Raman pump design using machine-learning approaches

Abstract

To provide methods for Raman pump design using machine-learning approaches.SOLUTION: Training data corresponding to an operation of a Raman pump system includes input parameters specifying Raman pump parameters and channel launch power corresponding to respective transmission band channels, and an output parameter specifying a Raman pump gain profile of the transmission band channels of the Raman pump system. A neural network configured to output inferred input parameters for the Raman pump system given a specified Raman pump gain profile includes an auto-encoder having: an input layer with input nodes representing the Raman pump gain profile and the channel launch power; one or more intermediate layers with intermediate nodes representing the Raman pump parameters and the channel launch power; and an output layer with output nodes representing the Raman pump gain profile and the channel launch power.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates generally to Raman pump design using machine learning approaches. [Background technology]

[0002] An optical fiber is a fiber configured to transmit light from a first end of the fiber to a destination at a second end of the fiber. Optical fibers are typically made of glass, certain plastics, or other materials with a low index of refraction that allows light to be transmitted over the length of the optical fiber by refracting the light through the optical fiber while mitigating absorption of the light by the optical fiber itself. Due to the low attenuation of light transmitted through optical fibers and the high propagation speed of light, optical fibers can be a useful medium for transmitting information from a first location to a distant second location. Optical fibers can be bundled together as cables and used to quickly and accurately achieve communication and computer networking between many different locations. To facilitate communication over optical fibers, a laser can be emitted through a first end of a particular optical fiber to a receiving element at the second end of the particular optical fiber. The wavelength of the emitted laser, along with the pulse rate of the emitted laser, can be interpreted by the receiving element or an associated computer system to present information to a distant location associated with the second end of the particular optical fiber.

[0003] Raman pumps can be used to amplify the output power of optical signals used in optical fiber communications. Furthermore, Raman pumps can achieve optical transmission of photons having different wavelengths along the same optical fiber. Thus, Raman pumps can improve optical fiber communications by increasing the distance over which an optical fiber can communicate information and achieving multiplexed (i.e., multi-channel) communications along a single optical fiber using photons having different wavelengths. Raman pumps involve launching a first photon (a "pump photon") toward a second photon (a "Stokes photon"), where the pump photon is launched at a specific wavelength designated to excite the Stokes photon. By designating a specific wavelength for the pump photon, the Stokes photon is launched at a different wavelength along the optical fiber.

[0004] The subject matter claimed in this disclosure is not limited to embodiments that solve any disadvantages or that operate only in environments such as those described above. Rather, this background is only provided to illustrate one example technology area where some embodiments described in this disclosure may be practiced. Summary of the Invention

[0005] According to an aspect of an embodiment, a method can include generating training data corresponding to operation of a Raman pump for a distributed Raman amplifier. The training data can include input parameters specifying Raman pump parameters and channel launch powers corresponding to each transmission band channel, and output parameters specifying Raman pump gain profiles of the transmission band channels of the distributed Raman amplifier, where the transmission band can be a single band, such as a C-band, or multi-band including two or more bands, such as a C-band and an L-band. The method can include training a neural network to estimate predicted input parameters for the distributed Raman amplifier given a particular Raman pump gain profile. The neural network can include an autoencoder having an input layer with input nodes representing the Raman pump gain profile and channel launch powers, one or more intermediate layers with intermediate nodes representing the Raman pump parameters and channel launch powers, and an output layer with output nodes representing the Raman pump gain profile and channel launch powers.

[0006] The object and advantages of the embodiments will be realized and attained at least by the elements, features, and combinations particularly pointed out in the claims. It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are not restrictive of the invention as claimed. [Brief description of the drawings]

[0007] Exemplary embodiments will be described and explained with additional specificity and detail through the accompanying drawings in which:

[0008] [Figure 1] FIG. 1 illustrates an example operating environment for training a neural network to estimate Raman pump parameters given a target Raman pump gain profile in accordance with one or more embodiments of the present disclosure.

[0009] [Diagram 2] 1 illustrates an example autoencoder architecture of a neural network configured to estimate Raman pump parameters, in accordance with one or more embodiments of the present disclosure.

[0010] [Diagram 3] 1 shows an example graph including a simulated Raman pump gain profile based on Raman pump parameters determined by a neural network compared to an estimated Raman pump gain profile and a target Raman pump gain profile determined by a neural network, in accordance with one or more embodiments of the present disclosure.

[0011] [Figure 4] 1 is a flowchart of an example method for training a neural network to estimate Raman pump parameters in accordance with one or more embodiments of the present disclosure.

[0012] [Diagram 5] 1 is an exemplary computer system in accordance with one or more embodiments of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] A Raman pump is typically designed to emit pump photons that can excite a target Stokes photon to a wide range of output wavelengths by inducing a phenomenon known as stimulated Raman scattering in the target Stokes photon. The pump photon is typically emitted at a specific short wavelength relative to the Stokes photon, which can absorb the pump photon while traveling through the optical fiber and be re-emitted at a different, longer wavelength. Based on the emission wavelength of the pump photon, the refractive index of the material in which the Raman pump is operating, and the material associated with the Stokes photon, the re-emission wavelength of the Stokes photon can be identified. Thus, a Raman pump can emit a first photon at a shorter wavelength based on parameters associated with the Raman pump and a longer target output wavelength of the second photon, whereas a conventional optical laser can only emit photons having an output wavelength that corresponds to the emission spectrum of the material associated with the emitted photon.

[0014] The efficiency and effectiveness of a particular Raman pump can be evaluated by a Raman gain profile, which measures or calculates the Raman gain of the Raman pump. Raman gain can indicate the optical amplification resulting from stimulated Raman scattering of Stokes photons caused by pump photons, and Raman gain can depend on the frequency offset between the pump wave corresponding to the pump photons and the signal wave corresponding to the Stokes photons. Raman gain can also depend on other properties of the Raman pump, such as the pump wavelength and the material of the optical fiber.

[0015] In some circumstances, one or more Raman pumps may be implemented as distributed Raman amplifiers used in combination with other optical fiber systems, such as an Erbium-Doped Fiber Amplifier (EDFA), to increase the optical gain of the optical fiber system. Increasing the optical gain of an optical fiber system can improve the optical signal to noise ratio, or OSNR, because the additional Raman gain enhances the optical signal without introducing a lot of additional noise into the optical fiber system. As a result, the transmission range of the optical signal can be increased by using distributed Raman amplifiers.

[0016] However, when a distributed Raman amplifier is implemented using a Wavelength-Division Multiplexed (WDM) optical fiber containing multiple channels, the Raman gain may or may not be uniform across the multiple channels, resulting in some channels having higher gain and better performance than others. To flatten the Raman gain profile across multiple channels, multiple Raman pumps can be implemented using WDM optical fiber. To improve the uniformity of the Raman gain profile, each Raman pump implemented using WDM optical fiber can include a carefully specified Raman pump wavelength and pump power. The design of Raman pumps in WDM optical fiber with multi-band transmission can be further complicated because additional stimulated Raman scattering can occur between different band channels, such as between C-band channels, L-band channels, S-band channels, U-band channels, or any other transmission band channels, causing interference effects between two or more Raman pumps contained in the WDM optical fiber.

[0017] In some cases, the Raman pumps can be arranged in a cascade configuration with two or more Raman pumps. A first Raman pump (i.e., a second-order or other higher-order Raman pump) in such a cascade configuration can emit a first photon having a first wavelength, which transfers energy to a second photon (i.e., emitted by a first-order or other lower-order Raman pump) having a second wavelength longer than the first wavelength. The energy provided by the first photon from the higher-order Raman pump to the second photon from the lower-order Raman pump can increase the Raman gain of the subsequent photon due to the propagation of energy from the first photon to the second photon, resulting in less attenuation in the optical fiber and a higher OSNR.

[0018] Additionally, distributed Raman amplifiers can be implemented with optical fiber systems as forward amplifiers, backward amplifiers, or forward / backward amplifiers, where a forward amplifier has the optical signal and the Raman pump propagating in the same direction, and a backward amplifier has the optical signal and the Raman pump propagating in opposite directions.

[0019] Modeling distributed Raman amplifiers, especially those with higher-order Raman pumps, using computer simulations and machine learning approaches may improve the efficiency of Raman pump designs. However, machine learning approaches for designing distributed Raman amplifiers may be inappropriate. The output value of a distributed Raman amplifier is often represented by a Raman pump gain profile that indicates the optical gain of a particular distributed Raman amplifier. Different distributed Raman amplifier configurations may generate the same or similar Raman pump gain profiles, which creates a one-to-many mapping problem between input Raman pump parameters and output Raman pump gain profiles that is difficult to solve with existing neural network architectures. In other words, multiple distributed Raman amplifier configurations with different Raman pump parameters, such as Raman pump wavelengths and pump power configurations, may generate the same or similar Raman pump gain profiles. For example, currently, inverse neural network designs are not trained to accurately ascertain which characteristics or parameters associated with a particular distributed Raman amplifier contribute to the generation of the corresponding Raman pump gain profile.

[0020] Furthermore, existing machine learning models for designing distributed Raman amplifiers typically only design the distributed Raman amplifier based on a small number of Raman pump parameters. For example, existing machine learning models may only control the Raman pump power or the wavelength configuration of the distributed Raman amplifier to generate a target Raman pump gain profile. Existing machine learning models for designing distributed Raman amplifiers do not specify parameters that correspond to specific channels included in the distributed Raman amplifier, such as channel launch power.

[0021] The present disclosure generally relates to training a neural network to design a Raman pump for a distributed Raman amplifier in a multi-band transmission system. In some embodiments, a neural network architecture described in accordance with one or more embodiments of the present disclosure can be used to determine weights associated with input parameters that may or may not affect a Raman pump gain profile. Additionally or alternatively, the neural network architecture of the present disclosure can provide for training the neural network to consider a wider range of input parameters, including channel launch power and propagation direction, which improves the calculations performed by the neural network in designing a distributed Raman amplifier by providing a more accurate machine learning model configured to consider and analyze a wider range of input variables.

[0022] Embodiments of the present disclosure will now be described with reference to the accompanying drawings.

[0023] 1 is a diagram of an example operating environment 100 for training a neural network to estimate Raman pump parameters given a target Raman pump gain profile, in accordance with one or more embodiments of the present disclosure. The environment 100 can include a neural network training module 120 configured to obtain training data 110 and determine node weights 125 corresponding to a trained neural network. The node weights 125 can be used, implemented, or otherwise referenced by a neural network 140 configured to obtain a Raman pump gain profile 130 and estimate Raman pump parameters 150 that may produce the obtained Raman pump gain profile 130.

[0024] In some embodiments, the neural network training module 120 and / or the neural network 140 (collectively referred to herein as "computing modules") may include code and routines configured to enable a computing system to perform one or more operations. Additionally or alternatively, one or more of the computing modules may be implemented using hardware, including a processor, a microprocessor (e.g., performing or controlling the execution of one or more operations), a field-programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). In some other examples, the computing modules may be implemented using a combination of hardware and software. In this disclosure, operations described as being performed by a computing module may include operations that the computing module may instruct one or more corresponding systems to perform. The computing module may be configured to perform a series of operations on the training data 110, the Raman pump gain profile 130, and / or the Raman pump parameters 150, as described in more detail below in connection with an example method 400 described with respect to FIG. 4.

[0025] In some embodiments, the training data 110 can describe parameters associated with the operation of a distributed Raman amplifier. For example, the training data 110 can include a specified power for a Raman pump or a wavelength of the Raman pump. As an additional or alternative example, a multi-channel distributed Raman amplifier, such as that used in a WDM optical fiber system, can include individual launch powers for various channels included in the distributed Raman amplifier, such that the training data 110 can include channel launch powers. In these and other embodiments, the training data 110 can include input parameters associated with the distributed Raman amplifier. In other words, the training data 110 can include parameters that can be set (e.g., by a design or implementation user) for the distributed Raman amplifier. Based on the input parameters included in the training data 110, the output parameters of the distributed Raman amplifier can be determined. For example, a particular set of input parameters can include Raman pump power, Raman pump wavelength, channel launch power, and channel wavelength specifications, along with Raman gain and attenuation coefficients of the optical fiber. A projected Raman pump gain profile associated with the distributed Raman amplifier can be generated by solving coupled differential equations based on the particular set of input parameters.

[0026] In some embodiments, the training data 110 may be randomly generated. For example, the Raman pump power, pump wavelength, and one or more channel launch powers may be randomly specified to generate a random Raman pump design. In these and other embodiments, each of the randomly generated input parameters may generate a corresponding randomly generated Raman pump gain profile. Randomizing the training data 110 may allow the neural network training module 120 to train with less bias than user-generated Raman pump designs due to user preferences or predispositions. Additionally or alternatively, randomized training data 110 may allow the neural network training module 120 to train with a wider range of Raman pump parameters, allowing more distributed Raman amplifier configurations to be considered and specified by the neural network training module 120.

[0027] The neural network training module 120 can obtain the training data 110 and generate node weights 125 indicative of the importance of different input parameters in determining the Raman pump gain profile. In some embodiments, the neural network training module 120 can be configured to generate the node weights 125 via a back-propagation training process that includes determining weights that quantify the importance of different variables represented by intermediate nodes in a particular neural network in calculating the Raman pump gain profile represented by an output node in the particular neural network. The neural network training module 120 can evaluate the difference between an observed Raman pump gain profile corresponding to an actual implementation of a distributed Raman amplifier based on the input parameters specified in the training data 110, or a simulation of a distributed Raman amplifier based on the input parameters, and a Raman pump gain profile calculated according to a differential equation based on the input parameters.

[0028] In some embodiments, the difference between the observed and calculated Raman pump gain profiles can be represented as a loss value that represents the accuracy of the calculated Raman pump gain profile. For example, the loss value can be calculated by summing or otherwise combining the mean squared errors between the observed and calculated Raman pump gain profiles across some or all of the channels included in the distributed Raman amplifier, with a particular mean squared error between the observed and calculated Raman pump gain profiles for a particular channel representing the mean squared error of the calculated Raman pump gain profile. If the calculation of the Raman pump gain profile is deemed too inaccurate (e.g., based on a loss value that exceeds a certain threshold), one or more weights associated with the intermediate nodes that represent the input parameters can be adjusted to reduce the loss values ​​of the calculated and observed Raman pump gain profiles. In these and other embodiments, the neural network training module 120 can be configured to iteratively compare the observed Raman pump gain profile to the calculated Raman pump gain profile and iteratively generate updated weight values ​​associated with the intermediate nodes until the loss value is below the threshold.

[0029] The neural network training module 120 may perform subsequent backpropagation to determine weights corresponding to one or more input nodes of the neural network that represent input parameters related to the distributed Raman amplifier. In some embodiments, the neural network training module 120 may fix weights corresponding to intermediate nodes during subsequent backpropagation and iteratively calculate loss values ​​corresponding to the input nodes based on differences between observed output values ​​(e.g., observed Raman pump gain profiles) and calculated output values ​​(e.g., calculated Raman pump gain profiles with fixed intermediate node weights).

[0030] In these and other embodiments, the neural network training module 120 can be configured to determine weights for the input nodes, the intermediate nodes, and the output nodes arranged according to an autoencoder architecture. FIG. 2 illustrates an example autoencoder architecture 210 of a neural network 200 configured to estimate Raman pump parameters, according to one or more embodiments of the present disclosure. The autoencoder architecture 210 can include a concatenation of an encoder 230 that can be configured for an inverse design of a distributed Raman amplifier and a decoder 220 that can be configured for forward modeling of a distributed Raman amplifier. The autoencoder architecture 210 including the encoder 230 concatenated with the decoder 220 can realize the neural network 200 that performs the inverse design of a distributed Raman amplifier by estimating a set of input parameters that can be used to generate a particular target Raman pump gain profile. Additionally or alternatively, the concatenation of the encoder 230 and the decoder 220 can realize the neural network 200 that performs the forward modeling of a distributed Raman amplifier by calculating an estimated Raman pump gain profile given a particular set of input parameters.

[0031] In some embodiments, the neural network 200 may include an input layer 240 including input nodes, one or more intermediate layers 242 including intermediate nodes, and an output layer 244 including output nodes. In some embodiments, the input nodes may correspond to expected input variables related to the design of a distributed Raman amplifier, including an input node 251 representing a target Raman pump gain profile, and an input node 252 representing a target parameter of a channel launch power in the C-band, L-band, U-band, S-band, or any other transmission band wavelength range of the distributed Raman amplifier. The intermediate nodes may be considered as output nodes of the encoder 230 or input nodes of the decoder 220, and may include, for example, an intermediate node 254 representing a target parameter of a channel launch power in the C-band and L-band wavelength range of the distributed Raman amplifier in the intermediate layer 242, and an intermediate node 253 representing a Raman pump parameter. The output nodes may include an output node 255 representing a target Raman pump gain profile at the output layer 244 level, and an output node 256 representing a target parameter of a channel launch power at the output layer 244 level.

[0032] In some embodiments, the output layer 244 may reproduce the information provided at the input layer 240 by having output nodes 255 and 256 reproduce the input nodes 251 and 252, respectively. In other words, the encoder 230 may be configured to generate values ​​corresponding to the hidden nodes 253 and 254 at the hidden layer 242 based on the provided values ​​corresponding to the input nodes 251 and 252 at the input layer 240. The decoder 220 may be configured to take the values ​​of the hidden nodes 253 and 254 and reproduce the provided values ​​of the input nodes 251 and 252 as values ​​of the output nodes 255 and 256 at the output layer 244 to verify the estimated values ​​of the hidden nodes 253 and 254.

[0033] In these and other embodiments, by including input nodes 252, intermediate nodes 254, and output nodes 256 to represent target parameters of channel launch power in the C-band and L-band wavelength ranges of the distributed Raman amplifier in each layer of the autoencoder architecture 210, a neural network 200 can be realized that considers the launch power of specific C-band and L-band channels in the design of the distributed Raman amplifier. Existing neural networks can omit the channel launch parameters in the input layer, the output layer, or both the input layer and the output layer, resulting in the channel launch power not being accurately or properly specified according to the target function of the distributed Raman amplifier being designed. Existing neural networks that do not consider the channel launch parameters in the input layer or the output layer may not be able to provide a practical channel launch power, since different distributed Raman amplifier configurations with different channel launch parameters may result in the same Raman pump gain profile. For example, the channel launch power may be set to a low value to reduce or eliminate the effect of the channel launch power on the designed distributed Raman amplifier, resulting in the designed distributed Raman amplifier failing to meet one or more target performance expectations. By expressing the channel launch power parameters as input node 252, intermediate node 254, and output node 256, the channel launch power of any C-band or L-band channel contained in the designed distributed Raman amplifier can be more accurately specified.

[0034] Returning to the discussion of Figure 1, the node weights 125 established by the neural network training module 120 may be implemented in a neural network 140 that is the same as or similar to the neural network 200 described in connection with Figure 2. For example, the neural network 140 may include an autoencoder architecture as described with respect to the neural network 200, and the node weights 125 may correspond to the weights of nodes included in the input layer 240, the hidden layer 242, and the output layer 244. In this manner, the neural network 140 may be configured to predictively calculate a Raman pump gain profile given Raman pump input parameters, and vice versa.

[0035] The neural network 140 can be configured to obtain the Raman pump gain profile 130 and output the Raman pump input parameters 150. Additionally or alternatively, the neural network 140 can be configured to obtain the Raman pump input parameters 150 and output the Raman pump gain profile 130. In some embodiments, a distributed Raman amplifier design can be evaluated against the output of the neural network 140 (e.g., the Raman pump gain profile 130 or the Raman pump input parameters 150). For example, a particular distributed Raman amplifier designed based on the Raman pump input parameters specified by the neural network 140 can be evaluated by comparing a Raman pump gain profile calculated based on the particular Raman pump input parameters to the target Raman pump gain profile, such as by calculating a loss value between the calculated Raman pump gain profile and the target Raman pump gain profile. In response to the calculated Raman pump gain profile differing from the target Raman pump gain profile (e.g., the comparison includes a loss value greater than a threshold), the node weights 125 used by the neural network 140 may be updated, such as by the neural network training module 120.

[0036] Modifications, additions, or omissions may be made to environment 100 without departing from the scope of the present disclosure. For example, the designations of different elements as described are meant to aid in explaining the concepts described herein and are not limiting. For example, in some embodiments, training data 110, Raman pump gain profile 130, and / or Raman pump parameters 150 are depicted in the particular manner described to aid in explaining the concepts described herein, but such depictions are not limiting. Additionally, environment 100 may include any number of other elements or be implemented in systems or environments other than those described.

[0037] 3 illustrates an example graph 300 including a simulated Raman pump gain profile 330 based on Raman pump parameters determined by a neural network compared to an estimated Raman pump gain profile 320 and a target Raman pump gain profile 310 determined by a neural network in accordance with one or more embodiments 310 of the present disclosure. The target Raman pump gain profile 310 may indicate a target gain level at which a distributed Raman amplifier corresponding to the target Raman pump gain profile 310 should be set. As shown in the graph 300, the target Raman pump gain profile 310 indicates that the distributed Raman amplifier should be set to 10 decibels (dB).

[0038] The estimated Raman pump gain profile 320 may be provided by a neural network, such as neural network 140 of FIG. 1 or neural network 200 of FIG. 2, that estimates Raman pump input parameters according to the target Raman pump gain profile 310. As shown in graph 300, for example, the estimated Raman pump gain profile 320 corresponding to the target Raman pump gain profile 310 may include an output gain across multiple channels that is set to about 10 dB on average. In some embodiments, the estimated Raman pump gain profile 320 may be generated by running a neural network.

[0039] Additionally or alternatively, graph 300 illustrates a fine-tuned Raman pump gain profile 330 observed by simulating a distributed Raman amplifier according to Raman pump input parameters estimated by a neural network. Fine-tuned Raman pump gain profile 330 can be compared to estimated Raman pump gain profile 320 to determine whether the Raman pump input parameters generating estimated Raman pump gain profile 320 and fine-tuned Raman pump gain profile 330 are the same or similar within a certain threshold. As shown in graph 300, estimated Raman pump gain profile 320 was generated by a neural network trained in accordance with one or more embodiments of the present disclosure, and fine-tuned Raman pump gain profile 330 includes a largely identical or similar Raman pump gain profile, indicating that the neural network generated Raman pump input parameters corresponding to target Raman pump gain profile 310.

[0040] 4 is a flowchart of an example method 400 of training a neural network to estimate Raman pump parameters, in accordance with one or more embodiments of the present disclosure. Method 400 may be performed by any suitable system, apparatus, or device. For example, training data 110, Raman pump gain profile 130, and / or Raman pump parameters 150 of FIG. 1 may perform one or more operations associated with method 400. Although illustrated by separate blocks, steps and operations associated with one or more of the blocks of method 400 may be divided into additional blocks, combined into fewer blocks, or eliminated, depending on the particular implementation.

[0041] Method 400 begins at block 402, where training data corresponding to operation of a distributed Raman amplifier is generated. In some embodiments, the training data may include training input parameters that specify Raman pump parameters, such as signal wavelengths or Raman pump powers, and channel launch powers corresponding to each C-band or L-band channel included in the distributed Raman amplifier. Additionally or alternatively, the training data may include training output parameters that specify Raman pump gain profiles of the C-band and L-band channels of the distributed Raman amplifier. In some embodiments, the training data may include randomly generated Raman pump parameters, and the Raman pump gain profiles may be calculated based on the randomly generated Raman pump parameters by solving coupled differential equations related to the Raman pump parameters.

[0042] In some embodiments, the training data set can be generated in a testbed that includes a Raman pump and a multi-band channel transmission link. In the testbed, random Raman pump parameters can be set to a particular range, such as a range of parameter values ​​that represent typical Raman pump operation. In these and other embodiments, the range of parameter values ​​included in the particular range can be set based on constraints associated with the corresponding Raman pump. For example, the Raman pump wavelength or pump power parameters can be set to a limited range of possible values ​​to account for manufacturing issues associated with a particular Raman pump. As an additional or alternative example, the Raman pump parameters can be limited to avoid optical fiber damage due to excessive power input. Based on the randomly set Raman pump parameters from the testbed, a Raman pump gain profile can be measured. The setting and random selection of possible Raman pump parameters can be repeated a certain number of times to generate the training data set.

[0043] At block 404, the training data can be used to train a neural network such that the neural network is configured to output estimated input parameters for a distributed Raman amplifier given a particular Raman pump gain profile. Additionally or alternatively, the neural network can be trained to output a Raman pump gain profile given a particular set of Raman pump parameters.

[0044] In some embodiments, the neural network may include an autoencoder having an input layer with input nodes representing the Raman pump gain profile and the channel launch power, one or more hidden layers with hidden nodes representing the Raman pump parameters and the channel launch power, and an output layer with output nodes representing the Raman pump gain profile and the channel launch power. The autoencoder may include an encoder between the input layer and the hidden layer, and a decoder between the hidden layer and the output layer. Training the autoencoder may include using a backpropagation method, in which decoding weights associated with the hidden layer and output layer nodes associated with the decoder are set based on a particular Raman pump gain profile, and encoding weights associated with the input layer and hidden layer nodes associated with the encoder are set based on the nodes included in the hidden layer and their respective decoding weights.

[0045] At block 406, the trained neural network may obtain output parameters indicative of a target Raman pump gain profile.

[0046] In block 408, the trained neural network may output estimated Raman pump parameters such that the distributed Raman amplifier exhibits behavior corresponding to a target Raman pump gain profile as indicated by the obtained output parameters.

[0047] In some embodiments, the trained neural network may be improved, fine-tuned, or otherwise modified. A first Raman pump gain profile corresponding to a simulated distributed Raman amplifier with estimated Raman pump parameters may be compared to a second Raman pump gain profile output by the neural network with the estimated Raman pump parameters. The comparison between the first and second Raman pump gain profiles may be based on a mean square error between the two Raman pump gain profiles, which may be quantitatively expressed as a loss value. In some embodiments, in response to the loss value being greater than a certain threshold, weights associated with nodes of the neural network, such as intermediate nodes representing Raman pump parameters and channel transmit powers of the decoder, may be fixed, while Raman pump input parameters obtained by the encoder may be fine-tuned to update the second Raman pump gain profile to more closely resemble the first Raman pump gain profile.

[0048] Modifications, additions, or omissions may be made to method 500 without departing from the scope of the present disclosure. For example, the designation of different elements as described is meant to help explain the concepts described herein, and is not limiting. Additionally, method 500 may include any number of other elements or may be implemented in systems or environments other than those described.

[0049] 5 is an exemplary computer system 500 in accordance with one or more embodiments of the present disclosure. Computing system 500 may include a processor 510, a memory 520, a data storage device 530, and / or a communication unit 540, all of which may be communicatively coupled. Any or all of the environment 100 of FIG. 1 may be implemented as a computing system consistent with computing system 500.

[0050] In general, the processor 510 may include any suitable special purpose or general purpose computer, computing entity, or processing device, including various computer hardware or software modules, and may be configured to execute instructions stored on any suitable computer-readable storage medium. For example, the processor 510 may include a microprocessor, a microcontroller, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or any other digital or analog circuitry configured to interpret and / or execute program instructions and / or process data.

[0051] 5, it is understood that the processor 1010 may include any number of processors distributed across any number of networks or physical locations and configured to individually or collectively perform any number of operations described in this disclosure. In some embodiments, the processor 510 may interpret and / or execute program instructions and / or process data stored in the memory 520, the data storage device 530, or the memory 520 and the data storage device 530. In some embodiments, the processor 510 may fetch program instructions from the data storage device 530 and load program instructions into the memory 520.

[0052] After the program instructions are loaded into memory 520, the processor 510 may execute the program instructions, such as instructions that cause the computing system 500 to perform the operations of the method 500 of Figure 5. For example, the computing system 500 may execute the program instructions to generate a set of training data, train a neural network using the set of training data, obtain output parameters indicative of a desired Raman pump operation, and / or output estimated input parameters that result in a desired Raman pump operation.

[0053] The memory 520 and the data storage 530 may include a computer-readable storage medium or one or more computer-readable storage media having computer-executable instructions or data structures stored thereon. Such computer-readable storage media may be any commercially available media that can be accessed by a general purpose or special purpose computer, such as the processor 510. For example, the memory 520 and / or the data storage 530 may include the high-level training data 110, the Raman pump gain profile 130, and / or the Raman pump parameters 150 of FIG. 1. In some embodiments, the computing system 500 may or may not include either the memory 520 and the data storage 530.

[0054] By way of example, and not limitation, such computer-readable storage media may include non-transitory computer-readable media including random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), compact disk read-only memory (CD-ROM) or other optical disk storage, magnetic disk storage or other magnetic storage, flash memory devices (e.g., solid-state memory devices), or any other storage medium that can be used to store desired program code in the form of computer-executable instructions or data structures and that can be accessed by a general-purpose or special-purpose computer. Combinations of the above may also be included within the scope of computer-readable storage media. Computer-executable instructions may include, for example, instructions and data configured to cause processor 510 to perform a particular operation or group of operations.

[0055] The communication unit 540 may include any component, device, system, or combination thereof configured to transmit or receive information over a network. In some embodiments, the communication unit 540 may communicate with other components at other locations, co-located devices, or in the same system. For example, the communication unit 540 may include a modem, a network card (wireless or wired), an optical communication device, an infrared communication device, a wireless communication device (e.g., an antenna), and / or a chipset (e.g., a Bluetooth device, an 802.6 device (e.g., a metropolitan area network (MAN)), a WiFi device, a WiMAX device, a cellular communication facility, etc.), etc. The communication unit 540 may enable data exchange with a network and / or any other device or system described in this disclosure. For example, the communication unit 540 may enable the system 500 to communicate with other systems, such as communication devices and / or other networks.

[0056] Those skilled in the art, after reviewing the present disclosure, may appreciate that modifications, additions, or omissions may be made to system 500 without departing from the scope of the present disclosure. For example, system 500 may include more or fewer components than explicitly shown and described.

[0057] The foregoing disclosure is not intended to limit the invention to the precise forms or particular fields of use disclosed. Thus, various alternative embodiments and / or modifications to the present disclosure, whether or not expressly described or shown herein, are contemplated in light of the present disclosure. Thus, having described embodiments of the present disclosure, it is understood that changes may be made in form and detail without departing from the scope of the present disclosure. Thus, the present disclosure is limited only by the claims.

[0058] In some embodiments, different components, modules, engines, and services than those described herein may be implemented as objects or processes (e.g., separate threads) executing on a computing system. Although some of the systems and processes described herein have been generally described as being implemented in software (stored on and / or executed by general-purpose hardware), dedicated hardware implementations or a combination of software and dedicated hardware implementations are also possible and contemplated.

[0059] The terms used in this disclosure, and particularly in the appended claims (the body of the appended claims), are generally to be construed as "broad terms" (e.g., the term "including" should be construed as "including but not limited to").

[0060] Moreover, where the recitation of a particular number of introduced claims is intended, such intent is expressly set forth in the claim, and in the absence of such recitation, no such intent exists. For example, as an aid to understanding, the following appended claims may include the use of the introductory phrases "at least one" and "one or more" to introduce the recitation of claims. However, the use of such phrases should not be construed to mean that the introduction of a claim recitation with the indefinite article "a" or "an" limits any particular claim that includes such an introduced recitation of claims to an embodiment that includes only one such recitation, even when the same claim includes the introductory phrase "one or more" or "at least one" and the indefinite article "a" or "an" (e.g., "a" and / or "an" should be interpreted to mean "at least one" or "one or more"). In other words, the same applies to the use of definite articles used to introduce a recitation of claims.

[0061] Moreover, where the recitation of a particular number of incorporated claims is explicitly recited, one of ordinary skill in the art will understand that such recitation should be interpreted to mean at least the number recited (e.g., a recitation of "two enumerations" without other qualification means at least two enumerations, or an enumeration of two or more). Furthermore, in instances where a recitation similar to "at least one of A, B, and C, etc." or "one or more of A, B, and C, etc." is used, such configurations are typically intended to include A alone, B alone, C alone, A and B together, A and C together, B and C together, or A, B, and C together, etc.

[0062] Furthermore, any disjunctive word or phrase expressing two or more alternative terms, whether in the description, claims, or drawings, should be understood to contemplate the possibility of including one of the terms, either one of the terms, or both terms. For example, the phrase "A or B" should be understood to include the possibilities of "A" or "B" or "A and B."

[0063] All examples and conditional language described in this disclosure are intended for instructional purposes to assist the reader in understanding the disclosure and the concepts that the disclosure contributes to further developing the art, and should not be construed as being limited to such specifically recited examples and conditions. Although the embodiments of the present disclosure have been described in detail, various modifications, substitutions, and alternatives may be made thereto without departing from the spirit and scope of the present disclosure.

[0064] In addition to the above embodiments, the following supplementary notes are further disclosed. (Supplementary Note 1) A method comprising: generating training data corresponding to operation of the Raman pump system, said training data comprising: training input parameters specifying Raman pump parameters and one or more channel launch powers corresponding to each transmission band channel; training output parameters that specify a Raman pump gain profile for the transmission band channels of the Raman pump system; training the neural network using the training data such that the neural network is configured to output estimated input parameters for the Raman pump system given a specified Raman pump gain profile; The method includes: (Supplementary Note 2) The neural network includes an autoencoder, and the autoencoder an input layer having input nodes representing the Raman pump gain profiles and the channel launch powers; one or more intermediate layers having intermediate nodes representing the Raman pump parameters and the channel launch powers; an output layer having output nodes representing the Raman pump gain profile and the channel launch power; The method of claim 1, comprising: (Supplementary Note 3) The autoencoder of the neural network includes an encoding function between the input layer and the hidden layer, and a decoding function between the hidden layer and the output layer, and is trained using a backpropagation method, the backpropagation method comprising: setting the decoding function and decoding weights associated with the one or more hidden layers based on the Raman pump gain profile of the output layer; setting the coding function and coding weights associated with the one or more hidden layers based on the decoding weights; The method of claim 2, comprising: (Supplementary Note 4) The method of Supplementary Note 2, further comprising a step of comparing a first Raman pump gain profile corresponding to a simulated Raman pump system having the estimated input parameters output by the neural network with a second Raman pump gain profile output by the neural network together with the estimated input parameters, wherein comparing the first Raman pump gain profile to the second Raman pump gain profile is based on a mean square error between the first Raman pump gain profile and the second Raman pump gain profile. (Supplementary Note 5) The method of Supplementary Note 4, further comprising a step of adjusting a weight associated with the input node, the intermediate node, or the output node in response to the first Raman pump gain profile and the second Raman pump gain profile differing by a threshold value. (Supplementary Note 6) Obtaining output parameters indicative of a target Raman pump gain profile by the trained neural network; outputting, by the trained neural network, the estimated input parameters that result in a Raman pump system exhibiting behavior corresponding to the target Raman pump gain profile indicated by the obtained output parameters; 2. The method of claim 1, further comprising: (Supplementary Note 7) The method of Supplementary Note 1, wherein the Raman pump parameters include a Raman pump wavelength and a Raman pump power. (Supplementary Note 8) The training data is randomly generated; 2. The method of claim 1, wherein the Raman pump gain profile is calculated by solving simultaneous differential equations related to input parameters. (Supplementary Note 9) The training data is generated in a testbed; The test bed comprises: generating random values ​​for the training input parameters; measuring the Raman pump gain profile based on the Raman pump system being designed using the random values; 2. The method of claim 1, wherein the test bed repeatedly generates random values ​​for the training input parameters and measures the Raman pump gain profile a specific number of times. (Supplementary Note 10) A method comprising: obtaining, by the neural network, output parameters indicative of a target Raman pump gain profile corresponding to the Raman pump system; outputting estimated input parameters by the neural network that will result in a Raman pump system exhibiting behavior corresponding to a target Raman pump gain profile indicated by the obtained output parameters, the estimated input parameters including Raman pump parameters and one or more channel launch powers corresponding to each transmission band channel of the Raman pump system; The method includes: (Supplementary Note 11) The neural network includes an autoencoder, and the autoencoder includes: an input layer having input nodes representing Raman pump gain profiles and the channel launch powers; one or more intermediate layers having intermediate nodes representing the Raman pump parameters and the channel launch powers; an output layer having output nodes representing the Raman pump gain profile and the channel launch power; 11. The method of claim 10, comprising: (Supplementary Note 12) The autoencoder of the neural network includes an encoding function between the input layer and the hidden layer, and a decoding function between the hidden layer and the output layer, and is trained using a backpropagation method, the backpropagation method comprising: setting the decoding function and decoding weights associated with the one or more hidden layers based on the Raman pump gain profile of the output layer; setting the coding function and coding weights associated with the one or more hidden layers based on the decoding weights; 12. The method of claim 11, comprising: (Supplementary Note 13) The method of Supplementary Note 11, further comprising a step of comparing a first Raman pump gain profile corresponding to a simulated Raman pump system having the estimated input parameters output by the neural network with a second Raman pump gain profile output by the neural network together with the estimated input parameters, wherein comparing the first Raman pump gain profile to the second Raman pump gain profile is based on a mean square error between the first Raman pump gain profile and the second Raman pump gain profile. (Supplementary Note 14) The method of Supplementary Note 13, further comprising a step of adjusting a weight associated with the input node, the intermediate node, or the output node in response to the first Raman pump gain profile and the second Raman pump gain profile differing by a threshold value. 15. The method of claim 10, wherein the Raman pump parameters include a Raman pump wavelength and a Raman pump power. (Supplementary Note 16) A system comprising: 1. A neural network configured to output estimated input parameters for a Raman pump system given a particular Raman pump gain profile, the neural network including an autoencoder, the autoencoder comprising: an input layer having input nodes representing the Raman pump gain profiles and channel launch powers; one or more intermediate layers having intermediate nodes representing Raman pump parameters and said channel launch powers; an output layer having output nodes representing the Raman pump gain profile and the channel launch power; A neural network including: one or more processors; one or more non-transitory computer-readable storage media configured to store instructions that, in response to being executed, cause the system to perform an operation; and the operation includes: generating training data corresponding to operation of the Raman pump system, the training data comprising: training input parameters specifying Raman pump parameters and one or more channel launch powers corresponding to the transmission band channels; training output parameters that specify Raman pump gain profiles of transmission band channels of the Raman pump system; training the neural network with the training data such that, given a particular Raman pump gain profile, the neural network is configured to output estimated input parameters for the Raman pump system; Including, the system. (Supplementary Note 17) The autoencoder of the neural network includes an encoding function between the input layer and the hidden layer, and a decoding function between the hidden layer and the output layer, and is trained using a backpropagation method, the backpropagation method comprising: setting the decoding function and decoding weights associated with the one or more hidden layers based on the specified Raman pump gain profile of the output layer; setting the coding function and coding weights associated with the one or more hidden layers based on the decoding weights; 17. The system of claim 16, comprising: (Supplementary Note 18) The system described in Supplementary Note 17, further comprising a step of comparing a first Raman pump gain profile corresponding to a simulated Raman pump system having the estimated input parameters output by the neural network with a second Raman pump gain profile output by the neural network together with the estimated input parameters, wherein comparing the first Raman pump gain profile with the second Raman pump gain profile is based on a mean square error between the first Raman pump gain profile and the second Raman pump gain profile. (Supplementary Note 19) The system of Supplementary Note 18, further comprising a step of adjusting a weight associated with the input node, the intermediate node, or the output node in response to the first Raman pump gain profile and the second Raman pump gain profile differing by a threshold value. (Supplementary Note 20) The system of Supplementary Note 16, wherein the Raman pump parameters include a Raman pump wavelength and a Raman pump power. (Supplementary Note 21) The training data is randomly generated, 17. The system of claim 16, wherein the Raman pump gain profile is calculated by solving a system of differential equations relating input parameters. [Explanation of symbols]

[0065] 110 Training Data 120 Neural Network Training Module 125 node weights 130 Raman pump gain profile 140 Neural Networks 150 Raman pump input parameters

Claims

1. 1. A method comprising: generating training data corresponding to operation of the Raman pump system, said training data comprising: training input parameters specifying Raman pump parameters and one or more channel launch powers corresponding to each transmission band channel; training output parameters that specify a Raman pump gain profile for the transmission band channels of the Raman pump system; training the neural network using the training data such that the neural network is configured to output estimated input parameters for the Raman pump system given a specified Raman pump gain profile; The method includes:

2. The neural network includes an autoencoder, the autoencoder comprising: an input layer having input nodes representing the Raman pump gain profiles and the channel launch powers; one or more intermediate layers having intermediate nodes representing the Raman pump parameters and the channel launch powers; an output layer having output nodes representing the Raman pump gain profile and the channel launch power; The method of claim 1 , comprising:

3. The autoencoder of the neural network includes an encoding function between the input layer and the hidden layer, and a decoding function between the hidden layer and the output layer, and is trained using a backpropagation method, the backpropagation method comprising: setting the decoding function and decoding weights associated with the one or more hidden layers based on the Raman pump gain profile of the output layer; setting the coding function and coding weights associated with the one or more hidden layers based on the decoding weights; The method of claim 2 , comprising:

4. 3. The method of claim 2, further comprising: comparing a first Raman pump gain profile corresponding to the simulated Raman pump system having the estimated input parameters output by the neural network with a second Raman pump gain profile output by the neural network with the estimated input parameters, wherein comparing the first Raman pump gain profile to the second Raman pump gain profile is based on a mean square error between the first Raman pump gain profile and the second Raman pump gain profile.

5. 5. The method of claim 4, further comprising adjusting a weight associated with the input node, the intermediate node, or the output node in response to the first Raman pump gain profile and the second Raman pump gain profile differing by a threshold value.

6. obtaining, by the trained neural network, output parameters indicative of a target Raman pump gain profile; outputting, by the trained neural network, the estimated input parameters that result in a Raman pump system exhibiting behavior corresponding to the target Raman pump gain profile indicated by the obtained output parameters; The method of claim 1 further comprising:

7. The method of claim 1 , wherein the Raman pump parameters include a Raman pump wavelength and a Raman pump power.

8. The training data is randomly generated; The method of claim 1 , wherein the Raman pump gain profile is calculated by solving simultaneous differential equations related to input parameters.

9. The training data is generated in a testbed; The test bed comprises: generating random values ​​for the training input parameters; measuring the Raman pump gain profile based on the Raman pump system being designed using the random values; The method of claim 1 , wherein the test bed repeatedly generates random values ​​for the training input parameters and measures the Raman pump gain profile a specific number of times.

10. 1. A method comprising: obtaining, by the neural network, output parameters indicative of a target Raman pump gain profile corresponding to the Raman pump system; outputting estimated input parameters by the neural network that will result in a Raman pump system exhibiting behavior corresponding to a target Raman pump gain profile indicated by the obtained output parameters, the estimated input parameters including Raman pump parameters and one or more channel launch powers corresponding to each transmission band channel of the Raman pump system; The method includes:

11. The neural network includes an autoencoder, the autoencoder comprising: an input layer having input nodes representing Raman pump gain profiles and the channel launch powers; one or more intermediate layers having intermediate nodes representing the Raman pump parameters and the channel launch powers; an output layer having output nodes representing the Raman pump gain profile and the channel launch power; The method of claim 10, comprising:

12. The autoencoder of the neural network includes an encoding function between the input layer and the hidden layer, and a decoding function between the hidden layer and the output layer, and is trained using a backpropagation method, the backpropagation method comprising: setting the decoding function and decoding weights associated with the one or more hidden layers based on the Raman pump gain profile of the output layer; setting the coding function and coding weights associated with the one or more hidden layers based on the decoding weights; The method of claim 11 , comprising:

13. 12. The method of claim 11, further comprising: comparing a first Raman pump gain profile corresponding to the simulated Raman pump system having the estimated input parameters output by the neural network to a second Raman pump gain profile output by the neural network with the estimated input parameters, wherein comparing the first Raman pump gain profile to the second Raman pump gain profile is based on a mean square error between the first Raman pump gain profile and the second Raman pump gain profile.

14. 14. The method of claim 13, further comprising adjusting a weight associated with the input node, the intermediate node, or the output node in response to the first Raman pump gain profile and the second Raman pump gain profile differing by a threshold value.

15. The method of claim 10 , wherein the Raman pump parameters include a Raman pump wavelength and a Raman pump power.

16. 1. A system comprising:

1. A neural network configured to output estimated input parameters for a Raman pump system given a particular Raman pump gain profile, the neural network including an autoencoder, the autoencoder comprising: an input layer having input nodes representing the Raman pump gain profiles and channel launch powers; one or more intermediate layers having intermediate nodes representing Raman pump parameters and said channel launch powers; an output layer having output nodes representing the Raman pump gain profile and the channel launch power; A neural network including: one or more processors; one or more non-transitory computer-readable storage media configured to store instructions that, in response to being executed, cause the system to perform an operation; and the operation includes: generating training data corresponding to operation of the Raman pump system, the training data comprising: training input parameters specifying Raman pump parameters and one or more channel launch powers corresponding to the transmission band channels; training output parameters that specify Raman pump gain profiles of transmission band channels of the Raman pump system; training the neural network with the training data such that, given a particular Raman pump gain profile, the neural network is configured to output estimated input parameters for the Raman pump system; Including, the system.

17. The autoencoder of the neural network includes an encoding function between the input layer and the hidden layer, and a decoding function between the hidden layer and the output layer, and is trained using a backpropagation method, the backpropagation method comprising: setting the decoding function and decoding weights associated with the one or more hidden layers based on the specified Raman pump gain profile of the output layer; setting the coding function and coding weights associated with the one or more hidden layers based on the decoding weights; The system of claim 16, comprising:

18. 20. The system of claim 17, further comprising: comparing a first Raman pump gain profile corresponding to a simulated Raman pump system having the estimated input parameters output by the neural network to a second Raman pump gain profile output by the neural network with the estimated input parameters, wherein comparing the first Raman pump gain profile to the second Raman pump gain profile is based on a mean square error between the first Raman pump gain profile and the second Raman pump gain profile.

19. 20. The system of claim 18, further comprising: adjusting a weight associated with the input node, the intermediate node, or the output node in response to the first Raman pump gain profile and the second Raman pump gain profile differing by a threshold value.

20. The system of claim 16 , wherein the Raman pump parameters include a Raman pump wavelength and a Raman pump power.

Images