All-optical Networks Research Articles

Backpropagation through nonlinear units for the all-optical training of neural networks

We propose a practical scheme for end-to-end optical backpropagation in neural networks. Using saturable absorption for the nonlinear units, we find that the backward-propagating gradients required to train the network can be approximated in a surprisingly simple pump-probe scheme that requires only simple passive optical elements. Simulations show that, with readily obtainable optical depths, our approach can achieve equivalent performance to state-of-the-art computational networks on image classification benchmarks, even in deep networks with multiple sequential gradient approximation. With backpropagation through nonlinear units being an outstanding challenge to the field, this work provides a feasible path toward truly all-optical neural networks.

Nonlinear All-Optical Diffractive Deep Neural Network with 10.6 μm Wavelength for Image Classification

A photonic artificial intelligence chip is based on an optical neural network (ONN), low power consumption, low delay, and strong antiinterference ability. The all-optical diffractive deep neural network has recently demonstrated its inference capabilities on the image classification task. However, the size of the physical model does not have miniaturization and integration, and the optical nonlinearity is not incorporated into the diffraction neural network. By introducing the nonlinear characteristics of the network, complex tasks can be completed with high accuracy. In this study, a nonlinear all-optical diffraction deep neural network (N-D2NN) model based on 10.6 μm wavelength is constructed by combining the ONN and complex-valued neural networks with the nonlinear activation function introduced into the structure. To be specific, the improved activation function of the rectified linear unit (ReLU), i.e., Leaky-ReLU, parametric ReLU (PReLU), and randomized ReLU (RReLU), is selected as the activation function of the N-D2NN model. Through numerical simulation, it is proved that the N-D2NN model based on 10.6 μm wavelength has excellent representation ability, which enables them to perform classification learning tasks of the MNIST handwritten digital dataset and Fashion-MNIST dataset well, respectively. The results show that the N-D2NN model with the RReLU activation function has the highest classification accuracy of 97.86% and 89.28%, respectively. These results provide a theoretical basis for the preparation of miniaturized and integrated N-D2NN model photonic artificial intelligence chips.

Robust light beam diffractive shaping based on a kind of compact all-optical neural network.

A kind of compact all-optical learning-based neural network has been constructed and characterized for efficiently performing a robust layered diffractive shaping of laser beams. The data-driven control lightwave strategy demonstrates some particular advantages such as smart or intelligent light beam manipulation, optical data statistical inference and incident beam generalization. Based on the proposed method, several typical aberrated light fields can be effectively modulated into the desired fashion including the featured flat-top beams, an arrayed sub-beam arrangement and complex annular fringes compared with conventional GS-based DOEs. An actual THz laser is utilized to evaluate the effectiveness of the method developed.

Controlled unidirectional reflection in cold atoms via the spatial Kramers-Kronig relation.

It is known that the Kramers-Kronig (KK) relation between real and imaginary parts of the optical susceptibility in the frequency domain can also be realized in the space domain, as first proposed in [Nat. Photonics9(7), 436 (2015)10.1038/nphoton.2015.106]. We here study a mechanism to implement spatial KK relations in a cold atomic sample and use it to control unidirectional reflectionless for probe light incident from either the left or right side of the sample at will. In our model, the complex frequency dependent atomic susceptibility is mapped into a spatially dependent one, employing a far-detuned driving field of intensity linearly varied in space. The reflection of an incident light from one side of the sample can then be set to vanish over a specific frequency band directly by changing the driving field parameters, such as its intensity and frequency. Also, by incorporating the Bragg scattering into the spatial KK relation, the reflectivity from the opposite side of the sample, though typically small for realistic atomic densities, can be made to increase to improve the reflectivity contrast. The present scheme bears potentials for all-optical network applications that require controllable unidirectional light propagation.

Bending behavior and its effect on switching performance of an all-optical switch

The all-optical switch is one of the key elements of all-optical high speed information networks and computing systems. Their bending behavior, actuated by optical gradient forces, dramatically influences their switching performance and is therefore the research topic of this paper. The coupled behavior between optical gradient forces and the bending deflection is taken into account in the present model. The effects of the effective refractive index and the stretching induced by bending are included and their influences on the resonance wavelength shift of the signal light are discussed. Results show that the resonance wavelength shift increases almost linearly at first, and then increases suddenly as the circulating power increases. The resonance wavelength shift increases as the central angle of the suspended part increases, while it decreases with the initial gap. The present theoretical model may be helpful when designing all-optical switches.

All-optical core node modeling and optimization using hybrid wavelength conversion for QoS provision

All-optical networks (AON) provide larger bandwidth, faster switching, and QoS provision. High-cost node architecture with full range wavelength converters (FRWC) is used to meet application and service demands. In this paper, hybrid wavelength converters (HDWC) and wavelength range (WR) supervision are considered. Fixed (FXWC), limited (LRWC), and FRWC are combined in the different output ports resulting in a new core node architecture design. The design depends on a node topology location and its ability to reduce WR usage while guaranteeing the needed demands. In wavelength converter (WC) combination, the core node architecture is optimized by defining the number and type of WC in the available output ports using physical topology and lightpath cost. In the WR reduction, the available resource is managed to extend the lightpath availability and assure jointly traffic switching and provision according to its loss and blocking delay. Simulation work is conducted where the achieved results indicate that the core node architecture cost is effectively reduced, and the available resource utilization is improved while offering traffic requirements.

Two-beam coupling by a hot electron nonlinearity.

Transparent conductive oxides such as indium tin oxide (ITO) bear the potential to deliver efficient all-optical functionality due to their record-breaking optical nonlinearity at epsilon near zero (ENZ) wavelengths. All-optical applications generally involve more than one beam, but, to our knowledge, the coherent interaction between beams has not previously been discussed in these materials, which have a hot electron nonlinearity. Here we study the optical nonlinearity at ENZ in ITO and show that spatial and temporal interference has important consequences in a two-beam geometry. Our pump-probe results reveal a polarization-dependent transient that is explained by diffraction of pump light into the probe direction by a temperature grating produced by pump-probe interference. We further show that this effect allows tailoring the nonlinearity by tuning the frequency or chirp. Having fine control over the strong and ultrafast ENZ nonlinearity may enable applications in all-optical neural networks, nanophotonics, and spectroscopy.

Design and analysis of photonic crystal ring resonator based 6 × 6 wavelength router for photonic integrated circuits

A photonic crystal ring resonator-based 6 × 6 router has been designed and reported. The router is designed using silicon pillars with a refractive index of 3.47 perforated in the air background of refractive index 1 in a square lattice with a lattice constant of a = 562 nm. The router is designed to operate in the third optical window wavelength which has low loss and most widely used in optical communication systems and networks. Plane-wave expansion and finite difference time domain method has been used to obtain the bandgap and performance of the designed router, which exhibits acceptable performance such as low insertion loss, low propagation delay, and low crosstalk. These routers will find applications in the photonic integrated circuit that paves a path to all-optical networks.

On Nash Equilibria in Non-Cooperative All-Optical Networks

We consider the problem of determining a routing in all-optical networks, in which some couples of nodes want to communicate. In particular, we study this problem from the point of view of a network provider that has to design suitable payment functions for non-cooperative agents, corresponding to the couples of nodes wishing to communicate. The network provider aims at inducing stable routings (i.e., routings corresponding to Nash equilibria) using a low number of wavelengths. We consider three different kinds of local knowledge that agents may exploit to compute their payments, leading to three corresponding information levels. Under complete information, the network provider can design a payment function, inducing the agents to reach a Nash equilibrium mirroring any desired routing. If the price to an agent is computed only as a function of the wavelengths used along connecting paths (minimal level) or edges (intermediate level), the most reasonable functions either do not admit Nash equilibria or admit very inefficient ones, i.e., with the largest possible price of anarchy. However, by suitably restricting the network topology, a constant price of anarchy for chains and rings and a logarithmic one for trees can be obtained under the minimal and intermediate levels, respectively.

The design of a reconfigurable all-optical logic device based on cross-phase modulation in a highly nonlinear fiber

A simple reconfigurable ultrahigh-speed optical device which can be configured to optical NOT, AND, or XOR logic operation via slight variations in the input signals applied to two identical parallel highly nonlinear fibers (HNLFs) is described. The cross-phase modulation (XPM) in the HNLF and phase shifter are used in the design to realize a reconfigurable all-optical logic gate operating at 120 Gb/s. The proposed reconfigurable ultrahigh-speed all-optical logic device could play a crucial role in the realization of future all-optical high-performance flexible optical networks.

Joint Hybrid Wavelength Conversion and Fiber Delay Line for All-Optical Core Node Modeling and Optimization Using Dynamic QoS Support

All-optical networks (AON) assured a large bandwidth, high-speed switching, and QoS (Quality of Service) support. But due to the development of applications and services, new requirements are created in terms of, specifically resource utilization and QoS provision. To meet these demands, AON enabled core node architecture with multiple wavelength converters’ types involving full range wavelength converter (FRWC) having a significant cost. To reduce the cost, a hybrid AON with composite node architecture combining wavelength converters (WC), Fiber delay line (FDL), and wavelength range (WR) are introduced. In this architecture, fixed and limited wavelength converters (FXWC, LRWC) are utilized while the FRWC converter is replaced by FDL. The proposed architecture decreased the presence and usage of WC, FDL, and WR while assuring QoS support. A lightpath cost formulation is computed based on WR, WC, and FDL enabled in the proposed node architecture to decrease converter and FDL numbers. The cost also managed the wavelength availability to ensure lightpath control and QoS support using real-time requirements, especially loss and blocking delay. The obtained results in the comparative study, conducted in the simulation work, showed that the novel approach performed better than the existing approaches. Moreover, our proposal optimized AON resources, particularly node architecture design, while providing QoS support according to the traffic parameters and requirements.

WDM Network Issues: QOT Impairments, Dynamic Bandwidth Allocation, Modeling and Monitoring Schemes based on ML

Machine Learning (ML) has recently drawn the interest of both academics and professionals in the optical networking area to solve several problems. This trend has been triggered largely by the vast amount of available data (i.e., signal strength metrics, network alerts, etc.) and the large range of optimization criteria available in today's optical networks (such as modulation format, light path routes, transport wavelength, etc.). WDM systems are popular in the telecommunications industry. It increases network bandwidth without laying more cable. WDM technology in optical communication transports multiple optical carrier signals on a single fibre using various wavelengths of laser light. Wavelength division multiplexing (WDM) is an efficient strategy for leveraging the wide bandwidth of optical fibres to meet the rapid rise in demand for bandwidth on the Internet. WDM plays an essential role in both core and access networks in the communication network and faces various issues like its network performance is affected due to QoT impairments, Dynamic Bandwidth Allocation problem, Modeling, and Monitoring issues. Physical impairments are an issue in WDM and can cause calls to be blocked if the transmission quality is calculated in terms of minimum bit-error-rate. Dynamic allocation of bandwidth in WDM provides a crucial challenge for the effective and equal usage of the passive optical network upstream bandwidth when supporting the QoS specifications of various traffic groups. Fast attack monitoring can prevent the failure of vast volumes of data on all-optical networks. So, monitoring and modeling in a WDM network is also a critical issue. To solve these issues in WDM networks, more intelligence resources in scalable optical networks are necessary. ML techniques are demonstrating superiority in solving these issues. In this paper, a small detail about the WDM network is presented. Then, some WDM issues like QOT impairments, Dynamic Bandwidth Allocation, and modeling and monitoring are discussed, and then, in the next section focus on some ML techniques for solving these issues are presented.

Recovery of a single link failure in all-optical networks based on the cuckoo search algorithm

Recovery of a single link failure in all-optical networks based on the cuckoo search algorithm

Recovery of a single link failure in all-optical networks based on the cuckoo search algorithm

Recovery of a single link failure in all-optical networks based on the cuckoo search algorithm

Liquid crystal on silicon technology and reconfigurable all-optical networks

Liquid crystal on silicon technology and reconfigurable all-optical networks

Solving Rwa Problem in All-optical Networks Based on Optimized Ant Colony Algorithm

The information bandwidth growth can be achieved using optical networking technology. Optical fiber has been used as a physical means through many network technologies. Wavelength Division Multiplexing (WDM) has emerged as a promising technology to conquer the use of large-bandwidth in optical networks. In WDM network, the essential problem is to serve the load traffic represented by the requests that arrive at the network and need to be served. For serve the request, the network will determine a path paired with a specific wavelength used to send the request's data from source to the destination node. This problem is called the routing and wavelength assignment (RWA). In this work, Multi-Node Optimized Ant Colony Algorithm (M-NOACA) proposed and implemented to solve the RWA problem in all-optical networks by simulating traffic with the RWA algorithms. A performance measurements applied to study the optimization problem of the RWA, and apply a comparison with the classic algorithms. Determining the best collection of parameters for an ant-based algorithm to obtain the best performance remains an open issue, in this work, a study is done to introduce a new contribution to obtain the optimal setting of Ant Colony parameters used to solve the RWA problem.

All-optical Networks Performance Improvement by Solving The Rwa Problem Using Genetic Algorithm

All-optical Networks Performance Improvement by Solving The Rwa Problem Using Genetic Algorithm

Modeling of all-optical modulation format conversion from amplitude-shift keying to phase-shift keying using a vertical cavity quantum wells absorber

A theoretical model for all-optical modulation format conversion from amplitude-shift keying (ASK) to phase-shift keying (PSK) has been proposed using a reflective vertical cavity semiconductor saturable absorber (R-VCSSA). The R-VCSSA is a nonlinear device. The phase response of the device has been studied numerically with different input intensities. It has been observed from the simulated result that a maximum phase of ± π / 2 with the reflected signal could be obtained by properly adjusting the input intensity. The conversion of an intensity-modulated return-to-zero (RZ) signal to a binary phase-shift keying signal has been analyzed based upon the simulated result. The pump signal is considered to be an intensity-modulated RZ signal with wavelength fixed at the cavity resonance of the R-VCSSA device, and a cavity-detuned continuous-wave source is used as a probe signal to carry out this operation. This device could be employed in all-optical future networks as an all-optical nonlinear signal converter for ASK to PSK conversion.

Free-space optical delay line using space-time wave packets

An optical buffer featuring a large delay-bandwidth-product—a critical component for future all-optical communications networks—remains elusive. Central to its realization is a controllable inline optical delay line, previously accomplished via engineered dispersion in optical materials or photonic structures constrained by a low delay-bandwidth product. Here we show that space-time wave packets whose group velocity is continuously tunable in free space provide a versatile platform for constructing inline optical delay lines. By spatio-temporal spectral-phase-modulation, wave packets in the same or in different spectral windows that initially overlap in space and time subsequently separate by multiple pulse widths upon free propagation by virtue of their different group velocities. Delay-bandwidth products of ~100 for pulses of width ~1 ps are observed, with no fundamental limit on the system bandwidth.

Multifunctional integration on optical fiber tips: challenges and opportunities

The flat endface of an optical fiber tip is an emerging light-coupled microscopic platform that combines fiber optics with planar micro- and nanotechnologies. Since different materials and structures are integrated onto the endfaces, optical fiber tip devices have miniature sizes, diverse integrated functions, and low insertion losses, making them suitable for all-optical networks. In recent decades, the increasing demand for multifunctional optical fibers has created opportunities to develop various structures on fiber tips. Meanwhile, the unconventional shape of optical fibers presents challenges involving the adaptation of standard planar micro- and nanostructure preparation strategies for fiber tips. In this context, researchers are committed to exploring and optimizing fiber tip manufacturing techniques, thereby paving the way for future integrated all-fiber devices with multifunctional applications. First, we present a broad overview of current fabrication technologies, classified as “top-down,” “bottom-up,” and “material transfer” methods, for patterning optical fiber tips. Next, we review typical structures integrated on fiber tips and their known and potential applications, categorized with respect to functional structure configurations, including “optical functionalization” and “electrical integration.” Finally, we discuss the prospects for future opportunities involving multifunctional integrated fiber tips.