As bit rates increase to 10 Gbit/s and higher in fiberoptic communication systems, polarization-related impairments in various system components have become increasingly acute. Such impairments include polarization-mode dispersion (PMD) in optical fibers, polarization-dependent loss (PDL) in passive optical components, polarization-dependent modulation (PDM) in electro-optic modulators, and polarization-dependent gain (PDG) in optical amplifiers. The time-varying characteristic of these impairments makes correcting them particularly difficult.
When chromatic dispersion and fiber nonlinearity impairments are successfully managed, polarization-mode dispersion is often cited as the next critical hurdle for high bit-rate transmission systems (10 Gbit/s and higher). To better understand polarization-mode dispersion, consider a light pulse passing through a retardation plate (or wave plate) having a slow and a fast axis. Upon entering the retardation plate, the pulse is decomposed into two polarization components along the two axes. Because the two components travel with different speed in the retardation plate, they exit the plate with a relative time delay called a differential group delay (see Fig. 1). When the differential group delay (DGD) is comparable with the bit separation of a data stream, the bit-error rate may significantly increase.
A fiber link can be modeled as a cascaded series of randomly oriented retardation plates. In the absence of polarization-dependent loss or polarization-dependent gain in the fiber link, these retardation plates are optically equivalent to a single retardation plate with an effective differential group delay and two effective orthogonal principal axes (either linear or circular) for a given optical frequency. Unlike a true retardation plate, the differential group delay and the effective principal axes of the fiber link are wavelength dependent and fluctuate in time as a result of temperature variations and mechanical stresses. Consequently, the corresponding pulse undergoes random broadening, both as a function of wavelength at a given moment in time and as a function of time at a given wavelength.
As a rule of thumb, the differential group delay value should not exceed 14% of the bit duration to ensure an outage probability of less than five minutes per year at a 3-dB power penalty. Applying this rule of thumb results in a 14-ps DGD for a 10-Gbit/s system and a 3.5-ps DGD for a 40-Gbit/s system. Applying the DGD requirements to installed links with distances exceeding 300 km results in a finding that 20% of currently installed fiber links are unsuitable for 10-Gbit/s transmission and 75% of the links are unsuitable for 40-Gbit/s transmission. Therefore, polarization-mode dispersion compensation is required to use these fiber links at the higher bit rates.
Unlike chromatic dispersion and fiber nonlinearity effects, which are deterministic and stable in time, PMD-induced penalties can be totally absent at a given moment and adversely large several days later, causing an unacceptable bit-error rate for no apparent reason. To ensure an acceptable outage probability for a fiberoptic system, PMD compensation must be dynamic in nature and adaptive to random time variations. Various schemes for mitigating the effect of polarization-mode dispersion are available (see Fig. 2). They generally contain three key components: a dynamic polarization controller, a PMD analyzer, and a feedback control circuit. For some schemes, a dynamic variable delay line may also be required.
Optical components exhibiting polarization-dependent loss act like partial polarizers having two orthogonal axes (either linear or circular). A light signal experiences a maximum loss when its state of polarization is aligned with one axis and a minimum loss when the state of polarization is aligned with the other axis. The polarization-dependent loss is defined as the difference between the maximum-insertion loss and the minimum-insertion loss in dB.
In a fiber link that contains many optical components with different PDL values, the total loss value fluctuates between a maximum and a minimum value depending on the polarization of the light signal transmitted in the link. When polarization-dependent loss is present in a fiber link, it also complicates polarization-mode dispersion compensation because the link no longer behaves as a single retardation plate. Instead the link is equivalent to two retardation plates with a partial polarizer sandwiched in between. Any PMD compensation scheme, therefore, must account for the partial polarizer, significantly increasing the complexity of the compensation arrangement.
Similar to the PMD loss, the polarization-dependent loss effect in a fiber link containing multiple PDL components separated by sections of single-mode fiber is also time dependent. At one point in time, the polarization states of different sections of the link may be favorably oriented such that the total link loss is low to enable an acceptable bit-error rate. At a different point in time, however, external thermal or mechanical stress on the fiber may cause the polarization states in different fiber sections to rearrange, resulting in link losses that are too high to achieve quality transmission.
To combat the time-varying polarization-dependent loss problem, dynamic polarization controllers may be used at selected locations in the fiber link (see Fig. 3). Each polarization controller ensures that the polarization of light passing through each PDL component is set to minimize losses.
Stronger polarization components saturate an optical amplifier and thus undergo less amplifier gain than weaker polarization components. The gain difference results in polarization-dependent gain. This polarization hole burning produces polarization fluctuations in a fiber laser system, resulting in mode hopping and increased super-mode noise in a modelocked laser.
In a fiberoptic link that includes multiple optical amplifiers and multiple components that exhibit polarization-dependent loss, the cumulative PDG effects are time dependent and can be significant at one point in time and negligible at another point in time. When a large number of optical amplifiers are cascaded in a long-haul fiber link, the performance degradation caused by polarization-dependent gain is significant, even though each amplifier may have a very small PDG (~0.1 dB). The performance degradation becomes even more severe when polarization-dependent gain effects are combined with the polarization-mode dispersion and polarization-dependent loss of the fiber and other components in the link. Polarization scrambling at a frequency above an amplifier`s response rate (inverse of its upper energy level lifetime, ~500 Hz for an erbium-doped fiber amplifier or EDFA) has proven to be effective in mitigating polarization-dependent gain impairment in long-haul systems (see Fig. 3c). A doubling in system Q -factor was demonstrated in an 8100-km link containing 181 EDFAs with such a scheme.
In addition to polarization-dependent loss, external modulators, such as LiNbO3-based electro-optical modulators and semiconductor electro-absorption modulators, also exhibit polarization-dependent modulation effects. These effects cause each polarization state to have a different modulation depth. As a result, the amplitude of the received data bits varies when the polarization state changes because of temperature fluctuations or other external strains on the fiber. The changing amplitude produces bit-error-rate fluctuations.
To assist in polarization alignment, many titanium-indiffused LiNbO3 modulators include a polarizer at an input or output of the waveguide and thus convert the polarization-dependent modulation problem into a more easily identified PDL problem. The LiNbO3 modulators made by a proton exchange process act like a polarizer itself without the embedded polarizer. Thus, one method of eliminating polarization-dependent modulation effects uses a fast-response dynamic-polarization controller placed in front of the modulator to ensure that light passing through the modulator has minimal loss.
Dynamic polarization controllers
Because polarization-induced penalties are time dependent, polarization-impairment mitigation must be dynamic and adaptive to random time variations. A dynamic polarization controller (DPC) is the single most important element for overcoming these impairments.
Fast speed, low polarization-dependent loss, low insertion loss, and low activation loss are all critical parameters in evaluating a DPC. Activation-induced loss measures the additional insertion loss caused by activation of the device and is defined as the difference of the maximum and minimum insertion losses of the device, considering all possible activation conditions. This specification is particularly important because all polarization-impairment compensation schemes utilize a feedback signal to activate the polarization controller. The activation-induced loss causes errors in the feedback signal and directly degrades the performance of the compensation apparatus. In addition, when an instrument for measuring the polarization-dependent loss of optical components includes a polarization controller, the activation-induced loss limits the resolution and accuracy of the measurement. Similarly, the polarization controller`s polarization-dependent loss also contributes to errors in a feedback system and complicates the design of compensation hardware and software.
Dynamic polarization controllers currently on the market include free-space retardation-plate-based and lithium niobate waveguide-based devices. The free-space devices contain multiple (sometimes three) retardation plates with different relative orientations. Applying a voltage to each retardation plate changes its retardation and hence the polarization of light passing through the plate. Because light exiting the fiber has to be collimated, passed through the plates, and focused back into the fiber when making such devices, the resulting labor costs, material costs, and insertion losses are high. The retardation plates may be made of liquid crystal or solid-state electro-optical materials. Liquid crystal devices also suffer from low speed (10 to 100 ms), narrow operating temperature ranges, and high polarization dependent loss.
Lithium niobate-based polarization controllers contain multiple waveguide sections. Each waveguide section has a different electrode (or crystal) orientation. Similar to retardation plate-based devices, each waveguide functions as an electrically variable retardation plate. Any polarization state can be generated by adjustment of the voltages applied to different waveguide sections. Lithium niobate waveguides, however, also have several drawbacks including high insertion loss (~4 dB), high polarization-dependent loss (~0.2 dB), low return loss (45 dB), high cost, and low power-handling capacity (50 mW).
Similar to other polarization controllers, a fiber-squeezer-based dynamic polarization controller/scrambler can be modeled as a cascaded series of retardation plates, each retardation plate having a corresponding orientation (see Fig. 4). Each retardation plate is generated by squeezing a section of fiber with a piezoelectric actuator. Varying the voltage applied to the piezoelectric actuator changes the retardation of the fiber section or "plate." By controlling the voltages on different fiber sections, any desired polarization state can be generated from any arbitrary input.
Because of its all-fiber nature, fiber-squeezer controllers have negligible insertion loss, no back reflection, and no polarization-dependent loss. With a response time less than 35 µs-approximately 100 times faster than a typical liquid-crystal-based controller-the device is capable of tracking fast polarization variations that may be caused by events such as a train passing by or by ocean waves in transoceanic fiber trunks.
Activation-induced losses in its fiber-squeezer controller have been reduced to less than 0.002 dB. The low activation-induced loss makes these components ideal for use in high-precision PDL instruments and in feedback loops for compensating for polarization-induced penalties. Another feature of the device is that its performance is wavelength independent. The device functions equally well for signals ranging from 1280 nm to 1650 nm. This one-device-fits-all feature simplifies system design, lowers implementation cost, and enables channel expandability of the system.
Implementation of fiber squeezers is also cost effective. Half-wave voltage requirements of the fiber-squeezer controllers have been reduced to less than 40 Vdc. The low voltage requirements allow the use of readily available low-cost electronics to drive and control the fiber-squeezer controller. In testing, reliability of the fiber squeezers has been shown in more than 10 billion activation cycles at half-wave voltages without any failure.
An additional application of the fiber-squeezer controller is as a polarization scrambler to effectively randomize polarization states. With a built-in resonant-enhanced circuit, the half-wave voltages of the device at scrambling frequencies (60 kHz, 100 kHz, and 140 kHz) are reduced to a few volts. Such a low voltage requirement makes the driving electronics simple and low cost. With properly selected driving parameters, the scrambler has a polarization sensitivity of less than 0.05 dB and a degree of polarization less than 1%.
1. T. Ono and Y. Yano, Optics and Photon. News, 61, May 2000.
2. H. Bulow, OFC 2000 Proceedings, ThH1-1, 110.
3. D. A. Watley et al., OFC 2000 Proceedings, ThB6-1, 37.5
4. Z. Pan et al., OFC 2000 Proceedings, ThH2-1, 113.
Steve Yao is chief technical officer with General Photonics, 13766 Arapahoe Place, Chino, CA 91710. Contact him at 909-590-5473 or firstname.lastname@example.org.
Figure 1. Real fiber is like many retardation plates in series with different orientations and birefringence. It is equivalent to a single retardation plate having slow and fast axes, and an effective birefringence. An optical pulse broadens because the two polarization components travel with different speeds.
Figure 2. Various polarization-mode dispersion mitigation schemes include a) principal state transmission method, b) multiple retardation plates method, c) optical delay line method, d) electrical delay line method, and e) chirped HiBi fiber grating method. In all schemes, dynamic polarization controllers require fast speed, low polarization loss, and low activation loss.
Figure 3. Dynamic polarization controllers are used for a) polarization stabilization, b) polarization-loss compensation, and c) polarization-dependent gain mitigation.
Figure 4. Fiber-squeezer dynamic polarization controllers have the advantages of no insertion loss, no back reflection, no polarization-dispersion loss, extremely low activation loss, and high speed.