Scalable optical switch with internal fiber network

Abstract

We present architecture of a large dynamic optical packet/burst switch comprising a number of smaller switching fabrics that are interconnected by internal fiber links. The switching fabrics are based on the broadcast-and-select architecture, while the internal interconnection network is a full mesh. An extensive performance study has been performed and results regarding scalability, packet/burst loss rate and achievable throughput for four scheduling algorithms are shown.

Appendix: Model of the gain-clamped SOA

Semiconductor optical amplifiers can be modelled using either single-section or multiple-section models. The gain and the phase-shift depend on the carrier density. For our purpose, we used a two-directional distributed time-domain model of SOA based of a modification of the transmission line laser modelling method [12]. The model divides device into n longitudinal sections. In each section of length ΔL, the carrier dynamics are described by the following rate equations for carrier density:

$$ \frac{dN_e}{dt}=\frac{J}{ed}-A N_e-B N_e^2-C N_e^3-\frac{v_g g_{nl}}{\Delta L} , $$

(8)

where J is the injection current density, e is the electron charge, d is the thickness of the active layer, v _g is the group velocity, N _e is the carrier density, S is the photon density, and A, B, and C are the monomolecular, bimolecular, and Auger recombination coefficients, respectively. Photon densities are obtained from field strengths in both longitudinal directions:

$$ S=|E_R|^2 + |E_L|^2 , $$

(9)

where E _R and E _L are fields travelling in the right and left direction, respectively. Thus, the model represents interactions of the local optical fields and the carrier density within a section, while the carrier density is assumed to be homogeneous within a section rather than along the whole device. This allows a consideration of inhomogenities along the length of the active region as well as an inclusion of spectral hole burning (SHB) effect in the model. The material gain of a section is governed by the carrier density within this section. It is calculated using the following expression:

$$ g_{nl}= a \varGamma \Delta L (N_e - N_{et}) \frac{S}{1+ \varepsilon S} , $$

(10)

where N _et is the carrier density at the transparency point (the point where g=0), and a is the gain cross section also known as the differential gain coefficient. Because the carrier density is assumed to be constant across a section, the unsaturated gain factor is also constant. The material gain nonlinearity induced phenomenologically by the gain compression parameter, ε and caused by the finite intraband relaxation times of the carriers implies a reduction of gain by 1/(1+εS). Thus, the gain saturates for high photon densities. The intraband processes include effects such as SHB and carrier heating, which do not affect the total carrier density, but lead to a change in the carrier distribution within the conduction and valence band. To describe the non-instantaneous nature of the gain nonlinearity, a time constant, t _nl , is used to filter the instantaneous photon density in the calculation of the nonlinear gain. The amplification of the field within a section is then given by:

$$ E_{R,L}^o (\Delta L)= E_{R,L}^i (0) \exp \bigl[G(1+i \alpha)\bigr] , $$

(11)

where G represents the field gain coefficient across the section, i.e., G=exp[(g _nl −a _s )/ΔL], a _s is the waveguide attenuation factor, and α is the linewidth enhancement factor. $E^{o}_{R,L}$ and $E^{i}_{R,L}$ are fields at the right and left inputs and the outputs of a model section, respectively. The phase shift between fields at the input and output of the amplifier in dependence of the carrier concentration induced refractive index change can be then calculated according to: Δφ=−Γαg _nl ΔL/2. A gain-clamped SOA is realized by placing two distributed Bragg reflectors (DBRs) at both sides of the active region, thereby achieving lasing in the SOA cavity at a wavelength different from the signal wavelength. These passive DBR sections were modelled using the same model as for the active region, but without the gain model. The index-coupling is represented by alternate low and high impedance subsections. In this manner, the modulation of the refractive index along the waveguide structure can be effectively modelled.