Introduction When you dial a number on your phone or press the “send” button for your email, you ultimately initiate a series of light pulses on an optical fiber, and the quality of the established communication link depends on both the strength and the shape of these pulses when they reach the receiving end. With technology advances, the pulses get shorter and shorter and the optical link length becomes longer. Deployed telecommunication systems are capable of sending very short optical pulses of less than 100ps over several hundred kilome- ters without conversion to an electronic form. These achievements come at a cost – new timescales and greater lengths lead to new impairments to the transmission quality, which arise from new and more subtle physical phenomena. In this paper we focus on the effect known as Polarization Mode Dispersion (PMD) – a potential hin- drance for Ultra-Long-Haul (greater than 1000km) high speed transmission systems. Ideally, the group velocity of polarized light pulses traveling along optical fibers should not depend on their polarization because fibers are, in essence, cylindrical waveguides. However, imperfections in the fiber manufacturing process, together with mechanical bends and stresses during cabling and installation, lead to small distortions in the otherwise perfect cylindrical shape. In other words, sections of the fiber have a small amount of birefringence – pulses of different polarizations have different group velocities – which leads to spreading of the pulses and dete- rioration of the transmitted signal. Because this birefringence is rather small, its effect had been negligible until recent advances required shorter optical pulses. The haphazard origins of the fiber birefringence makes it non-uniform along the fiber length. But a birefringence correlation length can be defined, and a fiber can be thought of as a large set of concatenated sections, each of constant birefringence. Obviously, the total PMD of a fiber should depend not only on the birefringence of each section but also on their relative orienta- tions, and thus PMD is not a simple scalar quantity. While the rig- orous mathematical formulation of PMD is beyond the scope of this paper, it will be necessary for our discussion to introduce some basic concepts. As recently reviewed [1], an elegant vectorial description of the problem has its basis in the Stokes Parameters and Poincare Sphere of classical optics, which provide a mapping of both the state of polarization and birefringence to a three dimensional space (called Stokes space). In this space, a frequency dependent birefringence operator ˆ β i (ω) transforms vectors from the beginning to the end of a section i of length L i by rotating them around a birefringence vec- tor  β i (ω) by angle β i (ω)L i . Also, in the same space, one can define a frequency dependent PMD vector  τ i (ω) = ∂(  β i (ω)L i ) ∂ω at the end of each fiber section. The length of this vector | τ i | is the Differential Group Delay (DGD) between the fast and the slow polarization modes in this section, and its direction has to do with orientation of the birefringence axis relative to the lab frame. Interestingly, evolu- tion of the state of polarization with frequency at the end of a given section i can be described as rotation about  τ i (ω) by angle  | τ i (ω)| ∂ω. Then a total PMD vector  τ  at the end of the fiber can be computed using a generalization of a straightforward concatena- tion rule [1]:  τ  = τ N + ˆ β N  τ N−1 + ˆ β N ˆ β N−1  τ N−2 + ... + ˆ β N ˆ β N−1 ... ˆ β 2  τ 1 . (1) In fact, this is just a pure summation of individual vectors  τ i after they are transformed by the concatenated rotations to the end of the fiber. Many properties of the total PMD vector  τ  are derived from this rule. The peculiarity of PMD, which makes it stand out from other impairments in optical transmission systems, is its ever-changing nature. Optical cables installed in the field are subject to all kinds of aging, relaxations, and environmental changes. These changes cause ˆ β i (ω) and  τ i (ω) to vary in time, which in turn result in a time-vary- ing total DGD of the link: | τ  (ω)|. A larger value of the DGD leads to a larger temporal spread of the information-carrying pulses and might cause a transmission system outage. Thus, knowledge of the PMD temporal dynamics (unsolved at the moment) is important for communications carriers as it would allow them to estimate the probability and duration of outage events. The first step in this direction would be to determine how many and which ˆ β i (ω) are changing in installed routes, and what implications these changes inflict on the total DGD of the route. In the absence of detailed knowledge, the conventional “retarder plate” model was put forward [2]. It treats all sections of the fiber as independent identically distributed random birefrin- gence plates, which change in time in a random fashion. In this case the total DGD of the fiber | τ  (ω)| is a random variable, which has a Maxwellian probability distribution function. In other words, for any optical frequency there exists a small but finite probability of having a very large value of DGD depending on the orientation of the independent vectors. Contrary to that, however, our field experiments established that the DGD of com- pletely buried fiber segments can be stable or “dead” for month- long timescales [3]. We have also shown that the DGD dynamics of installed telecom systems is determined largely by a limited number of “hinges” between these dead sections acting as polar- ization rotators [4]. These rotators can be either exposed sections of fiber such as bridge attachments (weakly responding to large diurnal changes in the ambient temperature) or some system com- ponents. In particular, we found that Dispersion Compensation Modules (for chromatic dispersion) responded strongly to small hourly temperature variations within the buildings that housed transmission equipment. 4 IEEE LEOS NEWSLETTER June 2004 Physical Mechanism for Polarization Mode Dispersion Temporal Dynamics Industry Research Highlights MISHA BRODSKY (1), MISHA BORODITSKY (1), PETER MAGILL (1), NICHOLAS J. FRIGO (1) AND MOSHE TUR (2) (1) AT&T LABS RESEARCH, 200 LAUREL AVE. S., MIDDLETOWN, NJ 07748 EMAIL: BRODSKY@RESEARCH.ATT.COM (2) MOSHE TUR FACULTY OF ENGINEERING, TEL AVIV UNIVERSITY, TEL AVIV, ISRAEL 69978