Signal noise reduction system, noise reduction method, and noise reduction program

Abstract

To provide a noise removal system, a signal processing system, a noise removal method, and a noise removal program for solving at least one of a first problem, a second problem and a third problem, where the first problem is that a distortion compensation function learnt using gathered signal data is inaccurate, the second problem is that the distortion compensation function having been learnt is suitable only in an operation environment in which signal data are gathered, and the third problem is that a large amount data are required to accurately learn the distortion compensation function.SOLUTION: A noise removal system includes distortion discrimination means and signal correction means. The distortion discrimination means discriminates distortion of an input signal from an optical receiver in a teacher-less method and then outputs a distortion parameter showing the discriminated distortion. The signal correction means uses the output distortion parameter to correct a signal input to an optical transmitter.SELECTED DRAWING: Figure 4

Description

Technical Field

[0001] The present invention relates to a signal noise removal system, a signal processing system, a noise removal method, and a noise removal program for distortion compensation, particularly for distortion compensation in optical communication.

Background Art

[0002] Patent Document 1 discloses an example of a conventional distortion compensation system. As shown in FIG. 1, the optical communication system shown includes an optical transmitter 101, an optical receiver 102, and pre-compensation means 103. The optical transmitter 101 includes a DAC (Digital to Analog Converter) 101a and transmission means 101b. The transmission signal input to the optical transmitter 101 is transferred to various sub-devices such as the DAC 101a that converts the digital input signal into an analog form and transmits it. The transmission means 101b may include an excitation amplifier and a Mach-Zehnder modulator shown as 101b. in

[0003] When the optical receiver 102 receives the optical transmission signal from the optical transmitter 101, the receiving means 102a demodulates the optical signal using an optical demodulation method such as coherent recovery. Then, the demodulated optical signal is supplied to the analog-to-digital converter 102b. The ADC 102b converts the analog optical signal into a digital signal. The DSP (Digital Signal Processing) 102c processes the digital signal. The DSP 102c includes various signal processes such as synchronization and filtering.

[0004]

[0005] The processing of the optical transmitter 101 is not ideal, causing undesirable distortion. Since these distortions affect the transmission performance, they need to be appropriately compensated. In order to add a component for pre-compensating the input signal, the pre-compensation means 103 is used. Conventional systems with such a structure operate as follows: First, an input signal consisting of the encoded message to be transmitted is provided to the DAC 101a of the optical transmitter 101. Next, this analog input is converted into an optical signal and transmitted to the optical receiver 102. This optical signal is demodulated by the receiving means 102a in the optical receiver 102. The input to the optical transmitter 101 and the output from the optical receiver 102 are used in the pre-compensation means 103 to learn a pre-compensation function. This can be achieved using various function learning methods for input and output data, such as filter coefficient learning, memory polynomials, and neural networks. This can also be done as shown in Non-Patent Literature 1, which describes function learning based on neural networks. As shown in Figure 1, the pre-compensation means 103, which includes a compensation function learned using input and output data, can also be implemented before the optical transmitter 101 during the implementation phase.

[0006] The method described above can also be implemented as a post-compensation scenario in which the compensation function is implemented after the optical receiver 102. In this scenario, the compensation means is implemented after the optical receiver 102 and not before the optical transmitter 101. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] U.S. Patent No. 7,756,421, Electrical domain compensation of non-linear effects in an optical communications system. [Non-patent literature]

[0008] [Non-Patent Document 1] G. Paryanti, H. Faig, L. Rokach, and D. Sadot, "A Direct Learning Approach for Neural Network Based Pre-Distortion for Coherent Nonlinear Optical Transmitter," Journal of Lightwave Technology, vol. 38, no. 15, pp. 3883-3896, August 1, 2020, doi: 10.1109 / JLT.2020.2983229 [Overview of the Initiative] [Problems that the invention aims to solve]

[0009] The first problem is that the distortion compensation function learned from the collected signal data is inaccurate. The reason for this first problem is that some of the distortion components are inherently indeterminate and therefore cannot be learned. In other words, some of the distortion cannot be learned because it may be noise or other factors that follow a distribution that makes it indeterminate.

[0010] The second problem is that the learned distortion compensation function is only suitable for the operating environment in which the signal data was collected. This second problem arises because the function learning is inaccurate and compensates for the noise profile of operating conditions, which may change as the operating conditions change. The noise profile of operating conditions refers to the system characteristics, such as the SNR and other noise characteristics. As the input power and other system conditions change, this noise profile also changes, and therefore the compensation function also needs to be changed.

[0011] The third problem is that a large amount of data is required to accurately train the distortion compensation function. The reason this third challenge arises is that a large amount of data is needed to eliminate the effects of nondeterminism and to train an accurate distortion compensation function due to the effects of determinism.

[0012] (Purpose of the present invention) One of the purposes of this disclosure is to provide a noise reduction system, a signal processing system, a noise reduction method, and a noise reduction program that can solve at least one of the above-mentioned problems. [Means for solving the problem]

[0013] A noise reduction system for optical communication systems, including optical transmitters and optical receivers, is proposed. This noise reduction system is A distortion identification means that identifies distortion in an input signal from an optical receiver using an unsupervised method and outputs distortion parameters indicating the identified distortion, The system includes signal correction means for correcting the signal input to the optical transmitter, with or without using a digital transmission signal, using the output distortion parameter.

[0014] A signal processing system for an optical communication system including an optical transmitter and an optical receiver is proposed. This signal processing system is A distortion identification means that identifies distortion in an input signal from an optical receiver using an unsupervised method and outputs distortion parameters indicating the identified distortion, A signal correction means that corrects the signal input to the optical transmitter using the output distortion parameters, Corrected signals and Digital transmission signal A compensation means for training a function, wherein the compensation means compensates for the distortion of at least one signal input to the compensation means. of include.

[0015] A noise reduction method performed by a computer is proposed for an optical communication system including an optical transmitter and an optical receiver. This method is: Using an unsupervised method, the distortion of the input signal from the optical receiver is identified, and distortion parameters indicating the identified distortion are output. The signal input to the optical transmitter is corrected using the output distortion parameters, either with or without a digital transmission signal.

[0016] A noise removal program for an optical communication system including an optical transmitter and an optical receiver is proposed. This program causes a computer to in an unsupervised manner, identify distortion of an input signal from an optical receiver, and output a distortion parameter indicating the identified distortion; and using a digital transmission signal or without using a digital transmission signal, correct a signal input to the optical transmitter using the output distortion parameter.

Advantages of the Invention

[0017] The first advantage is that accurate learning of a distortion compensation function becomes possible. The reason for the first advantage is that the noise removal means can remove or reduce the influence of unlearnability or non-determinism in the signal data.

[0018] The second advantage is that the distortion characteristics are accurately learned by a blind method or an unsupervised method and used for learning a compensation profile. The reason for the second advantage is that the distortion identification means can accurately identify the distortion characteristics from the provided signal data without labels.

[0019] The third advantage is that the requirement for learning data of the distortion characteristics is reduced. The reason for this advantage is that the noise removal means removes unlearnable or indeterminate elements in the data, so that less data is required for accurately inferring a distortion compensation function.

Brief Description of the Drawings

[0020] [Figure 1] It is a block diagram showing a configuration of a prior art of an optical communication system having pre-compensation. [Figure 2] It is a diagram showing a first exemplary embodiment having pre-compensation implemented according to the proposed method. [Figure 3] It is a flowchart showing a processing flow according to the first exemplary embodiment. [Figure 4]It is a key component of a noise reduction system. [Figure 5] It is a major component of a signal processing system. [Figure 6] This figure shows a second exemplary embodiment having ex post compensation implemented in accordance with the proposed method. [Figure 7] This is a flowchart showing the processing flow according to a second exemplary embodiment. [Figure 8] This figure shows a third exemplary embodiment having pre-strain compensation implemented in accordance with the proposed method. [Figure 9] This is a flowchart showing the processing flow according to a third exemplary embodiment. [Figure 10] This figure shows a fourth exemplary embodiment having ex post compensation implemented according to the proposed method. [Figure 11] This is a flowchart showing the processing flow according to a fourth exemplary embodiment. [Figure 12] This figure shows the functionality of the proposed method for QPSK constellation. [Figure 13] This figure shows the distortion characteristic learning for 16QAM back-to-back optical devices. [Figure 14] This is an example of pseudocode summarizing a third exemplary embodiment. [Modes for carrying out the invention]

[0021] (Example of the first exemplary embodiment) (Explanation of structure) First, a first exemplary embodiment of the present invention will be described with reference to the accompanying drawings.

[0022] Referring to Figure 2, an optical communication system 20 according to a first exemplary embodiment will be described. The optical communication system 20 comprises a distortion identification means 201, a signal correction means 202, and a pre-compensation means 203. The distortion identification means 201 and the signal correction means 202 correspond to a noise reduction system 200. These means 201 to 203 are embodied by at least one computer having at least one program. The computer may include various integrated circuits such as a CPU (Central Processing Unit), processor, data processing device, FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit).

[0023] These proposed distortion identification means 201 and signal correction means 202 can be included in an existing prior art setup of system 100, which includes an optical transmitter 101, an optical receiver 102, and a pre-compensation means 103. The pre-compensation means 203 may be identical to the pre-compensation means 103 shown in Figure 1.

[0024] These methods generally work as follows:

[0025] The distortion discrimination means 201 includes an unsupervised class of learning algorithms that learn parameters enabling the characterization of the statistical properties of a signal. The unsupervised class of learning algorithms aims to learn an object from a signal without labels. In this case, the label is the actual transmitted signal (i.e., the digital transmitted signal (ground truth)). The object is to learn the properties of the signal, such as the mean or variance of symbols, blindly, i.e., without labels and in an unsupervised manner. Specifically, for a K-order signal constellation, a typical learning algorithm aims to learn parameters such as the mean (i.e., center) and variance corresponding to each of the k unique transmitted symbols. Many more parameters can also be trained in this distortion discrimination means 201 to characterize the signal. Various unsupervised learning algorithms can be employed. For example, the learning algorithm could be a Gaussian mixture model (GMM) that aims to learn the properties of the signal under the assumption that the properties of the signal's features belong to the Gaussian distribution class. The distribution from which the parameters are assumed to be learned can be adjusted according to the operating conditions of the system. In other words, depending on the distribution to which the noise may belong, an appropriate mixture model can be adopted as an unsupervised learning algorithm.

[0026] The signal correction means 202 utilizes parameters learned from the distortion identification means 201, and these parameters are used to correct the signal. This correction can be performed using distribution-related characteristics and the learned parameters identified by the preceding distortion identification means 201. Identifying the distribution to which the distorted signal best corresponds is a crucial step achieved by the distortion identification means 201, which facilitates further correction of the signal in the signal correction means 202. One such characteristic is the "confidence interval" of the distribution, which indicates the probability that a signal point corresponds to a particular symbol among k input symbols. This correction can be performed using the actual signal, the digital transmission signal, with a confidence interval threshold "T", as follows:

number

[0027] The signal correction means 202 can also correct the signal without using a digital transmission signal, as described below.

number

[0028] The signal correction means 202 can select two modes of processing: a first mode that uses a digital transmission signal and a second mode that does not use a digital transmission signal. In the first mode that uses a digital transmission signal, the signal correction means 202 sets the signal correction / trimming using the digital transmission signal of the actual signal. The second mode can be used in situations where it is difficult or impossible to obtain the digital transmission signal of the actual signal. In the second mode that does not use a digital transmission signal, the signal correction means 202 forms a "pseudo-digital transmission signal" using likelihood estimation for each symbol from the distortion identification means 201. The signal correction means 202 then sets the signal correction / trimming using the "pseudo-digital transmission signal".

[0029] The output from the signal correction means 202 is supplied to the pre-compensation means 203. The pre-compensation means 203 corresponds to a compensation means that compensates for distortion of at least one input signal.

[0030] These distortion identification means 201 and signal correction means 202 work together to remove noise from the data from the optical receiver 102, and in particular, improve the performance of the noise reduction system 200 by improving the data to the pre-compensation means 203.

[0031] (Operation description) Next, the general process of the first exemplary embodiment will be described with reference to the flowchart in Figure 3.

[0032] First, in step S101 of Figure 3, a back-to-back (b2b) connection is established with the optical transmitter 101 and optical receiver 102 connected, meaning the optical device operates without compensation in a back-to-back setup. In this optical device, the input transmission symbols pass through the optical transmitter 101 and optical receiver 102, and the received symbols, which are the output from the optical receiver 102, are collected by the distortion identification means 201. The input transmission pattern consists of the message to be transmitted, encoded using appropriate constellation symbols. This input signal is converted to electrical format, then to optical format, finally back to electrical format at the optical receiver 102, and finally back to digital format at the output of the optical receiver 102. Then, in step S102, both the input signal to the optical transmitter 101 and the output signal from the optical receiver 102 are collected. The output signal to the optical receiver 102 is used by the distortion identification means 201. The input signal to the optical transmitter 101 is used by the signal correction means 202 and the pre-compensation means 203. The signal correction means 202 may or may not use the input signal depending on whether the signal trimming is supervised or unsupervised.

[0033] Furthermore, in step S103, a blind unsupervised clustering algorithm is trained; that is, a distortion discrimination means 201, which implements the unsupervised clustering algorithm described above as an unsupervised learning algorithm, is executed on the received signal output data collected in step S102. The distortion discrimination means 201 learns the characteristics of the received constellations with the aim of classifying the received signals as symbols of the original digital transmission signals. Generally, the unsupervised learning algorithm described in the distortion discrimination means 201 aims to classify (i.e., cluster) multiple signals. We use this algorithm to extract parameters used to classify multiple signals. The extracted parameters are provided to the signal correction means 202.

[0034] In step S104, confidence region T i The dataset (i.e., multiple signals) is modified. In this step, the output from the distortion identification means 201 and the learned parameters are provided to the signal modification means 202, where the noise in the signal is removed and modified. In this step, the output of the signal modification means 202 is generated using the parameters from the distortion identification means 201, as described in the paragraph describing the structure. This output is the modified signal.

[0035] In step S105, the pre-compensation means 203 is trained. In this step, the modified signal is provided to the pre-compensation means 203, and the training is performed using the digital transmission signal. The pre-compensation means 203 can use the digital transmission signal available from the optical transmitter 101.

[0036] After training converges, in step S106, the loss from the digital transmission signal is evaluated using a loss function. This evaluation of the loss from the digital transmission signal is an automated process, similar to that in neural network training. The evaluation of the loss from the digital transmission signal assesses the fit / accuracy of the pre-compensation mechanism 203. The evaluation criterion can be a loss function such as the "mean squared error loss." This loss is evaluated by comparing the learned compensation output with the expected ideal value generated from the available digital transmission signal. This loss evaluation method may vary depending on the implementation of the pre-compensation mechanism 203, e.g., a filter, memory polynomial, or artificial neural network. The output from the loss function is used to modify the weight function using an iterative algorithm such as backpropagation to ensure convergence and minimum loss.

[0037] In step S107, the evaluation of the back-to-back (b2b) connection is performed by a simple evaluation using a loss function or by calculating the BER (Bit Error Ratio). This process may be performed manually or automated by a defined function. In this step, the pre-compensation means 203 is first implemented in the back-to-back setup. In step S107, the performance of the complete system is evaluated. That is, the optical transmitter 101 is connected to the setup of the optical receiver 102 in the back-to-back setup, and the evaluation is performed with or without the use of a pre-distortion function. This step is performed to evaluate the performance improvement achieved by the distortion reduction by the pre-compensation means 203.

[0038] If an optical channel exists connected between the optical transmitter 101 and the optical receiver 102, the channel is configured in step S108. In other words, an optical fiber is connected between the optical transmitter 101 and the optical receiver 102. This configuration can be done manually or automatically. Back-to-back setup is used during the initial steps so that the distortion generated by the optical transmitter 101 is present only in the signal data used to learn the distortion minimization strategy. However, an online communication system may include at least one channel, for example, an optical fiber cable placed between the optical transmitter 101 and the optical receiver 102. Therefore, at the configuration stage, the learned compensation function is implemented in the optical communication system including the optical fiber.

[0039] Figure 4 shows the main components of a noise reduction system 200 for an optical communication system including an optical transmitter 101 and an optical receiver 102. The noise reduction system 200 includes a distortion identification means 201 and a signal correction means 202.

[0040] As described above, the distortion identification means 201 identifies distortion in the input signal from the optical receiver 102 in an unsupervised manner and outputs distortion parameters indicating the identified distortion. The signal correction means 202 corrects the signal input to the optical transmitter 101 using the output distortion parameters, with or without using the digital transmission signal, and removes noise from the signal. The corrected signal is used by a distortion function that compensates for signal distortion used in the optical communication system to learn using the digital transmission signal. Thus, distortion generated by the optical transmitter 101 is reduced.

[0041] Furthermore, the strain identification means 201 is configured to learn strain parameters by learning statistical information of the probability distribution of the identified strain.

[0042] Furthermore, the distortion identification means 201 is configured to learn statistical information by a learning algorithm.

[0043] Furthermore, the distortion identification means 201 is configured to cluster and separate the input signals according to at least one transmitted symbol of the input signals.

[0044] Furthermore, the signal correction means 202 is configured to correct the signal based on the statistical characteristics of the signal, which include the likelihood of the signal.

[0045] Figure 5 shows the main components of a signal processing system for an optical communication system including an optical transmitter 101 and an optical receiver 102. The signal processing system includes a distortion identification means 201, a signal correction means 202, and a pre-compensation means 203.

[0046] As described above, the pre-compensation means 203 learns a function that compensates for the distortion of at least one signal input to the pre-compensation means 203 using the digital transmission signal and the corrected signal. Thus, the distortion caused by the optical transmitter 101 is reduced.

[0047] Furthermore, the function of the pre-compensation means 203 compensates for distortion of at least one signal input to the optical transmitter.

[0048] (Explanation of effects) Next, the effects of the first exemplary embodiment will be described with reference to Figure 12. Figure 12 shows an ideal QPSK constellation of the noise reduction system according to the present disclosure. In Figure 12, letter d represents the variance related to the distribution of labeled symbols before the signal correction means. Letter d' represents the variance related to the distribution of symbols after the application of the signal correction means.

[0049] Constellation 001 shows the constellation after extracting distribution statistics from the signal. Constellation 002 shows the constellation after correcting the characteristics of the signal by modifying the signal statistics.

[0050] In the first exemplary embodiment, as shown in Figure 12001, the strain identification means 201 is configured to learn strain characteristics by an algorithm, so that the strain characteristics can be learned as learnable parameters.

[0051] Furthermore, as shown in Figure 12002, this exemplary embodiment is configured to correct the distortion components according to a distribution identified by the signal correction means 202 and a certain confidence threshold related to the likelihood function, so that the pre-compensation means 203 becomes more accurate in modeling and can reduce the amount of data required to converge to an accurate pre-distortion profile. Note that this denoising is performed on all symbols shown in Figure 12, i.e., all symbols in Figure 12b are de-de-de-noised.

[0052] (Second exemplary embodiment) (Explanation of structure) Next, a second exemplary embodiment of the present invention will be described with reference to the accompanying drawings.

[0053] Referring to Figure 6, an optical communication system 30 according to a second exemplary embodiment will be described. The optical communication system 30 comprises a distortion identification means 301, a signal correction means 302, and a post-compensation means 303. The distortion identification means 301 and the signal correction means 302 correspond to a noise reduction system 300. These means 301 to 303 are embodied by at least one computer having at least one program. The computer may include various integrated circuits such as a CPU, processor, data processing unit, FPGA, and ASIC.

[0054] The second embodiment aims to demonstrate how the proposed noise reduction system 300 is implemented in a post-hoc compensation scenario where the compensation block is positioned after the optical receiver 102.

[0055] These proposed means are included in an existing prior art configuration 100, which includes an optical transmitter 101 and an optical receiver 102. We focus on a post-compensation scheme using a post-compensation means 303 instead of a pre-compensation means 203. The post-compensation means 303 corresponds to a compensation means that compensates for distortion of at least one input signal.

[0056] The post-compensation means shown in Figure 1 represents a change from the prior art, but this does not indicate novelty. Post-compensation is a well-known technique for compensating for the effects of optical systems. The distortion identification means 301 and signal correction means 302 implemented before the post-compensation means 303 demonstrate the implementation of the proposal in this embodiment.

[0057] The post-compensation means 303 is similar to the pre-compensation means 103 and can operate using neural networks, filters, memory polynomials, and other techniques for inferring a desired function from available data.

[0058] These methods generally work as follows:

[0059] The distortion discrimination means 301 includes an unsupervised class of learning algorithms that learn parameters enabling the characterization of the statistical properties of a signal. The unsupervised class of learning algorithms aims to learn objects from signals without using labels. In this case, the labels are the actual transmitted signals (i.e., digital transmitted signals). For a K-order signal constellation, a typical learning algorithm aims to learn parameters such as the mean (i.e., center) and variance corresponding to each of the k unique transmitted symbols. This distortion discrimination means 301 can also learn many more parameters to characterize the signal. The learning algorithm can be a Gaussian mixture model (GMM), which aims to learn the properties of a signal under the assumption that the properties of the signal's features belong to the Gaussian distribution class. The distributions assumed to be learned for learning parameters can be adjusted according to the operating conditions of the system.

[0060] The signal correction means 302 uses parameters learned from the distortion identification means 301, which are used to correct the signal. This correction can be performed using distribution-related characteristics and learned parameters. One such characteristic is the "confidence interval" of the distribution, which indicates the possibility that a signal point corresponds to a particular symbol among the k input symbols. This correction is performed using the confidence interval as follows: threshold "T" and the actual signal Digital transmission signal This can be done using [a specific method / tool].

number

[0061] The signal correction means 302 does not necessarily have to use a digital transmission signal, as follows:

number

[0062] The output from the signal correction means 302 is supplied to the post-compensation means 303.

[0063] In another embodiment, the distortion identification means 301, the signal correction means 302, and the post-compensation means 303 can be used together to learn the optimal settings for compensating for optical distortion in a post-compensation manner. In the case of the neural network-based post-compensation means 303, this joint learning can be performed by a multiple loss objective function that requires the outputs and parameters from the distortion identification means 301 and the signal correction means 302. The multiple loss objective function corresponds to a scenario in which a weighted combination of multiple loss functions is used in training the neural network. A possible combination of two loss functions is the mean squared error and likelihood function for the digitally transmitted signal symbols. In this scenario, the likelihood function derives parameters from the distortion identification means 301 and the signal correction means 302. This can be implemented as follows:

number

[0064] These distortion identification means 301 and signal correction means 302 work together to remove noise from the data from the optical receiver 102, and in particular, improve the performance of the noise reduction system 300 by improving the data to the post-compensation means 303.

[0065] (Operation description) Next, the general operation of the second exemplary embodiment will be described with reference to the flowchart in Figure 7.

[0066] First, in step S201 of Figure 7, a back-to-back (b2b) connection is established, meaning the optical device operates without compensation in the back-to-back setup. In this step, the input transmit symbols pass through the optical transmitter 101 and the optical receiver 102, and the received symbols become the output from the optical receiver 102. Then, in step S202, the output data is collected by the distortion identification means 301.

[0067] Furthermore, in step S203, a blind unsupervised clustering algorithm is trained; that is, a distortion discrimination means 301, which implements the aforementioned unsupervised clustering algorithm as an unsupervised learning algorithm, is executed on the output data collected in step S202.

[0068] In step S204, confidence region T i The dataset (i.e., multiple signals) is modified. In this step, the parameters learned from the distortion identification means 301 are used by the signal modification means 302 to remove and correct noise from the received signal output.

[0069] In step S205, the post-compensation means 303 is trained. In this step, the modified signal output from the signal modification means 302 is provided to the post-compensation means 303, and training is performed using the digital transmission signal with the aim of extracting the original signal from the modified signal. This objective is in contrast to the objective of learning pre-compensation for the input of the pre-compensation means 203 in the first exemplary embodiment. Any known method of learning post-compensation using a dataset containing the input symbol pattern from the signal modification means 302 and the modified signal pattern can be employed for this training / fitting.

[0070] After the training has concluded, step S206 evaluates the loss from the digital transmission signal. Step S207 evaluates the back-to-back (b2b) connection. In this step, the post-compensation means 303 is first implemented in the back-to-back setup.

[0071] If an optical channel exists connected between the optical transmitter 101 and the optical receiver 102, the channel is configured in step S208.

[0072] (Explanation of effects) Next, the effects of a second exemplary embodiment will be explained with reference to Figure 12.

[0073] A second exemplary embodiment, as shown in Figure 12001, is configured to learn the strain characteristics by the algorithm of the strain identification means 301 so that the strain characteristics can be learned as learnable parameters including statistical information of the signal constellation.

[0074] Furthermore, as shown in Figure 12, 002, the second exemplary embodiment is configured to correct the distortion component according to the distribution identified by the signal correction means 302 using information learned from the distortion identification means 301 and a certain confidence parameter relating to likelihood. This improves the accuracy of the modeling of the post-compensation means 303 and reduces the amount of data required for convergence.

[0075] Furthermore, the function of the post-compensation means 303 compensates for the distortion of at least one signal output from the optical receiver.

[0076] (Third exemplary embodiment) (Explanation of structure) Next, a third exemplary embodiment will be described with reference to the accompanying drawings. Referring to Figure 8, the optical communication system 40 according to the third exemplary embodiment will be described. The third exemplary embodiment is identical to the first exemplary embodiment except for the feedback from the pre-compensation means 403 to the signal correction means 402.

[0077] The distortion identification means 401 is identical to the distortion identification means 201 in the first exemplary embodiment, and the signal correction means 402 is similar to the signal correction means 202 which uses an additional input for setting a confidence interval T. The distortion identification means 401 and the signal correction means 402 correspond to the noise reduction system 400.

[0078] The pre-compensation means 403 corresponds to the pre-compensation means 203 with an additional output indicating the level or accuracy of the fit achieved by the input. The feedback corresponds to the output of a loss function from the pre-compensation means 403, which can be a loss function. The output of the loss function indicates the accuracy of the learned function by comparing it with the output expected based on the digital transmission signal. This feedback is used to adjust parameters used by the signal correction means 402, such as the confidence interval "T". This adjustment of the confidence interval "T" is done with the aim of feeding back the loss function value in the next iteration, i.e., the pre-compensation means 403 is trained with the new output from the signal correction means 402. This ensures that the output from the signal correction means 402 also depends on the previous state of the pre-compensation means 403.

[0079] (Operation description) At least one program according to the third exemplary embodiment is loaded into at least one computer and controls the computer's processing. The computer includes various integrated circuits such as a CPU (Central Processing Unit), processor, data processing unit, FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit). The computer performs the following processes under the control of the program, and these processes are the same as those performed by the computer in the first exemplary embodiment.

[0080] Next, the general process of the third exemplary embodiment will be described with reference to the flowchart in Figure 9.

[0081] First, in step S301 of Figure 9, a back-to-back (b2b) connection is established, meaning the optical device operates without compensation in the back-to-back setup. In this step, the input transmit symbols pass through the optical transmitter 101 and the optical receiver 102, and the received symbols become the output from the optical receiver 102. Then, in step S302, the output data is collected by the distortion identification means 401.

[0082] Furthermore, in step S303, a blind unsupervised clustering algorithm is trained; that is, a distortion discrimination means 401, which implements the above-described unsupervised clustering algorithm as an unsupervised learning algorithm, is executed on the output data collected in step S302.

[0083] In step S304, confidence region T i The dataset (i.e., multiple signals) is modified. In this step, the parameters learned from the distortion identification means 401 are used by the signal modification means 402 along with the received signal and the confidence parameter "T" to remove noise from the signal.

[0084] In step S305, the pre-compensation means 403 is trained. In this step, the modified signal is provided to the pre-compensation means 403, and the training is performed using the digital transmission signal. After the training has converged, in step S306, the loss from the digital transmission signal is evaluated.

[0085] Based on the loss from step S306, the next step is determined. In step S307, it is determined whether the loss is within the expected range. Step S307 may be performed by the loss determination means of the noise reduction system 400. This loss determination means can be implemented by at least one computer having at least one program. This expected range corresponds to the acceptable range of loss. The loss function / accuracy criterion is used to evaluate the loss. If the selected loss function is "mean squared error", the loss under evaluation is compared to a default loss threshold. If the loss under evaluation is less than this default value, it is determined that the loss is within the expected range.

[0086] If the loss is not within the expected range (NO), in step S308, the confidence interval T used by the signal correction means 402 is updated based on feedback from the pre-compensation means 403. Step S308 may be performed by the updating means of the noise reduction system 400. The updating means can be implemented by at least one computer having at least one program. In this process, the feedback from the compensation means is a value such as the output of the loss function, which indicates the fit of the learned function to the digital transmission signal. This estimate indicates whether the correction of the data signal made in the confidence interval is accurate. If the value of the loss function used as feedback is high, the confidence interval used is appropriately adjusted by the new output from the signal correction means so that the value of the loss function decreases or the fit improves. Then, the process returns to step S304.

[0087] If the loss is within the expected range (YES), the process proceeds to step S309. In step S309, the back-to-back (b2b) connection is evaluated. In this step, the pre-compensation means 403 is first implemented in the back-to-back setup.

[0088] If an optical channel exists between the optical transmitter 101 and the optical receiver 102, the channel is configured in step S310.

[0089] (Explanation of effects) Next, the effects of a third exemplary embodiment will be described using Figure 12, which shows the effect on the QPSK signal constellation.

[0090] In a third exemplary embodiment, as shown in Figure 12001, the strain characteristics are configured to be learned by the algorithm of the strain identification means 401, so that the strain characteristics can be learned as learnable parameters including statistical information such as the mean or variance.

[0091] Furthermore, in a third exemplary embodiment, as shown in Figure 12, 002, the signal correction means 402 is configured to correct the distortion components according to the distribution identified by the signal correction means 402 using feedback from the pre-compensation means 403. That is, the multiple signal correction means 402 adaptively adjust the output of the signal correction means using at least one feedback from the pre-compensation means 403 that indicates the accuracy of the learned function. This allows the pre-compensation means 403 to be more accurate in modeling and reduce the data required for convergence with higher precision.

[0092] (Fourth exemplary embodiment) (Explanation of structure) Next, a fourth exemplary embodiment will be described with reference to the drawings. Referring to Figure 10, the optical communication system 50 according to the fourth exemplary embodiment will be described. The fourth exemplary embodiment shows the same implementation as the second exemplary embodiment, except for the post-compensation means 503.

[0093] The distortion identification means 501 shown in this embodiment is identical to the distortion identification means 301, and the signal correction means 502 is identical to the signal correction means 302, except for the additional input (feedback) from the post-compensation means 503 used to set the confidence interval "T". The distortion identification means 501 and the signal correction means 502 correspond to the noise reduction system 500.

[0094] The post-compensation means 503 is similar to the post-compensation means 303, with an additional output indicating the degree of accuracy of the achieved fit. The feedback may correspond to the output of a loss function from the post-compensation means 503, which may be a loss function. This feedback is used to adjust parameters of the signal correction means 502, such as a confidence interval "T".

[0095] (Operation instructions) At least one program according to a fourth exemplary embodiment is loaded into at least one computer and controls the operation of the computer. The computer may include various integrated circuits such as a CPU, processor, data processing unit, FPGA, and ASIC. The computer performs the following operations under the control of the program, which are the same operations performed by the computer in the second embodiment.

[0096] Next, the general process of the fourth exemplary embodiment will be described with reference to the flowchart in Figure 11.

[0097] First, in step S401 of Figure 11, a back-to-back (b2b) connection is established, meaning the optical device operates without compensation in the back-to-back setup. In this step, the input transmit symbols pass through the optical transmitter 101 and the optical receiver 102, and the received symbols become the output from the optical receiver 102. Then, in step S402, the output data is collected.

[0098] Furthermore, in step S403, a blind unsupervised clustering algorithm is trained; that is, a distortion discrimination means 501 that implements the above-described unsupervised clustering algorithm as an unsupervised learning algorithm is executed on the output data collected in step S402.

[0099] In step S404, confidence region T iThe dataset (i.e., multiple signals) is modified. In this step, the parameters learned from the distortion identification means 501 are used by the signal modification means 502 along with the received signal output to remove noise from the signal.

[0100] In step S405, the post-compensation means 303 is trained. In this step, the modified signal is provided to the post-compensation means 503, and training is performed using the digital transmission signal.

[0101] After the training has concluded, in step S406, the loss from the digital transmission signal is evaluated.

[0102] Based on the loss from step S406, the next step is determined. In step S407, it is determined whether the loss is within the expected range. Step S407 can be performed by the loss determination means of the noise reduction system 500. The loss determination means can be implemented by at least one computer having at least one program. This expected range corresponds to the acceptable range of loss. If the loss is not within the expected range (NO), in step S408, the confidence interval T used by the signal correction means 502 is updated based on feedback from the post-compensation means 503. Step S408 may be performed by the update means of the noise reduction system 500. The update means can be implemented by at least one computer having at least one program. Then, the process returns to step S404. If the loss is within the expected range (YES), the process proceeds to step S409, where the post-compensation means 503 is first implemented in a back-to-back setup.

[0103] In step S410, if an optical channel exists connected between the optical transmitter 101 and the optical receiver 102, the channel is configured.

[0104] (Explanation of effects) Next, the effects of the fourth exemplary embodiment will be described using Figure 12, which shows the proposed processing of the QPSK signal.

[0105] In a fourth exemplary embodiment, as shown in Figure 12001, the strain characteristics are configured to be learned by the algorithm of the strain identification means 501, so that the strain characteristics can be learned as learnable parameters corresponding to statistical information including the mean or variance from the mean.

[0106] Furthermore, in a fourth exemplary embodiment, as shown in 002 of Figure 12, the signal's statistical information is modified so that the distortion components are modified according to the distribution identified by the signal correction means 502. This allows the post-compensation means 503 to perform more accurate modeling and reduces the data required for convergence. The correct statistical information, along with feedback from the post-compensation means 503, is used to fit the signal, thus achieving accurate fitting by the post-compensation means 503.

[0107] (example) Next, the processing of modes for carrying out the present invention will be described using a specific example that embodies a third exemplary embodiment.

[0108] As shown in Figure 8, a pre-compensation means 403 for reducing distortion in advance is connected to the optical transmitter 101.

[0109] First, the optical transmitter 101 and the optical receiver 102 are connected by a back-to-back data connection. In this example, input symbols belonging to a 16-QAM constellation are transmitted. The output from the optical receiver 102 is a signal from a coherent receiver that has passed through the ADC 102b, which is further processed by the DSP 102c, which has an algorithm that includes filtering and sampling.

[0110] Next, the output from the optical receiver 102 is provided to the strain identification means 401. The strain identification means 401 is performed based on the assumption that the strain follows a Gaussian distribution. Thus, the unsupervised clustering algorithm (UCA) used in the strain identification means 401 is a Gaussian mixture model (GMM). In this case, the GMM is set up to have 16 clusters, which are in the same order as the constellation. The GMM uses the coordinates of the 16 cluster centers and the signals that converge to a solution containing the variance relative to the centers in two-dimensional coordinates. This information can be used to estimate the probability that a signal belongs to one of the clusters (i.e., one of the 16 transmitted symbols).

[0111] Figure 13 shows the clustering of a 16-QAM received constellation from a 32G baud back-to-back optical device with significant nonlinear distortion. The GMM can cluster the symbols and accurately identify the centers of the 16 clusters. For the i-th cluster, the cluster center is C i This is shown, and the variance is V i As shown, the signal and the learned parameters from the GMM are provided to the signal modification means 402. In this example, a simple signal modification using GMM parameters is proposed.

number

[0112] After this modification, the output of the signal modification means 402 is provided to the pre-compensation means 403, where this signal data, along with the input symbol pattern (digital transmission signal), is used to fit an appropriate compensation function. This fit can be used to learn FIR filter coefficients, memory or Volterra polynomial coefficients, or neural network weights.

[0113] After this fitting, the loss or inaccuracy of the fitting may be evaluated. If the loss exceeds an acceptable threshold, the value of the confidence threshold T is corrected, and the signal correction means 402 corrects the signal until the loss is within an acceptable range.

[0114] In Figure 14, pseudocode representing the above explanation is shown along with digital pre-distortion (DPD), which corresponds to the pre-compensation means. Note that DPD may be replaced with post-compensation means. After the loss is within an acceptable range, this adapted function is adopted and implemented in the flow before the transmission means 101b, and the input pattern is pre-distorted, accurately mitigating the effects of distortion. By removing noise (removing unlearnable noise and stochastic components) using the distortion identification means 401 and the signal correction means 402, the function to be learned is accurately extracted from the available data, thereby significantly reducing the amount of signal data required for function adaptation.

[0115] All other embodiments similar to the above-described examples of the third exemplary embodiment may be implemented. [Industrial applicability]

[0116] The present invention is applicable to pre-distortion systems relating to the effects of optical transmitters. Specifically, it is a signal modification program aimed at removing noise and other undesirable characteristics from a signal. The present invention is also applicable to post-compensation systems that compensate for the effects of optical transmitters. The present invention is also applicable to optical communications aimed at compensating for the effects of either pre-mode or post-mode optical communications.

[0117] Various other variations will be obvious to those skilled in the art and will not be described in further detail here. It should be noted that the proposed patent can be implemented directly in a complete optical communication system with channels, instead of the initial back-to-back data connection and training. In other words, the distortion compensation function can be learned using data from a complete optical communication system.

[0118] The program, when loaded into a computer, includes a set of instructions (or software code) for causing the computer to perform one or more of the functions described in the embodiments. The program may be stored on a non-temporary computer-readable medium or a physical storage medium. Examples, but not limited to, include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technologies, CD-ROM, digital versatile disk (DVD), Blu-ray® disc or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices. The program may be transmitted over a temporary computer-readable medium or a communication medium. Examples, but not limited to, include electrical, optical, acoustic or other forms of propagating signals. [Explanation of Symbols]

[0119] 10, 20, 30, 40, 50 Optical communication systems 101 Optical Transmitter 101a Digital-to-Analog Converter 101b Transmission means 102 Optical receiver 102a Receiving means 102b Analog-to-Digital Converter 102c DSP 103,203,403 Pre-compensation measures 200, 300, 400, 500 Noise Reduction System 303,503 Subsequent compensation measures 201,301,401,501 Strain Identification Means 202,302,402,502 Signal modification means

Claims

1. A noise reduction system for an optical communication system, comprising an optical transmitter and an optical receiver that receives an optical signal transmitted by the optical transmitter, A distortion identification means that identifies distortion in the input signal from an optical receiver and outputs distortion parameters indicating the identified distortion, A signal correction means that corrects the signal input to the optical transmitter using the output distortion parameter, Includes, The strain identification means is configured to learn the strain parameters by learning statistical information of the probability distribution of the identified strain. Noise reduction system.

2. The noise reduction system according to claim 1, wherein the distortion identification means is configured to cluster and separate the input signal according to at least one transmission symbol of the input signal.

3. The signal correction means is configured to correct the signal based on the statistical characteristics of the signal, The noise reduction system according to claim 1 or 2, wherein the aforementioned statistical characteristics include the likelihood of the signal.

4. A noise reduction method performed by a computer in an optical communication system including an optical transmitter and an optical receiver that receives an optical signal transmitted by the optical transmitter, Using a distortion identification means, the distortion of the input signal from the optical receiver is identified, and a distortion parameter indicating the identified distortion is output. The signal input to the optical transmitter is corrected using the output distortion parameter. The strain identification means is configured to learn the strain parameters by learning statistical information of the probability distribution of the identified strain. Noise reduction methods.

5. A noise reduction program for an optical communication system including an optical transmitter and an optical receiver that receives an optical signal transmitted by the optical transmitter, wherein the program provides a computer with the following capabilities: The steps include: using a distortion identification means to identify the distortion of the input signal from the optical receiver and outputting a distortion parameter indicating the identified distortion; The steps include: correcting the signal input to the optical transmitter using the output distortion parameter; Make it run, The strain identification means is configured to learn the strain parameters by learning statistical information of the probability distribution of the identified strain. Noise reduction program.

Images