effective reduction of the modulated optical power by reducing the AC modulation amplitude of the VCSEL, while maintaining a constant DC bias so that the VCSEL’s dynamic response was not affected. The optical power at the receiver reported in Fig. 10(b) is calculated (in linear units) as P0 · (V/Vmax) where P0 is the average optical power at the receiver, V is the modulation amplitude of the VCSEL (which was varied as part of the experiment) and Vmax is its value at the nominal operating point. It should be noted that this technique relies on operating the VCSEL within a linear region of its L-I characteristic, which we verified to be the case in our experiments. To calculate the power penalty introduced by the fiber under test we first measure the “back to back” BER with a short (approximately 1 m) length of fiber between the two connectors. We then insert the fiber under test by means of butt-coupled joints using micropositioning stages which we carefully aligned by maximizing the transmitted power. We estimate the power penalty from the comparison of the two BER curves as shown in Fig. 10(b), which shows a typical result for a 50-m long fiber with 682 MHz·km bandwidth, for which less than 0.5 dB of power penalty was observed. By introducing a controlled misalignment between the VCSEL and the photonic turn element we can create different launch conditions, i.e. different distribution of modal power in the fiber, and measure the change of power penalty with launch condition. Figure 10(c) shows the aggregated data that we obtained on 15 distinct fibers (all with 50 m length), each tested in two launch conditions, namely with perfect alignment and with a 15 µm lateral off-set. The horizontal axis in Fig. 10(c) is the calculated Effective Modal Bandwidth (EMBc), which once again we derived from a Differential-Mode-Delay (DMD) measurement on each fiber and the knowledge of the modal power distribution generated by our transmitter in each launch condition. For a 50 m link, a fiber with bandwidth of 232 MHz·km showed power penalty of at most ~1 dB. Fibers with higher bandwidth show progressively smaller power penalty: for bandwidth exceeding 800 MHz·km, the measured penalty was approximately 0.1 ± 0.1 dB, which we estimate to be the measurement repeatability of the power penalty with our experimental setup. Figure 10(c) shows a good correlation between the power penalty and the EMBc, proving that the EMBc metric is a useful predictor of system-level performance.
We discussed in detail a multimode fiber design optimized for short-reach consumer interconnects. A detailed analysis of coupling efficiency to enable ultra-low tolerance links using 10 Gb/s VCSELs and photodiode leads to the choice of core diameter of 80 µm and a numerical aperture of 0.29. A reduced glass diameter is implemented and confirmed to meet the lifetime requirements at 3 mm bend diameter, which are encountered in consumer cables. Our results also show that a trench-assisted refractive index profile ensures low bending loss at 3 mm diameter. Finally, we demonstrated fibers with bandwidth higher than 800 MHz·km, on which our system testing showed a power penalty smaller than 0.1 ± 0.1 dB at lengths up to 50 meters.
References and links
1. R. E. Wagner, J. R. Igel, R. Whitman, M. D. Vaughn, A. B. Ruffin, and S. Bickharn, “Fiber-based broadband-access deployment in the United States,” J. Lightwave Technol. 24(12), 4526–4540 (2006). [CrossRef]
2. S. Ten, “In home networking using optical fiber,” (Optical Society of America, 2012), NTh1D.4.
3. D. Z. Chen, W. R. Belben, J. B. Gallup, C. Mazzali, P. Dainese, and T. Rhyne, “Requirements for Bend Insensitive Fibers for Verizon's FiOS and FTTH Applications,” (Optical Society of America, 2008), p. NTuC2.
4. M. J. Li, P. Tandon, D. C. Bookbinder, S. R. Bickham, M. A. McDermott, R. B. Desorcie, D. A. Nolan, J. J. Johnson, K. A. Lewis, and J. J. Englebert, “Ultra-low Bending Loss Single-Mode Fiber for FTTH,” J. Lightwave Technol. 27(3), 376–382 (2009). [CrossRef]
5. M. J. Li, P. Tandon, D. C. Bookbinder, S. R. Bickham, K. A. Wilbert, J. S. Abbott, and D. A. Nolan, “Designs of bend-insensitive multimode fibers,” in Optical Fiber Communication Conference and Exposition (OFC/NFOEC), 2011 and the National Fiber Optic Engineers Conference (2011), 1–3.
6. O. Kogan, S. R. Bickham, M.-J. Li, P. Tandon, J. S. Abbott, and S. A. Garner, “Design and Characterization of Bend-Insensitive Multimode Fiber,” in 60th International Wire & Cable Symposium (IWCS) Conference (Charlotte Convention Center, Charlotte, North Carolina, USA 2011), 154–159.
7. B. Dunstan, “USB 3.0 Architecture Overview,” in SuperSpeed USB Developers Conference(2011).
8. Intel Corporation, “Thunderbolt Technology - Technology Brief,” (2012), http://www.intel.com/content/dam/doc/technology-brief/thunderbolt-technology-brief.pdf.
9. J. M. Trewhella, G. W. Johnson, W. K. Hogan, and D. L. Karst, “Evolution of optical subassemblies in IBM data communication transceivers,” IBM J. Res. Develop. 47(2.3), 251–258 (2003). [CrossRef]
10. T. Kibler, S. Poferl, G. Böck, H.-P. Huber, and E. Zeeb, “Optical Data Buses for Automotive Applications,” J. Lightwave Technol. 22(9), 2184–2199 (2004). [CrossRef]
12. M. J. Matthewson, C. R. Kurkjian, and S. T. Gulati, “Strength Measurement of Optical Fibers by Bending,” J. Am. Ceram. Soc. 69(11), 815–821 (1986). [CrossRef]
13. B. R. Lawn and T. R. Wilshaw, Fracture of Brittle Materials (Cambridge University Press, 1975).
14. S. T. Gulati, “Crack kinetics during static and dynamic loading,” J. Non-Cryst. Solids 38–39, 475–480 (1980). [CrossRef]
15. R. J. Charles, “Static fatigue of glass: I, II,” J. Appl. Phys. 29(12), 1657–1662 (1958). [CrossRef]
16. A. G. Evans and S. M. Wiederhorn, “Proof testing of ceramic materials. An analytical basis for failure predictions,” Int. J. Fract. 10(3), 379–392 (1974). [CrossRef]
17. G. S. Glaesemann, S. T. Gulati, and J. D. Helfinstine, “Effect of strain and surface composition on Young's modulus of optical fibers,” (Optical Society of America, 1988), TuG5.
18. G. S. Glaesemann, and S. T. Gulati, “Dynamic fatigue data for fatigue resistant fiber in tension vs bending,” (Optical Society of America, 1989), WA3.
19. R. Sugizaki, H. Inaba, K. Fuse, T. Nishimoto, and T. Yagi, “Small Diameter Fibers for Optical Interconnection and Their Reliability,” in Proceedings of the 57th International Wire & Cable Symposium(2008), 377–381.
20. M. Ohmura and K. Saito, “High-Density Optical Wiring Technologies for Optical Backplane Interconnection Using Downsized Fibers and Pre-Installed Fiber Type Multi Optical Connectors,” in Optical Fiber Communication Conference (OFC)(Optical Society of America, 2006), OWI71.
21. I. E. C. (IEC), “Optical fibres – Part 1-49: Measurement methods and test procedures – Differential Mode Delay”, IEC 60793–1-49:2006,” (2006).