Non-intrusive optical bandpass filter frequency offset monitoring and positioning method and device
Through the non-invasive optical bandpass filter frequency offset monitoring method, the signal processing of the optical communication system is used using the deep neural network model, which solves the high cost and low adaptability of filter frequency offset monitoring in the optical network, and realizes the problem of high frequency offset monitoring of the filter in the optical network. Efficient filter damage monitoring and positioning, reducing operating costs.
Patent Information
- Application Number
- CN202210923021.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-02
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2042-08-02
AI Technical Summary
The filter frequency offset monitoring scheme in existing optical networks has problems such as high implementation cost, difficulty in realizing and low adaptation to the network environment, and cannot realize intelligent monitoring.
Adopting the frequency offset monitoring and positioning method of non-invasive optical bandpass filter, the offline signal processing is performed by obtaining the received signal of the optical communication system, and the deep neural network model based on optical fiber digital backpropagation DBP is used, and the adaptive filter is adaptive The tap coefficient is used to identify damage and locate and trace the source to realize damage monitoring and positioning of the filter.
It realizes link abnormality monitoring through received signal processing and parameter characteristics without using measuring instruments, has accurate filter damage identification and positioning traceability capabilities, can promptly reflect link status information, and reduce operating costs.
Smart Images

Figure CN115426038B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of optical fiber communication, and in particular to a non-intrusive optical bandpass filter frequency deviation monitoring and positioning method and device. Background Art
[0002] Optical fiber communication has become an important bridge of modern communication network due to its advantages such as large information capacity, long transmission distance and strong anti-interference ability. It carries the interconnection of world information and provides more and more convenience for people's life, travel and work. However, as the network becomes more complex and large, once a link abnormality occurs, the communication quality will be reduced or even interrupted, and the consequences will be unbearable. Therefore, real-time and effective monitoring of complex and changeable large-capacity optical networks is a necessary technical means to ensure efficient and stable operation of the network. Optical network performance monitoring is an important means to achieve efficient and safe operation of the network. It can help network operation and maintenance managers to detect abnormal phenomena in time and shorten the repair time and operating costs of the optical network. In order to improve the intelligent management capabilities of optical networks, network fault management application technologies based on machine learning are constantly proposed and implemented. In most fault identification schemes, relevant features are mainly learned from a large amount of communication data received to complete network fault prediction, fault diagnosis and fault identification to ensure the reliable operation of high-speed optical fiber communication systems.
[0003] In optical transmission systems, optical filters are usually used to select signal wavelengths and filter out-of-band noise, and are essential optical devices in optical networks. However, if the filter has problems such as frequency offset during operation, the signal transmission quality will be reduced. Such failures may even interrupt established communication connections. In order to ensure communication quality, once abnormal problems such as filter frequency offset occur, they must be quickly located and repaired.
[0004] Existing filter frequency offset monitoring schemes in optical networks include: ① Based on traditional monitoring schemes, bulky and expensive hardware measurement equipment is used, such as deploying a spectrometer in the optical network to perform spectral analysis of the signal to achieve damage identification and location, but this method makes the network implementation cost and deployment complexity increase sharply, and is not conducive to subsequent network maintenance; ② Based on the solution of signal processing at the receiving end, the adaptive compensation algorithm of optical communication is usually used to achieve channel state estimation. When abnormal damage such as filter abnormality and polarization rotation occurs in the link, the adaptive compensation algorithm will try to compensate for certain damage in the link. This phenomenon is mainly viewed from the frequency response of the tap coefficient in the adaptive compensation algorithm; However, this scheme can only estimate the cumulative damage in the link, and cannot locate the abnormal filter. It is also easily affected by other damages in the channel, which interferes with the monitoring effect. ③ The scheme based on combining the characteristics of the received signal with artificial neural networks mainly performs preprocessing of the received data, takes the tap coefficient, the autocorrelation function of the signal, the power spectral density, the link length, the characteristics of the adjacent WSS (Wavelength Selective Switch), and the number of spans as input information, and uses the neural network to extract relevant identification features from the data to determine whether the filter has offset damage, thereby avoiding the use of additional measurement equipment; this type of scheme belongs to the conventional machine learning method, which uses neural networks for damage classification, but requires a lot of prior information and a large amount of actual network transmission data, which is difficult to implement, and changes in the network environment will lead to reduced accuracy of the identification results.
[0005] In summary, existing monitoring solutions generally have problems such as high implementation cost, difficulty in implementation, and low adaptability to network environments, and are unable to complete intelligent monitoring of the network. Summary of the invention
[0006] The present invention aims at the problems that existing monitoring schemes generally have, such as high implementation cost, difficulty in implementation, and low adaptability to network environments, and proposes the present invention.
[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0008] In one aspect, the present invention provides a non-intrusive method for monitoring and locating frequency offset of an optical bandpass filter, which is implemented by an electronic device and includes:
[0009] S1. Obtain a received signal of an optical communication system.
[0010] S2. Perform offline signal processing on the received signal to obtain a first branch signal and a second branch signal.
[0011] S3. Input the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation DBP to obtain tap coefficients of an adaptive filter in the deep neural network model.
[0012] S4. Based on the tap coefficients of the adaptive filter in the deep neural network model, the damage identification results and damage location and tracing results of the filter in the optical communication system are obtained.
[0013] Optionally, performing offline signal processing on the received signal in S2 to obtain the first branch signal and the second branch signal includes:
[0014] S21. Perform I / Q orthogonalization on the received signal.
[0015] S22. Perform dispersion compensation on the orthogonalized received signal.
[0016] S23, performing adaptive equalization and polarization demultiplexing on the dispersion compensated received signal.
[0017] S24. Estimating the frequency offset of the received signal after adaptive equalization and polarization demultiplexing.
[0018] S25. Divide the received signal after the frequency offset estimation into two branches to obtain a first branch signal and a second branch signal.
[0019] Optionally, in S3, the first branch signal and the second branch signal are input into a deep neural network model based on optical fiber digital back propagation DBP, and the tap coefficients of the adaptive filter in the deep neural network model are obtained, including:
[0020] S31. Add dispersion damage to the first branch signal, reconstruct the signal including dispersion damage, nonlinear damage and filter offset damage, and obtain the preprocessed first branch signal.
[0021] S32. Perform carrier phase recovery and judgment on the second branch signal to obtain a signal that eliminates dispersion damage, nonlinear damage, and filter frequency offset damage, and perform phase alignment based on the first branch signal to obtain a preprocessed second branch signal.
[0022] S33. Use the preprocessed first branch signal as the input signal of the deep neural network model based on DBP, use the preprocessed second branch signal as the reference signal of the deep neural network model based on DBP, train the deep neural network model based on DBP, and obtain the tap coefficients of the adaptive filter in the deep neural network model.
[0023] Optionally, in S33, the preprocessed first branch signal is used as an input signal of a deep neural network model based on DBP, the preprocessed second branch signal is used as a reference signal of a deep neural network model based on DBP, the deep neural network model based on DBP is trained, and the tap coefficients of the adaptive filter in the deep neural network model are obtained, including:
[0024] S331. Set a DBP-based deep neural network model.
[0025] S332. Use the preprocessed first branch signal as the input signal of the DBP-based deep neural network model, input it into the set DBP-based deep neural network model, and obtain the forward calculation output signal of the deep neural network model.
[0026] S333. Use the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP, train the deep neural network model based on DBP based on the forward calculation output signal, and obtain the tap coefficients of the adaptive filter in the deep neural network model.
[0027] Optionally, setting the DBP-based deep neural network model in S331 includes:
[0028] S3311. Set the number of layers of the DBP-based deep neural network model.
[0029] S3312, obtaining the number of spans of the optical communication system, and setting an adaptive filter in a DBP-based deep neural network model according to the position of each span of the optical communication system.
[0030] Optionally, the number of layers N of the deep neural network model based on DBP is set in S3311, as shown in the following formula (1):
[0031] N=L / ΔL (1)
[0032] Wherein, L is the total transmission distance of the optical communication system; ΔL is the compensation step size of DBP.
[0033] Optionally, in S333, the preprocessed second branch signal is used as a reference signal of a deep neural network model based on DBP, and the deep neural network model based on DBP is trained based on the forward calculation output signal, so as to obtain the tap coefficients of the adaptive filter in the deep neural network model, including:
[0034] S3331. Use the preprocessed second branch signal as the reference signal of the DBP-based deep neural network model, and train the DBP-based deep neural network model based on the forward calculation output signal to obtain the dispersion coefficient, nonlinear coefficient of each layer of the deep neural network model to be optimized, and the tap coefficient of the adaptive filter in each span.
[0035] S3332. Based on the gradient descent method, the optimal values of the dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span are solved to obtain the optimal dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span of each layer of the deep neural network model.
[0036] Optionally, the tap coefficients of the adaptive filter in the deep neural network model in S4 are used to obtain damage identification results and damage location tracing results of the filter in the optical communication system, including:
[0037] S41. Obtain the tap coefficient of each adaptive filter of multiple adaptive filters in the deep neural network model.
[0038] S42 . Draw a spectrum response diagram of each adaptive filter based on the tap coefficients of each adaptive filter.
[0039] S43. Based on the spectrum response diagram of each adaptive filter, damage identification is performed according to the spectrum characteristics of each adaptive filter, thereby obtaining damage identification results and damage location tracing results of the filters in the optical communication system.
[0040] On the other hand, the present invention provides a non-intrusive optical bandpass filter frequency offset monitoring and positioning device, which is used to implement a non-intrusive optical bandpass filter frequency offset monitoring and positioning method, and the device includes:
[0041] The acquisition module is used to acquire the received signal of the optical communication system.
[0042] The offline signal processing module is used to perform offline signal processing on the received signal to obtain a first branch signal and a second branch signal.
[0043] The input module is used to input the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation DBP to obtain tap coefficients of an adaptive filter in the deep neural network model.
[0044] The output module is used to obtain damage identification results and damage location tracing results of the filter in the optical communication system based on the tap coefficients of the adaptive filter in the deep neural network model.
[0045] Optionally, the offline signal processing module is further used to:
[0046] S21. Perform I / Q orthogonalization on the received signal.
[0047] S22. Perform dispersion compensation on the orthogonalized received signal.
[0048] S23, performing adaptive equalization and polarization demultiplexing on the dispersion compensated received signal.
[0049] S24. Estimating the frequency offset of the received signal after adaptive equalization and polarization demultiplexing.
[0050] S25. Divide the received signal after the frequency offset estimation into two branches to obtain a first branch signal and a second branch signal.
[0051] Optionally, the input module is further configured to:
[0052] S31. Add dispersion damage to the first branch signal, reconstruct the signal including dispersion damage, nonlinear damage and filter offset damage, and obtain the preprocessed first branch signal.
[0053] S32, performing carrier phase recovery and judgment on the second branch signal to obtain a signal that eliminates dispersion damage, nonlinear damage, and filter frequency offset damage, and performing phase alignment based on the first branch signal to obtain a preprocessed second branch signal.
[0054] S33. Use the preprocessed first branch signal as the input signal of the deep neural network model based on DBP, use the preprocessed second branch signal as the reference signal of the deep neural network model based on DBP, train the deep neural network model based on DBP, and obtain the tap coefficients of the adaptive filter in the deep neural network model.
[0055] Optionally, the input module is further configured to:
[0056] S331. Set a DBP-based deep neural network model.
[0057] S332. Use the preprocessed first branch signal as the input signal of the DBP-based deep neural network model, input it into the set DBP-based deep neural network model, and obtain the forward calculation output signal of the deep neural network model.
[0058] S333. Use the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP, train the deep neural network model based on DBP based on the forward calculation output signal, and obtain the tap coefficients of the adaptive filter in the deep neural network model.
[0059] Optionally, the input module is further configured to:
[0060] S3311. Set the number of layers of the DBP-based deep neural network model.
[0061] S3312, obtaining the number of spans of the optical communication system, and setting an adaptive filter in a DBP-based deep neural network model according to the position of each span of the optical communication system.
[0062] Optionally, the number of layers N of the DBP-based deep neural network model is set as shown in the following formula (1):
[0063] N=L / ΔL (1)
[0064] Wherein, L is the total transmission distance of the optical communication system; ΔL is the compensation step size of DBP.
[0065] Optionally, the input module is further configured to:
[0066] S3331. Use the preprocessed second branch signal as the reference signal of the DBP-based deep neural network model, and train the DBP-based deep neural network model based on the forward calculation output signal to obtain the dispersion coefficient, nonlinear coefficient of each layer of the deep neural network model to be optimized, and the tap coefficient of the adaptive filter in each span.
[0067] S3332. Based on the gradient descent method, the optimal values of the dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span are solved to obtain the optimal dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span of each layer of the deep neural network model.
[0068] Optionally, the output module is further configured to:
[0069] S41. Obtain the tap coefficient of each adaptive filter of multiple adaptive filters in the deep neural network model.
[0070] S42 . Draw a spectrum response diagram of each adaptive filter based on the tap coefficients of each adaptive filter.
[0071] S43. Based on the spectrum response diagram of each adaptive filter, damage identification is performed according to the spectrum characteristics of each adaptive filter, thereby obtaining damage identification results and damage location tracing results of the filters in the optical communication system.
[0072] On the one hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the above-mentioned non-intrusive optical bandpass filter frequency offset monitoring and positioning method.
[0073] On the one hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the above-mentioned non-intrusive optical bandpass filter frequency offset monitoring and positioning method.
[0074] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:
[0075] In the above scheme, a filter offset identification and positioning method for long-distance optical fiber communication transmission links is proposed. By processing the received signal and combining the optimization capability of the neural network, the optimized filter parameters can be obtained, and the abnormal condition of the link can be monitored without using any measuring instruments, only by using the received signal processing and optimized parameter value characteristics. The reverse reconstruction method of the optical transmission link is realized based on data and model drive, and the dynamic monitoring scheme of the damage and transmission quality of the optical link filter is realized. It has the ability to accurately identify and locate filter damage, and can timely and comprehensively reflect the status information of the link. BRIEF DESCRIPTION OF THE DRAWINGS
[0076] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0077] Figure 1 It is a schematic flow chart of a non-intrusive optical bandpass filter frequency offset monitoring and positioning method provided by an embodiment of the present invention;
[0078] Figure 2 is a principle block diagram of a system algorithm provided by an embodiment of the present invention;
[0079] Figure 3 It is a block diagram of the DBP algorithm technology principle provided by an embodiment of the present invention;
[0080] Figure 4 is a specific calculation example diagram of the DBP algorithm provided by an embodiment of the present invention;
[0081] Figure 5 This is an example diagram of a DBP-based deep neural network simulating reverse transmission of an optical fiber link provided by an embodiment of the present invention;
[0082] Figure 6 This is a diagram showing the effect of filter-free offset recognition in a system provided by an embodiment of the present invention;
[0083] Figure 7This is a recognition effect diagram of the offset of filter No. ① provided in an embodiment of the present invention;
[0084] Figure 8 This is a recognition effect diagram of the offset of filter No. ③ provided in an embodiment of the present invention;
[0085] Fig. 9 This is a recognition effect diagram of the offset of filter No. ⑤ provided in an embodiment of the present invention;
[0086] Fig.10 It is a block diagram of a non-intrusive optical bandpass filter frequency offset monitoring and positioning device provided by an embodiment of the present invention;
[0087] Fig.11 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0088] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.
[0089] like Figure 1 As shown, the embodiment of the present invention provides a non-intrusive optical bandpass filter frequency offset monitoring and positioning method, which can be implemented by an electronic device. Figure 1 The flowchart of the non-intrusive optical bandpass filter frequency offset monitoring and positioning method shown in the figure may include the following steps:
[0090] S1. Obtain a received signal of an optical communication system.
[0091] S2. Perform offline signal processing on the received signal to obtain a first branch signal and a second branch signal.
[0092] Optionally, performing offline signal processing on the received signal in S2 to obtain the first branch signal and the second branch signal includes:
[0093] S21. Perform I / Q orthogonalization on the received signal.
[0094] S22. Perform dispersion compensation on the orthogonalized received signal.
[0095] S23, performing adaptive equalization and polarization demultiplexing on the dispersion compensated received signal.
[0096] S24. Estimating the frequency offset of the received signal after adaptive equalization and polarization demultiplexing.
[0097] S25. Divide the received signal after the frequency offset estimation into two branches to obtain a first branch signal and a second branch signal.
[0098] In a feasible implementation, Figure 2 As shown, the process of performing offline digital signal processing on the received signal of the optical communication system can be: first, I / Q orthogonalization and dispersion damage compensation are realized, then demultiplexing of polarization multiplexed signals is completed, dynamic equalization is realized and inter-code interference damage is reduced, and then frequency offset estimation is performed to compensate for the frequency offset of the lasers at the transmitting and receiving ends; after that, the signal is divided into two paths, and the first branch (left branch) is re-added with dispersion to reconstruct the signal containing dispersion damage, nonlinear damage and filter offset damage, and then sent to the neural network to reversely optimize the link transmission process, and the link state information within this step can be obtained from the relevant optimized parameters according to the corresponding relationship between the number of layers of the neural network and the transmission position; the second branch (right branch) realizes signal recovery through phase recovery and judgment, so that the signal damage caused by dispersion, nonlinearity and filter frequency offset is completely eliminated, and then the phase difference between the two paths is estimated and compensated to ensure the phase alignment of the two paths of data.
[0099] Furthermore, after discrete digital signal processing, the difference between the two branch signals is the presence or absence of dispersion, nonlinearity, and filter offset damage.
[0100] S3. Input the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation DBP to obtain tap coefficients of an adaptive filter in the deep neural network model.
[0101] In a feasible implementation mode, combining Figure 3-5 Describe the specific structure of the deep neural network based on DBP (Digital Back-Propagation).
[0102] like Figure 3 This is a principle block diagram of conventional digital reverse propagation. Its basic technical theory is to know the basic parameters of the optical fiber link, such as the loss coefficient α, dispersion coefficient β, nonlinear coefficient γ and link configuration such as transmission distance, and then reversely simulate the optical fiber transmission process through digital signal processing technology to reversely deduce the original optical signal; compared with the transmission signal in the real optical fiber link, when the signal is reversely deduced in the virtual optical fiber link, the various damages suffered are the opposite of the real link damage, then the loss coefficient of the optical fiber is -α, the dispersion coefficient is -β, the nonlinear coefficient is -γ, and the amplifier gain is 1 / G, thereby achieving compensation for related damages.
[0103] like Figure 4This is a specific calculation process for optical fiber digital reverse propagation. During reverse calculation, the optical fiber transmission link is first divided into several segments with a step length of ΔL. In this step length, dispersion and nonlinearity are approximately regarded as acting separately. Dispersion compensation in the frequency domain is achieved through FFT (FastFourier Transform) and IFFT (Inverse Fast Fourier Transform), and then converted to the time domain to achieve compensation for optical fiber nonlinear damage. Furthermore, through N=L / ΔL iterations, L is the total link length, and ΔL is the reverse simulation step length of the link. When ΔL is small, the number of iterations required will increase, resulting in a sharp increase in calculation complexity. Therefore, it is necessary to balance the simulation accuracy and implementation complexity. Finally, through iterative calculations, a signal that is approximately the same as the transmitting end can be reversely derived, thereby achieving partial signal damage compensation.
[0104] like Figure 5 The figure is a schematic diagram of building a simple deep neural network based on DBP, and its structure is basically the same as that of DBP, that is, the transmission of the optical fiber link is reversely simulated by using the neural network, so that the neural network is not just a "black box" mode, but has a clear physical explanation. For example, dispersion compensation is a linear operation in the neural network, and nonlinear compensation is an activation function of the neural network. The total number of layers of the neural network is the total number of segments of the DBP, and corresponds to the transmission distance of the link. For example, if the neural network is in the third layer, the corresponding transmission distance of the optical fiber link is 3*ΔL. On this basis, compared with the standard DBP, the embodiment of the present invention adds a complex-valued filter to be optimized at the corresponding position in each span to better simulate the link transmission condition. Through training, the dispersion and nonlinear parameters within each ΔL distance in the link and the tap coefficients of the adaptive filter in each span can be obtained. By checking the frequency response of the tap coefficients, it is possible to monitor whether the working state of the filter is abnormal.
[0105] Optionally, in S3, inputting the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation DBP to obtain tap coefficients of an adaptive filter in the deep neural network model includes:
[0106] S31. Add dispersion damage to the first branch signal, reconstruct the signal including dispersion damage, nonlinear damage and filter offset damage, and obtain the preprocessed first branch signal.
[0107] S32, performing carrier phase recovery and judgment on the second branch signal to obtain a signal that eliminates dispersion damage, nonlinear damage, and filter frequency offset damage, and performing phase alignment based on the first branch signal to obtain a preprocessed second branch signal.
[0108] S33. Use the preprocessed first branch signal as the input signal of the deep neural network model based on DBP, use the preprocessed second branch signal as the reference signal of the deep neural network model based on DBP, train the deep neural network model based on DBP, and obtain the tap coefficients of the adaptive filter in the deep neural network model.
[0109] Optionally, in S33, the preprocessed first branch signal is used as an input signal of a deep neural network model based on DBP, the preprocessed second branch signal is used as a reference signal of a deep neural network model based on DBP, the deep neural network model based on DBP is trained, and the tap coefficients of the adaptive filter in the deep neural network model are obtained, including:
[0110] S331. Set a DBP-based deep neural network model.
[0111] Optionally, setting the DBP-based deep neural network model in S331 includes:
[0112] S3311. Set the number of layers of the DBP-based deep neural network model.
[0113] Optionally, the number of layers N of the deep neural network model based on DBP is set in S3311, as shown in the following formula (1):
[0114] N=L / ΔL (1)
[0115] Wherein, L is the total transmission distance of the optical communication system; ΔL is the compensation step size of DBP.
[0116] In a feasible implementation, the number of layers of the neural network is set to the number of compensations of the DBP.
[0117] S3312, obtaining the number of spans of the optical communication system, and setting an adaptive filter in a DBP-based deep neural network model according to the position of each span of the optical communication system.
[0118] In a feasible implementation, a complex-valued adaptive filter h is placed in each span to simulate the filtering operation of the signal, so as to better realize the transmission process of the inversely reconstructed signal.
[0119] S332. Use the preprocessed first branch signal as the input signal of the DBP-based deep neural network model, input it into the set DBP-based deep neural network model, and obtain the forward calculation output signal of the deep neural network model.
[0120] S333. Use the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP, train the deep neural network model based on DBP based on the forward calculation output signal, and obtain the tap coefficients of the adaptive filter in the deep neural network model.
[0121] Optionally, in S333, the preprocessed second branch signal is used as a reference signal of a deep neural network model based on DBP, and the deep neural network model based on DBP is trained based on the forward calculation output signal, so as to obtain the tap coefficients of the adaptive filter in the deep neural network model, including:
[0122] S3331. Use the preprocessed second branch signal as the reference signal of the DBP-based deep neural network model, and train the DBP-based deep neural network model based on the forward calculation output signal to obtain the dispersion coefficient, nonlinear coefficient of each layer of the deep neural network model to be optimized, and the tap coefficient of the adaptive filter in each span.
[0123] S3332. Based on the gradient descent method, the optimal values of the dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span are solved to obtain the optimal dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span of each layer of the deep neural network model.
[0124] In a feasible implementation, after N times of dispersion and nonlinear crossover operations, N layers of neural network calculations are realized, and the number of neural network layers corresponds to the distance of the transmission link. The number of filters is equal to the number of spans in the actual link, and then the forward calculation signal E of the neural network can be obtained. The error function I can be obtained by the output signal E of the neural network and the reference signal The relevant calculation is performed; the analysis shows that the calculated output signal E of the neural network contains the dispersion β to be optimized * , nonlinear γ * And the tap parameters of the filter h. The ultimate goal is to find these three types of optimization parameters, which can be expressed as shown in the following formula (2):
[0125]
[0126] This is a classic supervised machine learning method. The gradient descent method can be used to update the parameters to be optimized to find the optimal value. The neural network is set to have N layers, so Complex valued adaptive filter h * The number of is equal to the number of spans of the optical fiber network to simulate the link transmission process. If the optical fiber link is configured as M spans, then h * =[h 1,…h M ]; then the optimization formula can be used to update the value, as shown in the following formulas (3)-(5):
[0127]
[0128]
[0129]
[0130] Among them, k represents the number of layers of the neural network, k = 1, 2...N; x represents the number of spans, x = 1, 2...M, m is the number of updates, and μ is the learning rate.
[0131] Furthermore, by continuously optimizing parameters to gradually reduce the error value, the optimal parameter value in each compensation step can be obtained; the specific difference between the two signals is whether there is dispersion, nonlinearity, and filter offset damage. Therefore, the parameters optimized by the neural network try to find the optimal value to compensate for these damages. Then, if you want to monitor whether the filter is offset, you can get the optimal value from the adaptive filter h. * Specific identification features are found in the frequency response of the sensor because the adaptive filter in the neural network attempts to compensate for the damage caused by a certain filter offset. Based on this feature, identification, monitoring and positioning can be achieved.
[0132] S4. Based on the tap coefficients of the adaptive filter in the deep neural network model, the damage identification results and damage location and tracing results of the filter in the optical communication system are obtained.
[0133] Optionally, the tap coefficients of the adaptive filter in the deep neural network model in S4 are used to obtain damage identification results and damage location tracing results of the filter in the optical communication system, including:
[0134] S41. Obtain the tap coefficient of each adaptive filter of multiple adaptive filters in the deep neural network model.
[0135] S42 . Draw a spectrum response diagram of each adaptive filter based on the tap coefficients of each adaptive filter.
[0136] S43. Based on the spectrum response diagram of each adaptive filter, damage identification is performed according to the spectrum characteristics of each adaptive filter, thereby obtaining damage identification results and damage location tracing results of the filters in the optical communication system.
[0137] In a feasible implementation, the filter tap coefficients obtained by optimization can be obtained from each span, and a frequency response diagram can be drawn to observe the working status of the filter; because the reference value does not include damage, when the neural network encounters a filter abnormality at a certain position during reverse simulation, the corresponding adaptive filter during training will try to compensate for the damage. Based on this feature, the monitoring and positioning of the filter abnormality can be achieved.
[0138] Furthermore, in the analysis and verification of the embodiment of the present invention, a 5-span polarization multiplexing optical communication transmission system is used to actually verify the scheme. According to the previous analysis, the neural network will obtain 5 optimized adaptive filters, such as Figure 6 The frequency response of the filter tap coefficient of a normal transmission system shows that the spectrum response of the filter in the passband is relatively flat without obvious fluctuation linearity, which verifies the theoretical analysis of the embodiment of the present invention.
[0139] Figure 7 This is the identification result diagram when a frequency offset is introduced into the first filter in the transmission system. It can be found that the frequency response of the tap coefficient of this filter is obviously drastic. This is because the optimized filter attempts to compensate for the signal damage caused by the filter offset at this position, so it is quite different from the spectrum response of other filters, thereby realizing the identification and positioning of the filter offset.
[0140] In order to further verify the effectiveness of the scheme, the third and fifth filters are introduced into the optical transmission system with frequency offsets. Figure 8 and Fig. 9 This is the recognition effect diagram obtained. By observing its frequency response, we can still easily find the abnormal filter and its specific location. It can be seen that this method has good adaptability and recognition effect, which verifies the effectiveness and practicality of the proposed method.
[0141] In the embodiment of the present invention, a filter offset identification and positioning method for long-distance optical fiber communication transmission links is proposed. By processing the received signal and combining the optimization capability of the neural network, the optimized filter parameters can be obtained, and the abnormal condition of the link can be monitored without using any measuring instruments, only by using the parameter value characteristics of the received signal processing and optimization. A reverse reconstruction method for the optical transmission link is implemented based on data and model drive, and a dynamic monitoring scheme for the damage and transmission quality of the optical link filter is realized. It has the ability to accurately identify and locate filter damage, and can timely and comprehensively reflect the status information of the link.
[0142] like Fig.10As shown, an embodiment of the present invention provides a non-intrusive optical bandpass filter frequency offset monitoring and positioning device 1000, the device 1000 is used to implement a non-intrusive optical bandpass filter frequency offset monitoring and positioning method, the device 1000 includes:
[0143] The acquisition module 1010 is used to acquire a received signal of the optical communication system.
[0144] The offline signal processing module 1020 is used to perform offline signal processing on the received signal to obtain a first branch signal and a second branch signal.
[0145] The input module 1030 is used to input the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation DBP to obtain tap coefficients of an adaptive filter in the deep neural network model.
[0146] The output module 1040 is used to obtain damage identification results and damage location tracing results of the filter in the optical communication system based on the tap coefficients of the adaptive filter in the deep neural network model.
[0147] Optionally, the offline signal processing module 1020 is further configured to:
[0148] S21. Perform I / Q orthogonalization on the received signal.
[0149] S22. Perform dispersion compensation on the orthogonalized received signal.
[0150] S23, performing adaptive equalization and polarization demultiplexing on the dispersion compensated received signal.
[0151] S24. Estimating the frequency offset of the received signal after adaptive equalization and polarization demultiplexing.
[0152] S25. Divide the received signal after the frequency offset estimation into two branches to obtain a first branch signal and a second branch signal.
[0153] Optionally, the input module 1030 is further configured to:
[0154] S31. Add dispersion damage to the first branch signal, reconstruct the signal including dispersion damage, nonlinear damage and filter offset damage, and obtain the preprocessed first branch signal.
[0155] S32, performing carrier phase recovery and judgment on the second branch signal to obtain a signal that eliminates dispersion damage, nonlinear damage, and filter frequency offset damage, and performing phase alignment based on the first branch signal to obtain a preprocessed second branch signal.
[0156] S33. Use the preprocessed first branch signal as the input signal of the deep neural network model based on DBP, use the preprocessed second branch signal as the reference signal of the deep neural network model based on DBP, train the deep neural network model based on DBP, and obtain the tap coefficients of the adaptive filter in the deep neural network model.
[0157] Optionally, the input module 1030 is further configured to:
[0158] S331. Set a DBP-based deep neural network model.
[0159] S332. Use the preprocessed first branch signal as the input signal of the DBP-based deep neural network model, input it into the set DBP-based deep neural network model, and obtain the forward calculation output signal of the deep neural network model.
[0160] S333. Use the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP, train the deep neural network model based on DBP based on the forward calculation output signal, and obtain the tap coefficients of the adaptive filter in the deep neural network model.
[0161] Optionally, the input module 1030 is further configured to:
[0162] S3311. Set the number of layers of the DBP-based deep neural network model.
[0163] S3312, obtaining the number of spans of the optical communication system, and setting an adaptive filter in a DBP-based deep neural network model according to the position of each span of the optical communication system.
[0164] Optionally, the number of layers N of the DBP-based deep neural network model is set as shown in the following formula (1):
[0165] N=L / ΔL (1)
[0166] Wherein, L is the total transmission distance of the optical communication system; ΔL is the compensation step size of DBP.
[0167] Optionally, the input module 1030 is further configured to:
[0168] S3331. Use the preprocessed second branch signal as the reference signal of the DBP-based deep neural network model, and train the DBP-based deep neural network model based on the forward calculation output signal to obtain the dispersion coefficient, nonlinear coefficient of each layer of the deep neural network model to be optimized, and the tap coefficient of the adaptive filter in each span.
[0169] S3332. Based on the gradient descent method, the optimal values of the dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span are solved to obtain the optimal dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span of each layer of the deep neural network model.
[0170] Optionally, the output module 1040 is further configured to:
[0171] S41. Obtain the tap coefficient of each adaptive filter of multiple adaptive filters in the deep neural network model.
[0172] S42 . Draw a spectrum response diagram of each adaptive filter based on the tap coefficients of each adaptive filter.
[0173] S43. Based on the spectrum response diagram of each adaptive filter, damage identification is performed according to the spectrum characteristics of each adaptive filter, thereby obtaining damage identification results and damage location tracing results of the filters in the optical communication system.
[0174] In the embodiment of the present invention, a filter offset identification and positioning method for long-distance optical fiber communication transmission links is proposed. By processing the received signal and combining the optimization capability of the neural network, the optimized filter parameters can be obtained, and the monitoring of abnormal conditions of the link can be estimated by only using the parameter value characteristics of the received signal processing and optimization without using any measuring instruments. A reverse reconstruction method for the optical transmission link is implemented based on data and model drive, and a dynamic monitoring scheme for the damage and transmission quality of the optical link filter is realized. It has the ability to accurately identify and locate filter damage, and can timely and comprehensively reflect the status information of the link.
[0175] Fig.11 1 is a schematic diagram of the structure of an electronic device 1100 provided by an embodiment of the present invention. The electronic device 1100 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 1101 and one or more memories 1102, wherein the memory 1102 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 1101 to implement the following non-intrusive optical bandpass filter frequency offset monitoring and positioning method:
[0176] S1. Obtain a received signal of an optical communication system.
[0177] S2. Perform offline signal processing on the received signal to obtain a first branch signal and a second branch signal.
[0178] S3. Input the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation DBP to obtain tap coefficients of an adaptive filter in the deep neural network model.
[0179] S4. Based on the tap coefficients of the adaptive filter in the deep neural network model, the damage identification results and damage location and tracing results of the filter in the optical communication system are obtained.
[0180] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions, which can be executed by a processor in a terminal to complete the above-mentioned non-intrusive optical bandpass filter frequency offset monitoring and positioning method. For example, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0181] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.
[0182] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A non-intrusive optical bandpass filter frequency offset monitoring and positioning method, characterized in that: The method comprises: S1. Obtaining a received signal of an optical communication system; S2. performing offline signal processing on the received signal to obtain a first branch signal and a second branch signal; S3, inputting the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation DBP to obtain tap coefficients of an adaptive filter in the deep neural network model; S4. Based on the tap coefficients of the adaptive filter in the deep neural network model, obtain damage identification results and damage location tracing results of the filter in the optical communication system; The step of inputting the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation DBP in S3 to obtain the tap coefficients of the adaptive filter in the deep neural network model includes: S31, adding dispersion damage to the first branch signal, reconstructing a signal including dispersion damage, nonlinear damage and filter offset damage, and obtaining a preprocessed first branch signal; S32, performing carrier phase recovery and judgment on the second branch signal to obtain a signal that eliminates dispersion damage, nonlinear damage, and filter frequency offset damage, and performing phase alignment based on the first branch signal to obtain a preprocessed second branch signal; S33, using the preprocessed first branch signal as an input signal of a deep neural network model based on DBP, using the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP, training the deep neural network model based on DBP, and obtaining tap coefficients of an adaptive filter in the deep neural network model; In the S33, the preprocessed first branch signal is used as an input signal of a deep neural network model based on DBP, the preprocessed second branch signal is used as a reference signal of a deep neural network model based on DBP, the deep neural network model based on DBP is trained, and the tap coefficients of the adaptive filter in the deep neural network model are obtained, including: S331, setting a DBP-based deep neural network model; S332, using the preprocessed first branch signal as an input signal of a DBP-based deep neural network model, and inputting it into a set DBP-based deep neural network model to obtain a forward calculation output signal of the deep neural network model; S333. Use the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP, and train the deep neural network model based on DBP based on the forward calculation output signal to obtain tap coefficients of an adaptive filter in the deep neural network model.
2. The method according to claim 1, characterized in that The step of performing offline signal processing on the received signal to obtain the first branch signal and the second branch signal in S2 includes: S21, performing I / Q orthogonalization on the received signal; S22, performing dispersion compensation on the orthogonalized received signal; S23, performing adaptive equalization and polarization demultiplexing on the dispersion compensated received signal; S24, performing frequency offset estimation on the received signal after adaptive equalization and polarization demultiplexing; S25. Divide the received signal after the frequency offset estimation into two branches to obtain a first branch signal and a second branch signal.
3. The method according to claim 1, characterized in that The setting of the DBP-based deep neural network model in S331 includes: S3311. Set the number of layers of the DBP-based deep neural network model; S3312, obtaining the number of spans of the optical communication system, and setting an adaptive filter in a DBP-based deep neural network model according to the position of each span of the optical communication system.
4. The method according to claim 3, characterized in that The setting in S3311 is based on the number of layers N of the deep neural network model of DBP, as shown in the following formula (1): N=L / ΔL (1) Wherein, L is the total transmission distance of the optical communication system; ΔL is the compensation step length of DBP.
5. The method according to claim 1, characterized in that The step of using the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP in S333, training the deep neural network model based on DBP based on the forward calculation output signal, and obtaining the tap coefficients of the adaptive filter in the deep neural network model includes: S3331, using the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP, and training the deep neural network model based on DBP based on the forward calculation output signal to obtain the dispersion coefficient, nonlinear coefficient of each layer of the deep neural network model to be optimized, and the tap coefficient of the adaptive filter in each span; S3332. Based on the gradient descent method, the optimal values of the dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span are solved to obtain the optimal dispersion coefficient, nonlinear coefficient and tap coefficient of the adaptive filter in each span of each layer of the deep neural network model.
6. The method according to claim 1, characterized in that The step of obtaining the damage identification result and damage location tracing result of the filter in the optical communication system based on the tap coefficient of the adaptive filter in the deep neural network model in S4 includes: S41, obtaining a tap coefficient of each adaptive filter of a plurality of adaptive filters in a deep neural network model; S42, drawing a spectrum response diagram of each adaptive filter based on the tap coefficients of each adaptive filter; S43, based on the spectrum response diagram of each adaptive filter, damage identification is performed according to the spectrum characteristics of each adaptive filter, thereby obtaining damage identification results and damage location tracing results of the filters in the optical communication system.
7. A non-intrusive optical bandpass filter frequency offset monitoring and positioning device, characterized in that: The device comprises: An acquisition module, used for acquiring a received signal of an optical communication system; An offline signal processing module, used for performing offline signal processing on the received signal to obtain a first branch signal and a second branch signal; An input module, used to input the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation DBP to obtain tap coefficients of an adaptive filter in the deep neural network model; An output module, used to obtain damage identification results and damage location tracing results of the filter in the optical communication system based on the tap coefficients of the adaptive filter in the deep neural network model; The step of inputting the first branch signal and the second branch signal into a deep neural network model based on optical fiber digital back propagation (DBP) to obtain tap coefficients of an adaptive filter in the deep neural network model includes: S31, adding dispersion damage to the first branch signal, reconstructing a signal including dispersion damage, nonlinear damage and filter offset damage, and obtaining a preprocessed first branch signal; S32, performing carrier phase recovery and judgment on the second branch signal to obtain a signal that eliminates dispersion damage, nonlinear damage, and filter frequency offset damage, and performing phase alignment based on the first branch signal to obtain a preprocessed second branch signal; S33, using the preprocessed first branch signal as an input signal of a deep neural network model based on DBP, using the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP, training the deep neural network model based on DBP, and obtaining tap coefficients of an adaptive filter in the deep neural network model; In the S33, the preprocessed first branch signal is used as an input signal of a deep neural network model based on DBP, the preprocessed second branch signal is used as a reference signal of a deep neural network model based on DBP, the deep neural network model based on DBP is trained, and the tap coefficients of the adaptive filter in the deep neural network model are obtained, including: S331, setting a DBP-based deep neural network model; S332, using the preprocessed first branch signal as an input signal of a DBP-based deep neural network model, and inputting it into a set DBP-based deep neural network model to obtain a forward calculation output signal of the deep neural network model; S333. Use the preprocessed second branch signal as a reference signal of a deep neural network model based on DBP, and train the deep neural network model based on DBP based on the forward calculation output signal to obtain tap coefficients of an adaptive filter in the deep neural network model.
Citation Information
Patent Citations
Optical signal-to-noise ratio monitoring device and optical signal-to-noise ratio monitoring method
CN109474336A
CO-OFDM transmission system nonlinear damage compensation system and method
CN111355536A
Optical fiber nonlinear equalization method based on IPCA-DNN algorithm
CN113285758A
Method and System for Optical Impairments Mitigation for High-Speed Optical Communication Systems
US20150063818A1
Method and apparatus for high resolution monitoring of optical signals
US6493088B1
